Shayantharan Sivarajalingam

Modelling and Control of a Spray

Drying Process

Masters Thesis, November 2009

Shayantharan Sivarajalingam

Modelling and Control of a Spray

Drying Process

Masters Thesis, November 2009

Modelling and Control of a Spray Drying Process,

This report was prepared by

Shayantharan Sivarajalingam

Supervisors
Hans Henrik Niemann-Associate Professor, DTU Electrical Engineering
Ole Ravn -Associate Professor, DTU Electrical Engineering
Christer Utzen- GEA Niro,GEA Process Engineering A/S,Process Control

Department of Electrical Engineering

Automation and Control
Technical University of Denmark
Elektrovej building 326
DK-2800 Kgs. Lyngby
Denmark
http://www.dtu.dk/centre/aut
Tel:
(+45) 45 25 35 50
Fax:
(+45) 45 88 12 95

Release date:
Category:

30 November 2009
1 (public)

Edition:

First

Comments:

This report is part of the requirements to achieve the Master of

Science in Engineering (M.Sc.Eng.) at the Technical University
of Denmark. This report represents 30 ECTS points.

Rights:

c
Shayantharan,
2009

Preface

This masters thesis documents the process and results of the project Modelling and Control of a Spray Drying Process by Shayantharan Sivarajalingam.
The project has been conducted at the Technical University of Denmark
(DTU) at the Department of Electrical Engineering, Automation and Control in the period from June till November 2009 and represents a workload
of 30 ECTS points.
I would like to express my sincere thanks and appreciation to my supervisors
Associate Professor Hans Henrik Niemann and Associate Professor Ole Ravn
for their competent guidance and support throughout the development of
the ideas in my thesis.
All at once I would like to say thanks to Christer Utzen, for his assistance
during the thesis work and GEA Niro for giving me the opportunity to work
on this real system, which has given me a lot of experience.
Last but not least, I will express my gratitude to my friends: Kristian, Lars,
Soaban, Malcolm, Mickey, Varun and my Chellams for their patience and
great support during the project.

Abstract

This Master thesis is about modelling of a spray drying process. In a spray

drying process a liquid feedstock is dried by spraying the feed into heated
air. This process is utilized in the dairy industry, where milk is dried into
milk powder. Moreover, this process is also applied in the chemical and
medical industries.
The quality of the final product can vary, depending on how the system is
controlled. The purpose of modelling the spray dryer is to use the model for
development and testing of control strategies to the dryer.
The point of reference is a steady state model, which has been developed
to a dynamic model of a multi stage dryer with a mixed air flow. The
model is based on mass and energy equations, as well as product drying
characteristics. The model describes the drying conditions in the drying
chamber with regards to temperature and humidity.
The developed model is verified with data from a genuine Multi stage dryer
at GEA Niros test station. Furthermore, it has been demonstrated that
the model can be applied in connection with a controller and thus examine
various control systems.

Dansk Resum
e

Denne Master speciale omhandler modellering af et trsprayingsanlg. I

et trspraying anlg bliver et givent flydende stof trret ved at sende det
flydende stof gennem opvarmet luft. Det er anvendt i blandt andet mejeri
industrien, hvor blandt andet mlk trres til mlkepulver. Dette er ligeledes
anvendt i kemi og medicinal industrien.
Kvaliteten p
a det frdigtrrede product kan variere alt efter hvordan anlgget bliver styret. M
alet med modelleringen af anlgget er at kunne
anvende modellen til videre udvikling og test af reguleringssystemer til trspraying anlg.
Med udgangspunkt i en steady state model, er der blevet en udviklet en
dynamisk model af en multi stage dryer med mixed air flow. Denne er
baseret p
a masse og energi balance ligninger og produkt trrings karakterstikker. Modellen beskriver trringsforholdene i trspraying kammeret med
hensyn til temperatur og luftfugtighed.
Den udviklede model er blevet verificeret med data fra Multi stage dryer
anlg hos GEA Niros test station. Desuden er det blevet demonstreret at
modellen kan anvendes til at undersge reguleringssystemer.

Contents

List of Figures
List of Tables

xiii
xviii

1 Introduction

1.1

Description of the Problem . . . . . . . . . . . . . . . . . . .

1.2

The objective of the project . . . . . . . . . . . . . . . . . . .

2 Introduction to the Spray Drying Process and Spray Dryers

3 Modeling

13

3.1

Black Box model . . . . . . . . . . . . . . . . . . . . . . . . .

14

3.2

White Box model . . . . . . . . . . . . . . . . . . . . . . . . .

14

3.3

Grey Box model . . . . . . . . . . . . . . . . . . . . . . . . .

15

3.4

Chapter Summary . . . . . . . . . . . . . . . . . . . . . . . .

15

4 Modelling a Spray Dryer

17

4.1

Preparations and Assumptions for Modelling . . . . . . . . .

17

4.2

Steady State Mass and Energy Balances for Spray Dryers . .

20

4.2.1

Mass Balance . . . . . . . . . . . . . . . . . . . . . . .

24

4.2.2

Equilibrium Moisture Content . . . . . . . . . . . . .

25

4.2.3

Steady State Solution . . . . . . . . . . . . . . . . . .

29

4.3

4.4

Test on a Multi Stage Dryer- MSD20 . . . . . . . . . . . . . .

29

4.3.1

The Test on MSD-20 . . . . . . . . . . . . . . . . . . .

31

Steady State Calculations . . . . . . . . . . . . . . . . . . . .

33

4.4.1

Steady State Results . . . . . . . . . . . . . . . . . . .

35

4.4.2

Effect of varying Operation Variables

37

. . . . . . . . .

5 Dynamic Modelling of a Mixed Flow Spray Dryer

43

5.1

Total mass of air in the Spray Dryer . . . . . . . . . . . . . .

44

5.2

Drying Kinetic Mechanism . . . . . . . . . . . . . . . . . . . .

47

5.2.1

Mass Transfer Rate

. . . . . . . . . . . . . . . . . . .

49

5.2.2

Mass and Heat Transfer Coefficients . . . . . . . . . .

51

5.2.3

Droplet size . . . . . . . . . . . . . . . . . . . . . . . .

53

5.3

Matlab Simulink Implementation . . . . . . . . . . . . . . . .

54

5.4

Test:Dynamic model . . . . . . . . . . . . . . . . . . . . . . .

54

5.4.1

Step: Temperature of main inlet air . . . . . . . . . .

57

5.4.2

Step: Feed Flow rate . . . . . . . . . . . . . . . . . . .

58

5.4.3

Step: Main inlet air flow rate . . . . . . . . . . . . . .

60

Test:Drying Time for particle . . . . . . . . . . . . . . . . . .

62

5.5.1

Test 1: particle sizes . . . . . . . . . . . . . . . . . . .

65

5.5.2

Test 2: Temperature . . . . . . . . . . . . . . . . . . .

66

5.5.3

Test 3: Effective Diffusivity . . . . . . . . . . . . . . .

67

5.5.4

Test 4: Critical Moisture Content . . . . . . . . . . . .

67

5.6

Summary: tests . . . . . . . . . . . . . . . . . . . . . . . . . .

68

5.7

Modifications in the Dynamic Model . . . . . . . . . . . . . .

69

5.5

5.7.1
5.8

Implementation of longer drying times and change in

evaporation rate . . . . . . . . . . . . . . . . . . . . .

70

Summary: modifications . . . . . . . . . . . . . . . . . . . . .

73

6 Linearisation Analysis
6.1

Operating Point . . . . . . . . . . . . . . . . . . . . . . . . .

75
76

6.2

Linearised results . . . . . . . . . . . . . . . . . . . . . . . . .

77

6.3

Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

80

6.4

Comparison of the linear model with the non-linear model . .

81

7 System Identification of Residual Moisture Content

85

7.1

Applied Identification Methods . . . . . . . . . . . . . . . . .

86

7.2

Estimation Data and Validation Data . . . . . . . . . . . . .

87

7.3

System identification Results . . . . . . . . . . . . . . . . . .

89

7.3.1

ARX model . . . . . . . . . . . . . . . . . . . . . . . .

90

7.3.2

ARMAX Model . . . . . . . . . . . . . . . . . . . . . .

91

7.3.3

State Space model . . . . . . . . . . . . . . . . . . . .

93

7.4

Summary: System identification

. . . . . . . . . . . . . . . .

8 Control of spray dryers

8.1

8.2

8.4

95

Control Strategy . . . . . . . . . . . . . . . . . . . . . . . . .

95

8.1.1

97

PI controller . . . . . . . . . . . . . . . . . . . . . . .

PI controller for disturbance rejection of solids content variation 98

8.2.1

8.3

93

PI controller design . . . . . . . . . . . . . . . . . . .

98

Results from PI controller implementation in dynamic model 101

8.3.1

Step on reference temperature

. . . . . . . . . . . . . 101

8.3.2

Step on Solids content . . . . . . . . . . . . . . . . . . 101

Possible control strategies . . . . . . . . . . . . . . . . . . . . 102

9 Conclusion

105

Nomenclature

110

References

111

Appendix

114

A Appendix A

115

A.1 Desorption Isotherm at low and high humidity level . . . . . 116

A.2 General moisture Characteristic and food microbiology . . . . 117
B Appendix B
B.1 Modelling Variables

119
. . . . . . . . . . . . . . . . . . . . . . . 120

B.2 Steady State Calculation . . . . . . . . . . . . . . . . . . . . . 121

B.3 Results from the Dynamic Model before modification . . . . . 123
B.3.1 Main inlet air temperature step up . . . . . . . . . . . 123
B.3.2 Feed Flow step down . . . . . . . . . . . . . . . . . . . 124
B.3.3 Temperature SFB step down . . . . . . . . . . . . . . 125
B.3.4 Temperature SFB step up . . . . . . . . . . . . . . . . 126
B.3.5 SFB air flow step down . . . . . . . . . . . . . . . . . 127
B.3.6 SFB air flow step up . . . . . . . . . . . . . . . . . . . 128
B.3.7 Main air flow step down . . . . . . . . . . . . . . . . . 129
B.4 Results from the Dynamic Model after modification . . . . . 130
B.4.1 Feed Flow step down with modefication . . . . . . . . 130
B.4.2 Temperature MAIN step down with modification . . . 130
B.4.3 Air flow MAIN step up with modification . . . . . . . 131
B.4.4 Absolute Humidity in Dryer For feed step up with
modification . . . . . . . . . . . . . . . . . . . . . . . . 132
B.4.5 Response for the system G(s)evap . . . . . . . . . . . . 132
B.4.6 Simulink model . . . . . . . . . . . . . . . . . . . . . . 133
C Appendix C

135

C.1 MSD-20 test 24-7-2009 . . . . . . . . . . . . . . . . . . . . . . 135

C.2 Test Step & Responses . . . . . . . . . . . . . . . . . . . . . . 136
C.2.1 Test Program MSD-20 week 30 2009 . . . . . . . . . . 136
C.2.2 Test Step & Results for the entire test on MSD-20 . . 138
C.2.3 Moisture content of the particle from the SFB discharge for the entire test on MSD-20 . . . . . . . . . . 139
C.2.4 Test Step & Results for change in feed rate on MSD-20 140

C.2.5 Feed flow rate and Nozzle pressure results from test
on MSD-20 . . . . . . . . . . . . . . . . . . . . . . . . 141
C.2.6 Test Step & Results for change in Main inlet air temperature on MSD-20 . . . . . . . . . . . . . . . . . . . 142
C.2.7 Test Step & Results for change in Main inlet air flow
on MSD-20 . . . . . . . . . . . . . . . . . . . . . . . . 143
C.2.8 Test Step & Results for change in SFB inlet air temperature on MSD-20 . . . . . . . . . . . . . . . . . . . 144
C.2.9 Test Step & Results for change in SFB inlet air flow
on MSD-20 . . . . . . . . . . . . . . . . . . . . . . . . 145
C.2.10 Ambient Air Conditons At AIR intake (21/7-2009) . . 146
C.3 Logbook for MSD-20 test 24-7-2009
D Appendix D

. . . . . . . . . . . . . . 147
151

D.1 Humidity Calculation . . . . . . . . . . . . . . . . . . . . . . 151

D.2 Droplet calculations . . . . . . . . . . . . . . . . . . . . . . . 152
D.3 Thermal Conductivity Air . . . . . . . . . . . . . . . . . . . . 153
D.4 Thermal Diffusivity Air . . . . . . . . . . . . . . . . . . . . . 154
D.5 Kinematic Viscosity Air . . . . . . . . . . . . . . . . . . . . . 154
D.6 Mean Residense Time for the particle in the Spray dryer . . . 155
E Appendix E

159

E.1 Air and Particle Trajectory in Chamber . . . . . . . . . . . . 160

F Appendix F

161

F.1 Mass transfer and Drying time appendix . . . . . . . . . . . . 161

F.1.1

Default state operation condtion . . . . . . . . . . . . 161

F.1.2

TOut varied . . . . . . . . . . . . . . . . . . . . . . . . 161

G Appendix G

165

G.1 1st order system . . . . . . . . . . . . . . . . . . . . . . . . . 165

H Appendix H

167

xii
H.1 System Identification . . . . . . . . . . . . . . . . . . . . . . . 168
H.2 ARMAX models . . . . . . . . . . . . . . . . . . . . . . . . . 168
H.1.1 Model Misfit Vs Number parameters for ARX model . 169
H.1.2 Zero Pole plot for the ARX model . . . . . . . . . . . 170
H.2.1 ARMAX simulations . . . . . . . . . . . . . . . . . . . 171
H.2.2 ARMAX Zero-pole plot for 6th order model

. . . . . 174

H.2.3 State Space- continous time zero-pole plot . . . . . . . 178

I

Appendix I
I.1.1
I.1

I.2

179

Linearized model- State Space(Jacobians) . . . . . . . 179

Linearization of the dynamic Model - open loop . . . . . . . . 180

I.1.2

Frequency response - from inputs to output . . . . . . 181

I.1.3

Zero-Pole plot for the transfer functions - from each

inputs to output . . . . . . . . . . . . . . . . . . . . . 185

Comparison of linear and Non linear model . . . . . . . . . . 187

I.2.1

Feed flow . . . . . . . . . . . . . . . . . . . . . . . . . 188

I.2.2

Main inlet air flow . . . . . . . . . . . . . . . . . . . . 190

I.2.3

Main inlet air temperature . . . . . . . . . . . . . . . 193

I.2.4

Solids Content . . . . . . . . . . . . . . . . . . . . . . 196

I.2.5

Relative Humidity of Ambient air . . . . . . . . . . . . 197

List of Figures

1.1

Basic Spray drying process

. . . . . . . . . . . . . . . . . . .

2.1

Spray drying process stages . . . . . . . . . . . . . . . . . . .

2.2

Multi Stage Dryer . . . . . . . . . . . . . . . . . . . . . . . .

10

2.3

Agglomeration Process Schematic . . . . . . . . . . . . . . . .

11

4.1

Schematic of a CSTR and a plug flow reactor. . . . . . . . . .

18

4.2

Temperature profile for a mixed flow spray dryer . . . . . . .

20

4.3

Morphology of Particle with maltodextrin . . . . . . . . . . .

21

4.4

Basis Blockdiagram- variable description to the system . . . .

22

4.5

Desorption Isotherm Maltodextrin DE12 . . . . . . . . . . . .

27

4.6

MSD-20 Test Station Setup . . . . . . . . . . . . . . . . . . .

40

4.7

Test Centre at GEA Niro - Multi stage Dryer MSD-20 . . . .

41

5.1

Basic Dynamic Model . . . . . . . . . . . . . . . . . . . . . .

46

5.2

Particle Morphology . . . . . . . . . . . . . . . . . . . . . . .

47

5.3

Schematic of Drying mechanism . . . . . . . . . . . . . . . . .

48

5.4

Drying Proces of a Particle with a Shrinking Model . . . . . .

49

5.5

Dynamic model blockdiagram . . . . . . . . . . . . . . . . . .

55

5.6

Dynamic step response of the TOutAir for decrease in main

inlet air temperature . . . . . . . . . . . . . . . . . . . . . . .

58

xiv

LIST OF FIGURES
5.7

Dynamic step response of the TOutAir for a increase in feed

flow rate . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

59

Dynamic step response of the TOutAir for increase in Main

inlet air flow . . . . . . . . . . . . . . . . . . . . . . . . . . .

61

Mass tranfer rate and drying time test setup

. . . . . . . . .

62

5.10 Drying time for a particle at default operation state . . . . .

64

5.8
5.9

5.11 Drying time for various particle sizes at default operation state 65
5.12 Drying time for various for a single particle for various effective diffusivity coefficients . . . . . . . . . . . . . . . . . . . .

67

5.13 Drying time for various for a single particle for various critical
moisture contents . . . . . . . . . . . . . . . . . . . . . . . . .

68

5.14 Drying Time for single Particle of different sizes for Def f =
8e 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

70

5.15 Temperature response of the outlet air for the modified model
(Feed step) . . . . . . . . . . . . . . . . . . . . . . . . . . . .

72

6.1

Basic Dynamic Model . . . . . . . . . . . . . . . . . . . . . .

75

6.2

Bode plot- example: minimum phase & non minimum phase

79

6.3

Zero-Pole plot for the linearized model . . . . . . . . . . . . .

80

6.4

Comparison of linear and Non-linear model: Main inlet air

temperature. Step change= 10 from linearised input. No
change difference is observed . . . . . . . . . . . . . . . . . .

82

Comparison of linear and Non-linear model: Feed step : 2

& 10. For the small step no difference is observed. For the
larger step a small deviation is noted. . . . . . . . . . . . . .

82

Comparison of linear and Non linear model: Solids content

step from 50 % to 80 % Such a large change is not possible
in reality. Small difference between the non- linear and linear
model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

83

7.1

Simulated ARX model output and measured output . . . . .

90

7.2

ARMAX model: 6th order & 10th order . . . . . . . . . . . .

91

7.3

State Space model : 4th order . . . . . . . . . . . . . . . . . .

93

8.1

Input and output variables for process control . . . . . . . . .

96

6.5

6.6

LIST OF FIGURES

xv

8.2

Feed rate control . . . . . . . . . . . . . . . . . . . . . . . . .

97

8.3

The control system . . . . . . . . . . . . . . . . . . . . . . . .

99

8.4

Bode plot for the open loop transfer function . . . . . . . . . 100

8.5

PI controller:step on reference temperatur . . . . . . . . . . . 102

8.6

PI controller:step on solids content . . . . . . . . . . . . . . . 103

8.7

PI controller:Continuous Solids content Disturbance . . . . . 104

A.1 Desorption Isotherm Maltodextrin DE12 . . . . . . . . . . . . 116

A.2 General moisture Characteristic and Food microbiology . . . 117
B.1 Dynamic step response of the TOutAir for increase in main
inlet air temperature . . . . . . . . . . . . . . . . . . . . . . . 123
B.2 Dynamic step response of the TOutAir for decrease in main
inlet air temperature . . . . . . . . . . . . . . . . . . . . . . . 124
B.3 Dynamic step response of the TOutAir for decrease in SFB
inlet air temperature . . . . . . . . . . . . . . . . . . . . . . . 125
B.4 Dynamic step response of the TOutAir for decrease in SFB
inlet air temperature . . . . . . . . . . . . . . . . . . . . . . . 126
B.5 Dynamic step response of the TOutAir for increase in SFB inlet
air flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
B.6 Dynamic step response of the TOutAir for decrease in SFB
inlet air flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
B.7 Dynamic step response of the TOutAir for decrease in Main
inlet air flow . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
B.8 Dynamic step response of the TOutAir for decrease in main
inlet air temperature . . . . . . . . . . . . . . . . . . . . . . . 130
B.9 Dynamic step response of the TOutAir for decrease in MAIN
inlet air temperature(modified) . . . . . . . . . . . . . . . . . 130
B.10 Dynamic step response of the TOutAir for decrease in MAIN
inlet air flow(modified) . . . . . . . . . . . . . . . . . . . . . . 131
B.11 Absolute Humidity in Dryer For feed step up with modification132
B.12 Response for the system G(s)evap . . . . . . . . . . . . . . . . 132
B.13 Simulink dynamic model . . . . . . . . . . . . . . . . . . . . . 133

xvi

LIST OF FIGURES
C.1 Test Step & Results for the entire test on MSD-20 . . . . . . 138
C.2 Moisture content of the particle from the SFB discharge . . . 139
C.3 Test Step & Results for change in feed rate on MSD-20

. . . 140

C.4 Feed flow rate and Nozzle pressure results from test on MSD-20141
C.5 Test Step & Results for change in Main inlet air temperature
on MSD-20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
C.6 Test Step & Results for change in Main inlet air flow on MSD-20143
C.7 Test Step & Results for change in SFB inlet air temperature
on MSD-20 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
C.8 Test Step & Results for change in SFB inlet air flow on MSD-20145
C.9 Ambient Air Condition at Intake . . . . . . . . . . . . . . . . 146
C.10 Logbook From Test on MSD-20 week 30 2009 . . . . . . . . . 147
C.11 Logbook From Test on MSD-20 week 30 2009 . . . . . . . . . 148
D.1 Thermal Conductivity of Air vs. Temperature . . . . . . . . . 154
D.2 Thermal Diffusivity of Air vs. Temperature . . . . . . . . . . 155
D.3 Kinematic Viscosity of Air vs. Temperature . . . . . . . . . . 156
E.1 Air Stream and particle trajectory in mixed flow chamber . . 160
F.1 Mass evaporated for various particles sizes . . . . . . . . . . . 161
F.2 Mass Transfer coefficient for various particles sizes . . . . . . 162
F.3 Crust resistance f for various particle sizes . . . . . . . . . . . 162
F.4 Drying time for particle -various feed flow . . . . . . . . . . . 163
F.5 Crust resistance f for various feed flow rates . . . . . . . . . . 163
F.6 Drying time for particle -various Main inlet air flow

. . . . . 164

G.1 Control variable as a first order system . . . . . . . . . . . . . 165

H.1 Model Misfit Vs Number parameters for ARX model . . . . . 169
H.2 Zero-Pole plot for the 10th order ARX model . . . . . . . . . 170
H.3 ARMAX model 1 . . . . . . . . . . . . . . . . . . . . . . . . . 171

LIST OF FIGURES

xvii

H.4 ARMAX model 1 . . . . . . . . . . . . . . . . . . . . . . . . . 172

H.5 ARMAX model 1 . . . . . . . . . . . . . . . . . . . . . . . . . 172
H.6 ARMAX model 1 . . . . . . . . . . . . . . . . . . . . . . . . . 173
H.7 ARMAX model 1 . . . . . . . . . . . . . . . . . . . . . . . . . 173
H.8 Zero-Pole plot for ARMAX 6th model . . . . . . . . . . . . . 174
H.9 Zero-Pole plot for ARMAX 6th model 1 . . . . . . . . . . . . 175
H.10 Zero-Pole plot for ARMAX 6th model 2 . . . . . . . . . . . . 176
H.11 Zero-Pole plot for State space model 4th order . . . . . . . . 177
H.12 Zero-Pole plot for State space model 4th order continuos time 178
I.1

Frequency response of the linearized model process input to

output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

I.2

Frequency response of the linearized model- Feed flow input

to output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

I.3

Frequency response of the linearized model- disturbance input

(Solids content) to output . . . . . . . . . . . . . . . . . . . . 183

I.4

Frequency response of the linearized model- disturbance input

(Ambient Relative humidity) to output . . . . . . . . . . . . . 184

I.5

Zero-Pole plot for the transfer functions - from each process

input to output . . . . . . . . . . . . . . . . . . . . . . . . . . 185

I.6

Zero-Pole plot for the transfer functions - from disturbance

inputs to output . . . . . . . . . . . . . . . . . . . . . . . . . 186

I.7

Comparison of linear and Non linear model: feed flow . . . . 188

I.8

Comparison of linear and Non linear model: feed flow 1 . . . 189

I.9

Comparison of linear and Non linear model: feed flow 2 . . . 189

I.10 Comparison of linear and Non linear model: Main inlet air flow190
I.11 Comparison of linear and Non linear model:Main inlet air flow 1191
I.12 Comparison of linear and Non linear model: Main inlet air
flow 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
I.13 Comparison of linear and Non linear model: Main inlet air
temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
I.14 Comparison of linear and Non linear model: Main inlet air
temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194

xviii

LIST OF FIGURES

I.15 Comparison of linear and Non linear model: Main inlet air
temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
I.16 Comparison of linear and Non linear model: Main inlet air
temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
I.17 Comparison of linear and Non linear model: Solids content . 196
I.18 Comparison of linear and Non linear model: Solids content . 196
I.19 Comparison of linear and Non linear model: Relative humidity of ambient air . . . . . . . . . . . . . . . . . . . . . . . . . 197
I.20 Comparison of linear and Non linear model: Relative humidity of ambient air . . . . . . . . . . . . . . . . . . . . . . . . . 197

List of Tables

3.1

Modelling Approach . . . . . . . . . . . . . . . . . . . . . . .

15

4.1

The manipulated variables in test of the dynamic process

(MSD-20) . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

31

The manipulated range of the variables in test of the dynamic

process(MSD-20) . . . . . . . . . . . . . . . . . . . . . . . . .

32

4.3

Default Operation Variable Values for Testing . . . . . . . . .

34

4.4

Steady State Results for the drying air temperature TOutAir .

36

4.5

Steady State Results of absolute humidity and equlibrium

moisture content . . . . . . . . . . . . . . . . . . . . . . . . .

37

4.6

Energy Level of the components

. . . . . . . . . . . . . . . .

38

5.1

Ranz-Marshall correlation Parameter Values . . . . . . . . . .

53

6.1

Operating Point for linearization: Stationary state . . . . . .

78

7.1

Sensor Description for Test system(MSD-20)

88

4.2

. . . . . . . . .

B.1 Default Operation Variable Values for Testing APPENDIX . 120

B.2 Steady State Results for the drying air tempereture TOutAir
APPENDIX . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
B.3 Steady State Results of absolute humidity and equlibrium
moisture content . . . . . . . . . . . . . . . . . . . . . . . . . 122

xx

LIST OF TABLES
C.1 Sensor Description for Test system(MSD-20)

. . . . . . . . . 135

C.2 Test Program for Test On MSD-20 . . . . . . . . . . . . . . . 136

H.1 ARMAX models . . . . . . . . . . . . . . . . . . . . . . . . . 168
I.1

Operating Point for linearization: Stationary state . . . . . . 180

LIST OF TABLES
0

LIST OF TABLES

Chapter

1
Introduction

Baby food, milk powder, coffee whitener, flavours, and various other products are produced in a spray drying process.
Spray drying is a process that transforms a given feedstock from a fluid state
into a dried particulate form by spraying the feed into a gaseous hot drying
medium. The feedstock can be a solution, emulsion or fluid paste, though it
is required that the feed is pumpable so it can be atomised into droplets. The
extensive contact between the droplets and the drying medium is the main
principle of the spray drying process, where the drying medium provides the
energy for the evaporation of the solvent in the feed. The resulting dried
product is conformed to a powder of either single particles, agglomerates,
granules, or pellets. The shape and structure of the particles depends on
the physical and chemical properties of the feed, spray dryer design, and the
operation conditions used.
The spray drying process, compared to other drying processes, is unique in
its ability to dry liquid feedstock to powder with specific physical properties,
particle morphology, and moisture content. The broad range of spray dryer
designs available fulfils the specifications stipulated by various industries
both in terms of product properties and production capacity. The spray
dried product reduces transportation cost, as there is less liquid to transport,
but also simplifies storage and handling.
Furthermore the spray dryer has the great advantage due to the ability to
dry both non-heat sensitive and heat sensitive materials. Thus the product
can be dried without any loss or changes in the volatile compounds of the
product. These compounds can be the aromatic characteristics of the prod-

Introduction

Feed

Air

Drying
chamber

Parculates
separaon /
collecon

Air

Stac
fluid bed

Total dried
product disharge
Figure 1.1: Spray drying process is characterised by a liquid feedstock that is
sprayed into a chamber in which contact with the heated drying
medium(air) results in evaporation of moisture from the droplets.
Drying takes place as the droplet moves through the drying chamber. If the particle reaches the bottom of the chamber and the drying
is complete the product is discharged. Some feedstock requires post
treatment of the particles to reach a specific powder characteristic. To
reduce the amount of particles in the exhaust air, a particle collector
is used to clean the air before leaving the drying system.

uct or some proteins in the product, which is especially important in the

food and dairy industries. Here spray drying helps with preservation of the
product which gives product stability and extended shelf life by reducing
the moisture content to levels where microbiological growth is not possible.
In addition to this the spray dryer can handle materials under aseptic and
hygienic drying condition which makes it applicable in the pharmaceutical
industry. It has also proved its worth in the protection of the environment as
it is capable of evaporating organic solvents which are potentially explosive
or have toxic risks. Thus spray dryers have a variety of industrial and
commercial uses and just a few examples were named here.

1.1 Description of the Problem

1.1

Description of the Problem

The objective of the spray drying process is to produce a product of a desired

quality at minimum cost, with maximum throughput and sustain the desired
dried product quality regardless of the disturbances in the drying operation
and variations in feed supply. Appliance of automatic control systems offers
an opportunity to improve the dryer operation and its efficiency. However,
the product quality, which is the parameter that is wished to be controlled,
consists of sub parameters such as moisture content, thermal degradation,
aroma retention, and structure and size of the particles. These parameters
are difficult to predict, but also to measure online in the system, with the
aim to be used as a control variable in the controller.
By experience it is identified that the most effective parameter to control
quality is the moisture content in the product. This can be measured with
a suitable moisture sensor, but these are not used for controller purposes in
spray dryers yet as these are expensive and have a low reliability at present
time. It is recognised that the quality of the final product is the outcome
of the drying conditions in the chamber, with respect to temperature and
humidity. Therefore the spray dryer uses the outlet drying air leaving the
chamber to describe the drying condition and uses it as a control variable
in a feedback control. So the moisture content is controlled indirectly by
maintaining a specific outlet drying air temperature by varying either the
feed flow into the dryer or the inlet drying air temperature.
So far the main spray drying operation controller is based on a single control
variable and a single manipulated variable, a SISO control system, for which
a PI controller is utilised. The proportional gain and integral time constants
are selected from the experience of setting up previous spray dryers. This
is often considered to be sufficient when the product moisture content can
be kept within a narrow limit. In case large variations or fluctuations are
observed these parameters are adjusted, which is an accepted and easy procedure.
However, there is obviously potential in optimising the existing control
system or developing a more advanced control system by using analytical
methods(MIMO- or Model Predictive Control System). This could minimise
the variations in moisture contents and improve the product quality. At the
same time this can lead to a more cost-effective drying, either by increasing
the throughput or reducing the energy consumption in the dryer, since spray
dryers are known to have a relatively poor thermal efficiency compared to
other drying processes.
With the purpose of being able to develop and analyse a control system, a
model of the spray drying system is required. There are two types of models:

Introduction
An equipment model, which combines the factors that affects the spray
drying process and describes the environment the particles are experiencing as drying takes place.
A particle model, which describes how the particles respond to the
drying environment.

There are several papers 1 on the study of how a single particle of a specific
product reacts in certain conditions. There are also very detailed Computational Fluid Dynamic (CFD) models that are able to illustrate the flows
of the particles and the air in the spray dryer very precisely. Common for
both cases are that the models have not been studied in a view to be used
in the development of the control systems in the spray dryer. But are more
used to examine the product for chemical purposes or in the development
of the spray drying design. For the CFD models the calculation times are
very long, which is not acceptable.

1.2

The objective of the project

The object of this project is to construct a general dynamic equipment model

that is fundamental and is able to predict and estimate the drying conditions
in the spray drying chamber as this is the factor that affects the quality of
the powder. The model is developed in preparation to be further used in
the development of control systems for the spray dryer. However, the main
focus in this project is the modelling and development process.
The spray dryer that is modelled is of the type: multi stage dryer with
a mixed air flow
The focus is on the spray drying chamber and the condition inside this for a running process. The start-up of the process is
neglected.
Modelling the temperature of the drying air is the main target.
Subsequently the humidity level inside the chamber is of interest.
These parameters have a significant influence on the drying of the
product.
The aim is to model the temperature and humidity levels in the chamber by using first principle methods.
1

(Langrish and Kockel (2001)),( Shabde (2006)),( Ireneusz Zbicinski and Delag (2002)),
(Ruud E.M. Verdurmen and JONG (2002)),( Lixin Huang and Mujumdar (2005)), (Kieviet
(1997))

1.2 The objective of the project

Mass and Energy balance equations are applied.

Initially a steady state model of the spray dryer is examined.
With the steady state model as the underlying basis a dynamic
model is developed .
The drying characteristics of the feed is examined.
The equilibrium moisture content and the drying kinetics is studied for maltodextrin which has been used as an example during
the modelling.
The particle behaviour, such as the agglomeration process and
fines are neglected.
For test and verification of the model, experiments on a multi stage
spray dryer (MSD-20) at GEA Niros test station is conducted.
step responses for this system was examined and data recorded.
Data is used to compare steady state model results and the the
dynamic response times for the dynamic model.
The default operating settings for this dryer has been applied in
the models
The dynamic model is linearised in order to analyse the system.
System identification principles has been tried on the data collected
from the experiment on the MSD-20.
A simple linear model of the moisture content in the particles as
a function of the main process variables is tried to be estimated.
In the last part of the project, the developed model is used to demonstrate that the model can be used for development and test of controller.
A PI controller is applied to control the outlet temperature of the
spray drying model.

Introduction

Chapter

2
Introduction to the Spray
Drying Process and Spray
Dryers

The spray drying process can be described by four stages: Atomization,

spray and air contact, evaporation of the moisture from the droplets and
product discharge.
STAGE 1

STAGE 2

ATOMIZATION

SPRAYAIR CONTACT

Liquid feed into

a spray of droplets

Mixing and flow

pa"ern

STAGE 3

DRYING OF SPRAY
Moisture
evapora#on

STAGE 4

PRODUCT DISCHARGE
Separa#on of
dried product
from the air

Figure 2.1: Spray drying consists of four process stages. This involves spraying
of liquid, contact and mixing with drying air, droplet drying particle formation and at last powder collection. These four stages are
illustrated in figure 2.2

The plant in which the spray drying process takes place is a spray dryer.
Spray dryers exists in various designs with regards to size, type(conventional,
compact, tall-form, multistage etc), air flow characteristics (Co-current,
counter, and mixed air flow) and mobility. The spray drying design and
mode of operation, together with the physical and chemical properties of

10

Introduction to the Spray Drying Process and Spray Dryers

the feed determines the final characteristics of the dried product (particle
size and structure). The spray dryer used in this project is a Multi Stage
Dryer (MSD) as illustrated in figure 2.2. The multi stage refers to the fact
that post-treatment equipment for the powder is a part of the spray drying
system, which will be elaborated below.

MULTI STAGE DRYER

STAGE 1

Feed
Drying air - Main

Fines

Exhaust air

Air out
Cyclone /
Bag filter
STAGE 2

STAGE 3

Drying air - SFB

STAGE 4

Powder out

SFB
Fan

Heater

Vibro Fluidizer
Figure 2.2: Multistagedryer with mixed air flow. Feed enters the spray drying
chamber from the top. Air is drawn from atmosphere by a fan and
heated with a heater. The heated air is mixed with the feed, which
falls down while it dries. The base of the chamber is the static fluidbed
(SFB) and is used for agglomeration and to finalize the drying of the
powder. Air outlet is at the top of the chamber. A cyclone and bag
filter is to filter the exhaust air for fine particles. The vibro fluidizer
is for post treatment of powder [GEA Niro].

At the first stage the feed is pumped from the feed tank to the atomizer1 .
1

In some cases the feed is send through a preheater/evaporator with the intention of

11
The atomizer is either a rotary atomizer (rotating disc) or a nozzle which
makes a spray of droplet. The nozzle atomizer which is operated in the
system that is examined in this project utilizes pressure to create droplets
through an orifice. The atomizer is located at the ceiling of the chamber.
The formed droplets are mixed with the drying air and evaporation commences.
The air is drawn from the atmosphere by a fan and passed through a heater.
In this spray dryer setup the air flow is mixed, which means the drying air
enters the dryer both from the top of the chamber and from the bottom.
The drying air entering from the ceiling of the chamber is denoted as the
Main inlet air. The air from the base of the chamber is the Static Fluid Bed
(SFB) inlet air. The air leaves the chamber from the top of the chamber as
well, which is mentioned as the outlet air.
Following the evaporation of moisture from the droplets the majority of the
dried particles fall down to the base of the drying chamber. A Multi stage
dryer has a Static Fluid Bed at this place which serves to finalize the drying
of the product and for agglomeration. Agglomeration is the process where
wet or partially dried droplets get in contact with dry particles and forms
larger particles. This is illustrated in figure 2.3. The air leaves the chamber
from the top of the chamber as well, which is mentioned as the outlet air. A
cyclone and/or a bag filter are used to filter out particles from the exhaust
air. The filtered particles are referred to as fines and these are led back into
the chamber to increase the agglomeration process.

Figure 2.3: Agglomeration Process Schematic. Dry particles(fines)collision with

wet particles, and thus gets into a structure or increase its size by
getting more layers[GEA Niro].
increasing the viscosity of the feed or increasing the solids content in the feed.

12

Introduction to the Spray Drying Process and Spray Dryers

The choice of spray drying setup, which includes drying chamber design,
atomizer, air inlet and disperser has an influence on the resulting powder
size and how the product reacts to the temperature and humidity profiles
existing in the dryer due to the selected operation conditions. Besides the
design of the spray dryer and its equipment, the chemical composition of
the solids, affects the particle shape and formation during the drying.
Particles are continuously discharged from the SFB. These are led to a vibrating fluid bed, which is equipment for post treatment of the powder.
However, the focus in this project has only been on the spray drying chamber.A more detailed description of the spray dryer design and its equipment
is found in (Masters (2002)).
The terms droplet and particle has been alternately used throughout the
report to describe the element which is dried. To elaborate this: an element
which enters the drying chamber is at the outset a droplet and turns into a
particle as it solidifies.

Chapter

3
Modelling

With the intention to be able to evaluate more advanced or novel control

strategies, an understanding of the static and dynamic properties of the
spray drying process is needed. This can be obtained with an estimated
mathematical model of the plant. Describing and developing a completely
accurate mathematical model of a spray drying process is a complicated
assignment due to high complexity of the physical, chemical, and mechanical
properties in such a system. This embraces for instance the heat and mass
transfer both within the particle at the boundary between the solid phase
and liquid phase, but also the particle and surroundings. Another aspect
of this multifaceted process is the accounting for the various entrances of
the drying medium into the spray dryer and the resulting flow patterns for
the gas and the particles. However, in all models some uncertainty in the
process behaviour will arise due to unmeasured disturbances, unmodelled
dynamics and nonlinearities. Although the mathematical model only will
be an approximation of the real process, it will be acceptable if it is capable
of giving a practically realistic representation of the process and thus satisfy
the previously defined objective.
The use of steady state models are well established in chemical engineering
for plant analysis that can be used to calculate the necessary process conditions for an optimal exploitation of the system concerning powder properties
and energy consumption. However, a dynamic model is an important part
of operability study, both in assessing the consequences of plant malfunction
and in the mitigation of possible effects. Moreover it gives a better understanding of process performance and is therefore a significant instrument for
process optimisation.

14

Modeling

To decide which approach to use for a modelling task the required level of
flexibility, time frame or validity goal, available resources , and the number
of approximations that is acceptable, has to be considered. In the forthcoming section different methods to develop or estimate a dynamic model are
explained (Labspace (2009)),(OCallagan and Cunningham (2005)).

3.1

Black Box model

The black box modelling strategy is used for investigating a complex system
with no or minimal knowledge and assumptions about the process and the
internal structure. Such a model is represented by an empirical description
or a set of transfer parameters that relate the output of the model to a set
of inputs. With the sufficient data available, containing the significant dynamics of the system, an estimate of a model is achievable and is known as
system identification. Thus the need of experimental data for this method
involves data collection, determination of model structure, parameter estimation, and model validation. However correct the dynamics are revealed in
the model, the physical details of the process are excluded. The determined
model is specific to the system, operating region, and the product which
data is extracted from.
This lack of flexibility is the main disadvantage of black box modelling, since
the effect of changes in any of the process conditions outside of those met
during the structuring of the model cannot be concluded. Another constraint
on this type of model, is the lack of any form of physical meaning, which
makes it difficult to relate it to the real object being modelled. Nonetheless
it has proven its effectiveness in situations where important parameters are
complicated to identify and measure online, such as the residual moisture
content in the final powder. This subject will be elaborated in section 7.

3.2

White Box model

A white box model is the most detailed and comprehensive category within
modelling. It is based on a first principle approach, which describes the
physical processes at the lowest level. The result will be a true nonlinear
dynamic model and as close as possible to the true description of the plant.
In contrast to the black box model, this type of model will be fully predictive,
even in the situation when changes in process conditions are outside the
normal operating conditions. In spite of the fact that the model is flexible
and realistic, the outcome could be a model of great complexity. The more
complex a model is, the more difficult it will be to identify the increased

3.3 Grey Box model

15

number of parameter values. A pure white box model cannot exist as it is

essentially a copy of the reality. So what is needed, is a model with a simple
approach but which demonstrates realistic process phenomena. Thus the
majority of simulation models are grey box models.

3.3

Grey Box model

A grey box model provides a physical representation of the system, though

some of the physical parameters are simplified or approximated by an empirical model. This hybrid model structure is the result of the combination
of the best properties from the white and black box model, and the method
that is used in this project: the flexibility enables one to model the plant
design and determine the effect of variations in chamber size or changes in
material parameters. In addition to this it is physically close to the real process, however, a convection heat transfer coefficient is utilized to describe the
heat changes around a particle instead of a model of the actual laminar flow
that requires an airflow model. An empirical model is chosen for determination residual powder moisture content. The reasonable trade-off between
complexity and performance is suitable from a control point of view.

3.4

Chapter Summary

Type
White box

Black box

Grey

Modelling Approach
Advantage
Disadvantage
Extremely flexible
High Complexity
Realistic
Large computer power
slow
Parameter identification
Low flexibility
Minimal computer power
Non physical
Fast
Flexibility
Error checking
physicality

Time Frame
Long

Short

medium

Table 3.1: Summing up the pros, cons and the time frame for the three modeling
approaches. The time frame is connected to flexibilty requirement.

16

Modeling

Chapter

4
Modelling a Spray Dryer

In this chapter the white box modelling method described in chapeter 3 is

applied to model the spray drying process. Explaining the reality completely
by physical equations is difficult. For that reason some simplification are
made which is elaborated in the follwing section.

4.1

Preparations and Assumptions for Modelling

Modelling the spray dryer can be done at various levels and degrees of details,
from describing the flow, reaction rate and the effect on circumstances of
droplet of a liquid to the overall energy flow and mass flow for the total
spray dryer. The purpose with modelling the spray dryer in this project is
to be capable of estimating and predicting the temperature of the drying air
in the spray drying chamber, which is to be used further in the control of
the moisture content of the final product. There are four main phenomena
in a spray drying operation:
1. Atomisation of the liquid feed
2. Drying of the droplets once they are formed
3. Motion of the droplet in the spray drying unit
4. Product discharge
The region of interest is the spray drying chamber and on the drying of
the droplets once they are formed. Modelling the motion of the droplet is

18

Modelling a Spray Dryer

recognised as being dependent on the geometry of the chamber and the mechanical setup of the system. Thus it is more of a mechanical problem, since
the motion of the particles cannot be directly controlled with the flow and
temperature parameters we have at hand. To describe this class of problems, rate based models, which are dynamic models that describe the rate
at which the solvent removed from the droplets as they travel through the
spray drying chamber can be used(Gauvin and Katta (1976)). Otherwise the
particle-source-in-cell models, that assumes the droplets to be a source
of mass, energy and momentum in a grid of the drying gas can also be used
Papadakis (1988). Due to the complexity of calculating the heated drying
gas flow and particle motion, it usually requires Computational Fluid Dynamic (CFD) techniques, which are tools that use numerical methods and
algorithms, to solve these models. The disadvantages using this approach
are the long calculation times and model parameter values that may have
no physical meaning.
Therefore a simpler mass and energy balance model, incorporating equilibrium relationships on the amount of moisture in the particle is utilized.
Chemical reaction engineering techniques have been considered and used to
model the spray dryer, in view of the fact that the drying process can be
viewed as a reaction (mixing) between gas and liquid/solids (vapour). The
reaction process can be represented either as in a Continuously Stirred
Tank Reactor (CSTR), a plug flow reactor or a sequence of these.(4.1)

Figure 4.1: Schematic of a CSTR and a plug flow reactor. The effluent composition of the CSTR is identical to the conditions that exist in the
reactor. For the plug flow reactor the outlet condition varies along
the length of the tube.

In the CSTR the contents of the reactor are assumed to be ideally well mixed
and the reactants and products flow into and out of the reactor continuously.
This means that the temperature, pressure and concentration levels are in-

4.1 Preparations and Assumptions for Modelling

19

dependent of spatial position within the reactor. Accordingly it also implies

that the composition and the temperature of the effluent flow are identical
to the gas in the chamber.
The plug flow reactor is an ideal flow assumption in a tube in which the fluid
is well mixed in the radial and angular directions. The velocity, composition
and temperature of the fluid are functions of the axial position (along the
length of the tube) only. The plug flow can also be described as an infinite
number of CSTRs in a cascade connection. To model a co-current spray
dryer (Air inlet from top and outlet air in the bottom), it is anticipated that
this reactor type will give the best representation. But due to the fact that a
model of a mixed flow spray dryer is wanted, which has an inlet air flow from
the top and the bottom of the chamber, it is expected that the well mixed
CSTR reactor model is most suitable and will illustrate the inlet air mixing
best. Moreover the air leaving the dryer is from the top of the chamber and
thus using a CSTR model, this temperature will not dependent on the travel
length of the air flow.
Additionally, when looking at the temperature profile for a mixed flow spray
drying chamber (figure 4.2), it is observed that the chamber temperature to
a great extent is the same and well mixed, except at the air inlet entrances
where the temperature is higher due to the heated air. Therefore it is assumed that the spray dryer can be modelled as a CSTR process.
The model prepared is based on the following assumptions and simplifications:
1. The Spray drying process is modelled as Continuously Stirred Tank
Reactor (CSTR). In this the drying gas and the feed are continuously
injected into the chamber at uniform flow rates. The state of the gas
in the chamber is identical to the state of the gas leaving the chamber.
2. The model will be based on mass and energy balance with equilibrium
relationship incorporated.
3. The gas is assumed to be a composition of dry air and vapour, which
behaves as an ideal gas and flows as a perfect mixture. This has an
influence on the calculations of the gas density and as well on the
relation between absolute humidity and the partial vapour pressure.
4. The liquid feed is assumed to be completely atomised, that is all the
droplets are of uniform size and homogeneous. In figure 4.3 particles
with various amount of maltodextrin are depicted and as expected
the particles are not perfectly spherical. However for simplicity the
particles are modelled as having a spherical shape. They are all well
mixed in the chamber and do not interact with one another. Because of

20

Modelling a Spray Dryer

Figure 4.2: Temperature profile for a mixed flow spray dryer. The figure to the
left is the temperature profile given by Masters (2002). The arrows
indicate air flow. The dotted arrows point towards product flow direction. The figure to the right is a temperature profile with relative
temperatures(GEA Niro). Red is hot. Blue is less warm. It is seen
on the figures that a great part of the chamber, except from the air
inlet entrances, has the same temperature. Due to this information it
is assumed that the spray dryer can be modelled as a CSTR process.

this simplification, the agglomeration process is disregarded, as there

will be no small particles (fines) nor will there be created any, due to
no collisions of particles.

4.2

Steady State Mass and Energy Balances for

Spray Dryers

The spray dryer operating requirements are found by solving mass and heat
balance calculations in steady state which is very common in chemical engineering. With the production rate requests, feedstock, dried product, and
ambient air properties at hand, the air flow rate requirements can be estimated. Correspondingly the moisture content of the final product, for a
presented drying air flow, can be calculated at certain conditions. For this
to succeed it is presumed that the dryer is well mixed and hence the gas
condition is uniform inside the drying chamber. It is expected that the outlet gas and outlet particles are in equilibrium, such that the temperatures
of these elements are equal. Accordingly the solids moisture content of the
outlet product is in equilibrium with the gas temperature and humidity. The

4.2 Steady State Mass and Energy Balances for Spray Dryers

21

Figure 4.3: Morphology of rice starch with little maltodextrin in particle on the
left picture and increased amount of maltodextrin on the right picture.
Though the particles are not completely spherical, this is assumed in
the modelling. The density of the structure is also dependent on the
drying temperature.[GEA Niro]

equilibrium moisture content is the moisture content at which the product is

neither gaining nor losing moisture; this however, is a dynamic equilibrium
and changes with relative humidity and temperature.
In the following the equations of the mass and energy balances, for an open
cycle mixed flow spray dryer with an aqueous feedstock, are illustrated. The
conservation of mass and energy in a steady state flow process is expressed
as: the rate of mass/energy flow into the system is equal to the rate of
mass/energy flow out of the system. First the energy balance is considered
in (4.1). In figure 4.4 a diagram illustrating input and output to the system
and a variable description for the following equations.

FM aindry HairIn + FSF Bdry HairIn + Ff eed Hf eedIn

= FOutdry HairOut + Fpowder HpowderOut

(4.1)

Where FM aindry is the Main inlet air flow, FSF Bdry is the inlet air flow from
the SFB while FOut is the outlet air flow, all of them in dry form( Kg
s ).
The dry components are easier to handle in the equations and later in this
chapter it is shown, how these are determined from the true humid air flow.
The humid air is a mixture of mv mass of water vapour and mass mdryAir
of dry air. FF eed is the flow rate of the feed in ( Ls ) and Fpowder is the flow
of powder out of the system( Kg
s ). The enthalpy H is a composite energy of
the internal energy of the constituent atoms and the flow work associated
with forcing streams in and out of a system against a pressure. It has the
J
). The enthalpy of mixtures such as the humid
unit energy per unit mass( Kg
drying medium, the feedstock etc. is the sum of the partial enthalpies of
the components and a residual enthalpy term which for example takes into
account the heat of mixing. However, in this report the influence of the

MAIN INLET DRYING AIR

Modelling a Spray Dryer

Fmain
Yin
Hmain
Tmain
Cdry air
Cvapour

Flow
Absolute humidity
Enthalpy
Temperature
Specific heat capacity dry air
Specific heat capacity vapour

FEED IN
Kg/s
Kgmoist/Kg dry air
KJ/Kg
C
J/Kg"K
J/Kg"K

Feed (liquid)

Ffeed
Xin
Hfeed
Tfeed
Csolid
Cwater

Moisture content
Enthalpy
Temperature
Specific heat capacity solid
Specific heat capacity water

Fout
Yout
Hout
Tout
Cdry air
Cvapour

Flow
Absolute humidity
Enthalpy
Temperature
Specific heat capacity dry air
Specific heat capacity vapour

Flow

Kg/s
Kgmoist/Kgsolids
KJ/Kg
C
J/Kg"K
J/Kg"K

OUT LET AIR

AMBIENT AIR

SPRAY
DRYER
WEEL MIXED

Temperature
Rela!v humidity

SFB INLET DRYING AIR

Flow
Absolute humidity
Enthalpy
Temperature
Specific heat capacity dry air
Specific heat capacity vapour

KJ/Kg
C
J/Kg"K
J/Kg"K

OUT LET PRODUCT

Kg/s
Kgmoist/Kg dry air
KJ/Kg
C
J/Kg"K
J/Kg"K

Powder out

Fout
Xout
Hout
Tout
Csolid
Cwater

Flow
Moisture content
Enthalpy
Temperature
Specific heat capacity solids
Specific heat capacity water

Kg/s
Kgmoist/Kg solids
KJ/Kg
C
J/Kg"K
J/Kg"K

22

FSFB
Yin
HSFB
TSFB
Cdry air
Cvapour

Kg/s
Kgmoist/Kg dry air

Figure 4.4: Basis Blockdiagram- variable description for the system used in energy and mass balance equation. The block describes the
input and output for air and product. The block shows the true flows of the air flow and feed flow. In the calculations these
are modified into dry flows for easier use.

4.2 Steady State Mass and Energy Balances for Spray Dryers

23

residual enthalpy is neglected as this value often is very small compared

to the enthalpy of the main components ( 1%). With this definition the
humid air enthalpy for both inlet and outlet airflow is defined in 4.2:

HhumAir = HdryAir + Y Hvapor

(4.2)

Y is the absolute humidity or moisture content in the

by the

 air expressed
Kgvapour
mv
relation between mass of vapour and dry air mdryAir , KgdryAir . From being
YIn at the inlet it increases during the spray drying operation to Yout . In
terms of specific heat the enthalpy is given by:
HhumAir = CdryAir (Tair Tref ) + Y ( + Cvapor (Tair Tref ))

(4.3)

Kj
where CdryAir is the specific heat capacity of dry air ( KgK
), Cvapor is the

Kj
. The heat capacity is defined
specific heat capacity for water vapor KgK
as the energy required to raise the temperature a unit mass of a substance
by a unit temperatur. The specific heat capacity is temperature dependent,
however, it is convenient to use mean values for this parameter, which is
the heat capacity evaluated at the arithmetic mean temperature for a given
temperature range. This has been used through the entire project. Tair is
the air temperature(o C), Tref is the reference temperature. (0o C), is used
as the reference temperature at which there is zero enthalpy. is the latent
heat of vaporization, which is the heat required for water to change from
liquid- to gas phase (vaporize).

(4.3) is a simplification since it is assumed that the final enthalpy is independent of the vaporisation path, accordingly the vaporisation is assumed
to take place at (0o C) at which the enthalpy is chosen to be zero and then
superheated to the air temperature Tair . Originally to reach vapour state,
the vaporisation occurs at the dew point temperature, which is the temperature at which the air become saturated and then heated up to the final air
temperature. This becomes of practical importance for the calculations if
water
the absolute humidity is above 0.05 Kg
kgdry (Mujumdar (2007)). The enthalpy of the feed entering the dryer is the sum of the enthalpy of the dry
solid and the moisture liquid in the product
Hf eed = Csolid (Tf eed Tref ) + XIn Cwater (Tf eed Tref )
Hpowder = Csolid (Tpowder Tref ) + Xout Cwater (Tpowder Tref )

(4.4)
(4.5)

where Csolid and Cwater are the specific heat capacity of dry solid and water.
XIn/out is the solids moisture content and is based on a unit weight of dry

24

Modelling a Spray Dryer

water
product ( Kg
Kgsolids ) . The reason for using a dry basis in the equations above
for the air and powder moisture content is that the flow rates of the dry
air and the dry solids is the same at both the inlet and outlet, which makes
the calculations more straightforward as the moisture now is directly related
to the dry substance. (4.5) for the enthalpy of powder leaving the dryer is
similar to the enthalpy of feed (4.4). It is assumed that all the moisture
evaporated from the feed is absorbed by the outlet air and taken out of the
dryer. Hence the moisture content in the final powder can be related to the
outlet humidity of the dryer by a mass balance.

4.2.1

Mass Balance

The mass balance over the spray dryer relates the moisture entering the
dryer with the outgoing moisture and gives (4.6). Due to the assumption of
a well mixed dryer and equilibrium state between outlet air and solids, the
outlet moisture content of the powder X0 is expected to be the equilibrium
moisture content of the solid in the respective air conditions. The outlet
absolute humidity Yout is then isolated.

FM ainDry YIn + FSF BDry YIn + MsIn XIn = FOutDry YOut + MsOut XOut
(4.6)
FM ainDry + FSF BDry = FOutDry
MsIn = MsOut
Ms (XIn Xout ) = FOutdry (YOut YIn )
Ms
(XIn Xout )
YOut = YIn +
FOutdry

(4.7)

(4.7) is inserted into the previously stated energy balance equation, which
results in (4.8). Hereafter the unknown and unspecified parameters are:
Primary and secondary inlet airflows, FM ain and FSF B respectively and the
belonging air temperatures, TM ain and TSF B , the moisture content of the
air going into the system YIn , the moisture contents of the feed XIn and the
final product XOut , and the in- and outlet solids rate MsIn and MsOut .
FM ainDry (CdryAir TM ain + YIn ( + Cvapor TM ain ))
+ FSF BDry (CdryAir TSF B + YIn ( + Cvapor TSF B ))
+ Ms (Csolid Tf eed + XIn Cwater Tf eed )
Ms
(XIn Xout ))( + Cvapor TOutair )
= FOutDry (YIn +
FOutDry
+ CdryAir TOutair ) + Ms (Csolid Tpowder + Xout Cwater Tpowder )

(4.8)

4.2 Steady State Mass and Energy Balances for Spray Dryers

TOutair = Tpowder

25

(4.9)

In our present situation, the temperature of the outlet drying air and the
resulting moisture content of the final product are the variables that are
required to be estimated. As it can be seen from the two equations above
these variables are influenced by the input operational variables and material
parameters outlined in the block diagram in figure 4.4. In the calculation
process the input operational variables to the plant are assumed to be known
and predetermined. Thus by first determining the outlet powder moisture
content, (4.7) can be solved for the outlet drying air humidity. Taking the
assumption into account that the temperature of the gas and the product is
similar, this temperature can be computed from (4.8).

4.2.2

Equilibrium Moisture Content

The moisture content of the outlet powder is approximated to be the equilibrium moisture content. The equilibrium moisture content is the resulting state of an interaction between the environment and the substance, to
which the moisture content of the substance converges to either by moisture
uptake(adsorption) or by drying(desorption). It is noted that equilibrium
moisture content may vary depending on whether the substance is exposed to
adsorption or desorption. So changes in the moisture content of a substance
are dependent on the surrounding partial vapour pressure and temperature
condition, but also on the nature of the solids. After an adequate amount of
time has passed with steady state condition an internal moisture diffusion
balance takes place until the equilibrium moisture content is attained. Thus
for the vapour pressure at a given temperature the substance will have a
state where it will neither gain nor lose any moisture.
This relationship between the equilibrium vapour pressure and the moisture
content in the substance can be presented by a moisture sorption isotherm
function. This sorption isotherm designates the equilibrium moisture content for a certain humidity value, at a constant temperature and hereby
gives a description of a products ability to bind water. Due to the complexity of the sorption process, the isotherm cannot be determined analytically,
but instead measured experimentally. Different products and materials have
different hygroscopic properties, which is affected by their molecular structure and their solubility. There are various empirical relations describing
the sorption characteristics for food ingredients using different models in
literature.
The desorption isotherm for maltodextrin, which is the test material that has
been used in this project, is determined in (Jes
us M. Fras and Schittkowski

26

Modelling a Spray Dryer

(2001)). Maltodextrin is a polysaccharide and consist of dextrose(glucose)

molecules connected in a chain of variable length. The length of the molecule
chain is described by a DE(Dextrose equivalent) number and explains its
properties(flavour). The sorption isotherm found in (Jes
us M. Fras and
Schittkowski (2001)) is for maltodextrin DE 12 and the maltodextrin used
in our test setup is of type DE 10. The difference was discussed with a
chemist from GEA NIRO and it was found acceptable to use the sorption
isotherm for maltodextrin DE 12.
As explained there will be a slight dissimilarity in moisture binding properties though it is insignificant and can be neglected. The equilibrium moisture content model is based on Guggenheim-Anderson-de Boer (GAB) model
equation, as recommended by the European Project Group COST 90 on
the Physical Properties of Foods 1 for the characterisation of water sorption
in food materials.

Xeq (T, aw ) =

Ceq Keq Weq aw

(1 Keq aw )(1 Keq aw + Ceq Keq aw )

(4.10)

Where the model parameters Wm , Ceq and Keq are determined by (Jes
us
M. Fras and Schittkowski (2001)). All the parameters are dependent on the
temperature of the solid in celsius.

1257.14
)
Tsolid + 273
144.57
= 0.65exp(
)
Tsolid + 273
99.27
= 0.05exp(
)
Tsolid + 273

Ceq = 0.04exp(

(4.11)

Keq

(4.12)

Weq

(4.13)

To characterise equilibrium vapour pressure in the sorption isotherm the

relative humidity content of the drying air is applied, as the vapour pressure
in the solid is equal to the partial vapour pressure in the drying air when
no more moisture can be lost to the surroundings and thus be equilibrium.
This is also known as the water activity of the product aw . The equilibrium
moisture content model is derived from desorption isotherm measurements
performed at four different temperatures (4o C, 25o C, 37o C, 50o C). Since
it is expected that the spray dryer will work at higher temperature rates,
the behaviour of the model is tested at higher temperatures. In figure (4.5)
a graph is produced in which the equilibrium moisture content is plotted
1

http://www.esf.org/ (2/10-09)

4.2 Steady State Mass and Energy Balances for Spray Dryers

27

against the relative humidity for various temperatures. If the relative humidity of the surrounding air is close to zero, then the equilibrium moisture
inside the dry product also is nearly zero independent of the temperature.
At higher temperatures a larger variation in equilibrium moisture content
is noted. The model is unrealiable for water activities above 0.9. For large
values of water activities values (4.10) gets negative.
Desorption Isotherm for Maltodextrin DE12 at various Temperatures
4oC

0.9

25oC
0.8

37oC

Equilibrium Moiosture Content (KgH O/Kgsolid)

50oC

0.7

65oC
85oC

0.6

100oC
115oC

0.5

increasing T
0.4
0.3
0.2
0.1
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Water activity a

Figure 4.5: Desorption Isotherm Maltodextrin DE12: Equilibrium moisture content as function of water activity for temperatures between 40 C and
1150 C. If the relative humidity of the surrounding air is close to
zero, then the equilibrium moisture inside the dry product also is
nearly zero independent of the temperature. At higher temperatures
a larger variation in equilibrium moisture content is noted. The model
is unrealiable for water activities above 0.9

The relative humidity of the vapour gas mixture is measured as the fractional saturation with moisture and is defined as the ratio of the partial
vapour pressure Pv to the saturated pressure Psat at the same temperature:

Pv
Psat

(4.14)

For drying to take place the relative humidity of the surrounding drying air
must be lower than the water activity of the product. The partial vapour

28

Modelling a Spray Dryer

pressure is related to the absolute humidity of the surrounding air and is

recalled to be the ratio of vapour mass mv to mass of dry air mdryAIR . Using
the gas law for the two fractions at constant temperature T and total volume
J
V results in (4.15). R is the universal gasconstant (8.314 (molK)

Pv V
Mw
RT
PdryAir V
mdryAir =
MdryAir
RT
Mw
Pv
Y =
PdryAir MdryAir
mv =

(4.15)

Daltons law (4.16) states that the total pressure exerted by mixture of gas
is equal to the sum of the partial pressures of each fraction in the mixture
g
and knowing the molar mass of water is Mw = 18.01 mol
and and dry air
g
MdryAir = 28.96 mol (4.15) becomes:

Ptotal = Pv + PdryAir
Pv
Y = 0.622
Ptotal Pv

(4.16)
(4.17)

The outcome of rearranging the above written equation is the partial vapour
pressure as function of absolute humidity Y, since Ptotal is assumed to be
equal to the standard atmospheric pressure : 101325 Pa(Langrish (2008)).

Pv =

Y
( 0.622
)Patm
Y
)
1 + ( 0.622

(4.18)

At 100 % relative humidity, the partial vapour pressure equals the vapour
pressure of liquid and the drying air or the surface of the substance is said
to be saturated with vapour. There are many formulations to calculate the
saturation vapour pressure.(4.19) (Richard Shelquist (2009)) used here is
simple with only 3 parameters and still offers good results when compared
to the Smithsonian reference table for vapour pressure found in (Wiederholt
(1997)). A deviation of 1 percent at high temperatures (above 100o C) and
much less for lower temperatures from reference value is acceptable, as this
will only have a very small effect on the relative humidity calculation and
equilibrium moisture content (see appendix A).

4.3 Test on a Multi Stage Dryer- MSD20

29

C1 T

Psat = 100 C0 10 C2 +T

(4.19)

C0 = 6.1078
C1 = 7.5
C2 = 237.3
With the system of equations put forward in this section, the temperature
and moisture content of the outlet air, Tout and Yout respectively can be
determined. To that, the moisture content of the final product is estimated.
The procedure is written in a Matlab script and solved with Matlab.

4.2.3

Steady State Solution

There are six unknown variables (Xout , Yout , , Pv , Psat , and Toutair ) and six
corresponding equations ((4.19), (4.18), (4.14), (4.10), (4.8) , and (4.7))
which means the system has a unique solution that is found by the following
iterative process.
1. X0 is initialised to have the same value as XIn , moisture in the feed
2. Yout is calculated using equation (4.7)
3. ToutAir is solved for equation (4.8), where Tpowder is set to be equal to
ToutAir .
4. Pv is calculated using equation (4.18)
5. Psat is calculated using equation (4.19)
6. is calculated using equation (4.14)
7. Xout is calculated using equation (4.10). Hereafter the process is returned to the second step and a new outlet moisture value Yout is
calculated. This is reiterated until the process converges to the final
values. In the Matlab script the process stops when the difference
between the previous determined temperature and the presently calculated temperature is less than (1/1000).

4.3

Test on a Multi Stage Dryer- MSD20

With the purpose of being able to validate the correctness of the dynamics
and the steady state values of the created model, a test has been prepared

30

Modelling a Spray Dryer

and completed on a real spray dryer system. The tests was basically about
putting in step changes at the most important input operation variables and
examine the resulting step responses for some significant output variables.
Such a test on an actual system would depict the known but also the hidden
dynamics of the system and thus explain its behaviour for certain changes
in the system.
The test was completed on a Multi Stage Dryer(MSD)-20 open cycle2 system
at GEA NIROs test station in Soeborg, Denmark. Maltodextrin DE 10 was
used as the material to be spray dried. As drying gas atmospheric air was
applied with the use of an electric heater. The system setup was as shown
in figure 4.6.
The drying chamber has a diameter of 2 m and a height of 2.30 m, with
a total volume of approximately 10.3 m3 . Compared to the largest spray
dryers used in the dairy industry which can be up to 16 m in diameter with
a total volume of 1920 m3 , this is a small one. This spray dryer can produce
kg
approximately 70 hour
of powder.
The input variables which are chosen to be manipulated are the most significant ones and have the greatest effect on the spray drying process. That is
the feed flow rate, given by sensor 1626, the Main inlet airflow and temperature, (sensor 1701) and 1702 respectively. Similarly the SFB inlet airflow
(1703) and temperature (1704) are controlled. In the end the airflow into
the Vibrofluidizer is operated, but this is only commented superficially given
that the focus in this project has been on the drying chamber and the process
in it. The central output variables are the outlet air temperature (1709),
outlet powder particle size and residual moisture content. (see table in app.
C.1 and figure 4.6)
Furthermore in the figure the variables are marked as either manually (M ) or
automatically (A) controlled. Automatical control indicates that a variable
is controlled by a PI regulator when the specific variable value is set. For
the manually controlled variable an operator sets the controller output value.
As described previously, under normal circumstances the moisture content
of the powder leaving the dryer, is controlled indirectly by maintaining a
constant outlet drying air temperature through the regulation of the speed
on the feed rate pump.
Throughout the test this feedback loop was disconnected, so the main system functioned as an open loop and the feed rate was controlled manually
with pumps 1639 and 1606. The inlet air flows and temperatures were automatically controlled, likewise the pressure inside the chamber. This is
to prevent the chamber from crumpling up due to the air flows. The PI
2
In a open cycle spray dryer the drying medium is not reused. Air enters one place
and exits at another place

4.3 Test on a Multi Stage Dryer- MSD20

31

parameters are shown in the logbook from the test in appendix C.3. The
Sensor Name
MAINKGH
T1702
SFBKGH
T1704
F1626

Manipulated variables
Description
Main air flow into chamber
Temperature of MAINKGH
SFB air flow into chamber
Temperature of SFBKGH
Feed flow into chamber

Control
Auto
Auto
Auto
Auto
Manual

Unit
Kg/h
oC
Kg/h
oC
L/h

Table 4.1: The manipulated variables in test of the dynamic process(MSD-20).

Auto= automatic control (PI), manual: Manual control

blue lines in figure 4.6 depict air flows, while the yellow lines show product
flow. A cyclone was used to filter out the fines from the outlet air. These
were returned at the top of the chamber to be applied in the agglomeration
process. The bag filter shown was disengaged.
Each variable is controlled in a number of steps within a maximum and
minimum range(step up and step down) which is predetermined by the Test
engineers at GEA Niro, so the system stays in stable and workable process states. The standard values indicate the value the variables are kept
at in normal operation. The variables are changed one at a time, while all
the others are kept constant and the entire dynamic process is recorded.
It is essential that nothing else is modified throughout the test, thus only
the response due to the specific step change is observed. However, disturbances both measurable and immeasurable are still present which can affect
the responses. As disturbances the relative humidity of the ambient air
(measured by sensor 1618), fluctuations in feed concentration and height of
powder layer in SFB can be mentioned (1706).
The feed that was used in the test was a mixture of 50% solids and 50%
water with variations within the measurement uncertainty (2%). Five tanks
of mixture were used during the entire test and with no stirring of the feed,
which contributes as a small disturbance. In an industrial dryer there is
often a preheater/evaporator on the feed input, which is advantageous not
only from a bacteriological point of view, but it also decreases the viscosity
to improve atomizer performance. Moreover, increasing the solids concentration in the feed is a more energy effective, since less moisture has to be
evaporated in the spray dryer.

4.3.1

The Test on MSD-20

The total number of test steps are 22 included those for the VFB. From
rehearsal tests (24/6-2009) the settling time for the responses of the output

32

Modelling a Spray Dryer

Manipulated
Sensor Name
MAINKGH
T1702
SFBKGH
T1704
F1626

variables Step range

Min Default Max
1600
1800
2000
150
160
170
350
500
600
80
90
100
65
75

Table 4.2: The manipulated range of the variables in test of the dynamic
process(MSD-20). The standard values indicate the value the variables are kept at in normal operation. Min and Max values are the
step values used for test. In total there was 22 test steps included
those for the VFB

variables for step changes was experienced. From this the test period for a
step change in Feed rate and Main inlet air was estimated to have a length
of 1 hour and 30 minutes as it was expected that the responses would have
reached their final states. For SFB and VFB the length of the test period
was 1 hour, since their influence on the final output values are small and
less time is required for the response to settle.
The experiment was carried out in a continuous process ( 30 hours) in
order to avoid spending unnecessary amount of time on both system start-up
and terminating the process. All the variables are sampled and logged every
second and every tenth second. The moisture content and particle size of
the outlet powder though is measured off-line, for this reason a sample of the
powder from the SFB outlet and the VFB outlet is taken every fifth minute
and send to the laboratory. The result from the laboratory is an average
particle size and moisture content. Triple tests in the laboratory of samples
verified, that the results from the used method for determining the moisture
content has a measurement uncertainty of 1 percent when tests are repeated.
For the particle size determination the uncertainty is 2 percent. The powder
data is linearly interpolated in the intervals between two samples, in this
way the data can be applied with the one or ten second measurements of
the system.
The tests were completed successfully, despite small adjustments on an oscillating controller and on the height of the powder layer in the SFB which
is an indirect description of the powder residence time in the SFB. This is
controlled by setting the time on how often powder is sent out from the SFB
through a lock.
The results are used in the further study of the spray dryer and for model
verification in the steady state and dynamic modelling (see section 4.2 and
5.4), where the results are examined and discussed. Furthermore they are

4.4 Steady State Calculations

33

used for estimating a model for the prediction of the powder moisture content
by system identification (see section 7). Since some of the steps have been
repeated, some part of the tests could be applied as estimation data and the
rest of the test data is used as validation data for the identified model. The
entire test program and graphical plots of the time responses are found in
appendix C

4.4

Steady State Calculations

It is desired to compare the results derived from the model with the results
from the experiment on the real system MSD-20(see section (4.3) for test
description). Hence the same operation conditions, as were used on the
real plant, were applied to the steady state model to create similar process
conditions. The operation parameters and the specific values for the plant to
work at a default normal state are explained in table 4.3. It should be noted
that the flow rates are given in mass per hour. These have to be divided by
3600 sec to get the flow rate in seconds. As drying gas atmospheric air is
applied and the solvent in the feed is water.
As stated previously the air and feed flow rates in the mass and energy
equations are based on dry gas and dry solids flow, which are calculated
from the initial moisture content in air YIn and feed XIn . The dry air flow
rate is decided by determining the ratio of mass of dry air to the mass of
total air as displayed in (4.20).
1
Ff low
YIn + 1
Ms = FF eed f eed Sconc

Fdry =

(4.20)
(4.21)

Likewise the dry solids flow is estimated from the amount of solid, Sconc ,
in the feed (4.21). The feed flow process value is given in Ls on the real
plant and therefore also used in the model. This is converted into kg
s by
multiplying the feed flow with the density of the feed, f eed . Since it has not
been possible to find any functions describing the density of the maltodextrin
that was used in the experiment, the density function is approximated, after
consultancy with Chemists from GEA NIRO, to be the same as for milk
concentrate without any fat and is given in (4.22). The total feed density
is a combination of solids amount in percentage of total feed S, the density
of lactose (Carbonhydrate)D1 , and density of water D2, which are both

34

Modelling a Spray Dryer

Variable Name
FM AIN
Tmain
FSF B
TSF B
Ff eed
f eed
Sconc
Tf eed
Tamb
amb
Solid
XIn
Patm
YIn
Cdryair
Cvapour
Cdryair
Cwater

UChamber
AChamber

Default Operation Values

Description
Value
Main air flow IN
1800
Temperature of MAIN
160
SFB air flow IN
500
Temperature of SFB
90
Feed flow IN
65
Density of feed
1.208
Solid concentration
0.5
Temperature of feed
50
Temperature of ambient air
30
Rel. humidity ambient air
28
solids of total feed
50
Initial Moisture Content
1
Standard Atmospheric pressure
101325
Absolute Humidity air IN
0.0079
Specific Heat Capacity
Dry air
1
vapour
1.8
Maltodextrin
1.5
water
4.2
latent heat of vaporisation
2500
Qloss
heat transfer coefficient (chamber) 16.75
Surface area(chamber)
26.2

Unit
Kg/h
oC
Kg/h
oC
L/h
Kg/L
Kg/Kgf eed
oC
oC
%
%
Kg/Kgsolid
Pa
Kg/Kgdryair
KJ/(Kg K)
KJ/(Kg K)
KJ/(Kg K)
KJ/(Kg K)
KJ/(Kg)
KJ/m2 /h
m2

Table 4.3: The manipulated variables default operation values for test of the
models. These are equal to the operation values used at the test of
MSD-20.Note that the flow rate are not on dry basis, but the total
amount(sum of Moist and either dry air or dry solids.)

functions of the feed temperature Tf eed (Refstrup).

f eed =

1
S( D1

100
1
D2
)+

(4.22)

100
D2

D1 = 1.635 0.0026 Tf eed + 2 105 Tf eed 2

4

D2 = 1.0020825 1.14 10

(4.23)
6

Tf eed 3.325 10

S = Solids in %

Tf eed

(4.24)
(4.25)

The initial moisture contents XIn is known as the ratio of water to solids in
the feed.
XIn =

water
0.5 kg
kg feed
solid
0.5 kg
kg feed

(4.26)

4.4 Steady State Calculations

35

Moreover the table shows the ambient air humidity and temperature condition that are used in the model calculation. Obviously these circumstances
changed during the 1 21 day experiment as it can be seen on the plot of the
temperature and humidity in appendix C.2.10. Therefore an approximated
average value for ambient temperature and relative humidity are chosen for
the calculations of the absolute humidity in the air going into the chamber,
YIn ((eq. 4.14), (4.19),and (4.7)).

4.4.1

Steady State Results

The model has been exposed for the same tests and changes on the process
variables as for the experiment on the real spray dryer and followed the same
test procedure. At the outset the model is at the default state given by the
parameters in table 4.3. For each test a process variable was changed, while
all the other variables were kept at default state. The resulting estimate of
the outlet air temperature compared to the results from the tests on MSD20 is shown in table 4.4. To this point the heat losses from the spray dryer
have been neglected in the energy balance equation. Nonetheless, as it is
observed from the results, the temperature calculated without any heat loss
is significantly higher than the temperatures measured on MSD-20.
In view of the fact that the model is ideal and the drying is complete (all
possible moisture is evaporated), the drying course is not the possible reason
for the higher temperature at the outlet, as more energy cannot be used on
evaporation and thus decreasing the temperature. It is expected that the
lower temperatures from the experiment to some extent can be attributed
to the heat losses through the spray dryer outer cladding and structural
supports. Another possible reason for the lower temperatures in the real
experiment is due to the cooling air for the air disperser at the top of the
chamber which has been omitted in the model. Moreover the system is
modeled as a CSTR, in which the temperature is well mixed through the
entire chamber. However, on the real system the outlet temperature is
measured at the exit of the chamber, while at the inlet air entrances, the
temperature is a little higher. Thus there are some small regions in the
chamber that will have a higher temperature than the rest of the chamber,
which the model does not take into account.
The heat loss is expressed by the standard heat transfer equation in (4.27)
and added to the right hand side in the energy balance equation (4.8).
Qloss = Uchamber Achamber T
T = TOutAir Tamb

(4.27)
(4.28)

The heat transfer coefficient Uchamber for properly insulated drying chambers

36

Test
Description
Default System
Ff eed = 75 L/h
Tmain = 150 o C
Tmain = 170 o C
FM AIN = 2000 kg/h
FM AIN = 1600 kg/h
FSF B = 600 kg/h
FSF B = 350 kg/h
TSF B = 80 o C
TSF B = 100 o C
Tamb = 50 o C
Tamb = 10 o C
RHamb = 75 %
RHamb = 10 %
Sconc = 0.8 Kg/Kg
Sconc = 0.2 Kg/Kg

Modelling a Spray Dryer

Steady State Results
No Loss
Loss Included
TOutAir
TOutAir
Temperature Temperature
101.5 o C
86.1o C
o
95.3 C
80.6o C
94.2 o C
79.9o C
o
108.9 C
92.4o C
o
106 C
90.9o C
96.3 o C
80.6o C
o
101.1 C
86.2o C
102.3 o C
85.9o C
o
99.5 C
84.4o C
o
103.6 C
87.9o C
102.1 o C
86.7o C
o
101.3 C
85.9o C
102 o C
86.7o C
o
101.4 C
85.9o C
o
124.7 C
105.7o C
78.6 o C
66.7o C

TEST
TOutAir
Temperature
86o C
81o C
79o C
90o C
89o C
84o C
83o C
82.5o C
85.6o C

Table 4.4: Steady State Results for the drying air temperature TOutAir calculated
with the operation variables values as used in the real test on MSD20, with an energy loss function included and without a loss function.
These are compared with the results from the Test on MSD-20

Kcal
with mineral wool or similar materials is in the order of 1-2 Kcal
(1 (m
2 h) =
m2 h
4.187 KJ). For non isolated drying chambers the heat transfer coefficient is
. The heat transfer coefficient is in the model chosen to be
around 5-7 Kcal
m2 h
,
although the MSD-20 spray dryer is non-isolated and preequal to 4 Kcal
m2 h
sumably have a higher heat transfer coefficient than the chosen value. But
to compensate for the complete drying of the powder, which gives the lowest
possible drying air temperature due to more energy being used for evaporation, the lower value is selected. Including this into the calculations of the
iterative process with a loss equal to the initial temperature of the chamber,
reveals the modelled outlet temperature to be much closer to reality.

The reason for the grand effect of including heat loss is explained by the
energy levels of the various components. The energy level lost through
the chamber wall is high compared to energy from the incoming feed and
the outlet powder, which is the reason for its great effect on the the air
temperature in the chamber ToutAir (see tabel 4.6).
In table 4.5 the resulting humidity and moisture contents are listed.

4.4 Steady State Calculations

Steady State Results for humidity
Loss Included Loss Included
Test
Eq Moist.
YOut
Description
Content
Abs. Hum
Default System
0.0033
0.0251
Ff eed = 75 L/h
0.0047
0.0277
o
Tmain = 150 C
0.0045
0.0251
Tmain = 170 o C
0.0024
0.0251
FM AIN = 2000 kg/h
0.0025
0.0237
FM AIN = 1600 kg/h
0.0047
0.0267
FSF B = 600 kg/h
0.0032
0.0244
FSF B = 350 kg/h
0.0035
0.0263
TSF B = 80 o C
0.0036
0.0251
TSF B = 100 o C
0.0030
0.0251
o
Tamb = 50 C
0.0051
0.0397
Tamb = 10 o C
0.0026
0.0198
RHamb = 75 %
0.0050
0.0386
RHamb = 10 %
0.0026
0.0199
Sconc = 0.8 Kg/Kg
7.75e-4
0.0148
Sconc = 0.2 Kg/Kg
0.0135
0.0354

37

TEST
Abs. Hum
Outlet Air
0.010
0.012
0.011
0.010
0.010
0.0115
0.012

Table 4.5: Steady State Results of absolute humidity and equlibrium moisture
content calculated with the variables and values as used in the real
test on MSD-20. Calculated with a energi loss funtion included. These
are compared with the results from the Test on MSD-20. The absolute
humidity is calculated from sensor 1616 and 1614.

4.4.2

Effect of varying Operation Variables

Feed Flow
It is observed that the outlet air temperature is inversely proportional to the
feed flow rate. An increase in the feed flow decreases the temperature and
vice versa. This is explained by the fact that an increase in feed flow results
in more water to be evaporated which requires a lot of energy. This results
in an increased absolute humidity value in (4.7) and the relative humidity
increases, which causes the moisture content in the final powder to increase.
(See table 4.5)
Feed solids content
The change in feed solids content resulted in a change in outlet temperature
due to the same reason as for the feed flow rate. Decreasing the solids content
in the feed obviously gives rise to an increased amount of solvent (water)
per unit weight of feed that requires more energy to evaporate. From the

38

Modelling a Spray Dryer

Energy level of the components (Default)
Energy
Test
KJ/s=KW
FM AIN Inlet
90.33
Ff eed Inlet
3.10
Fpowder Outlet
1.4
Fout outlet
96.8
Qloss
10.5

Table 4.6: Energy Level of the components at default state. Energy in the air is
very high compared to the other parts. However it is noted that the
energy lost in the chamber compared to the energy in the feed and
powder is high as well, and therefore has an important effect on the
temperature in the chamber ToutAir . The energy is shown in KJ
s .

large changes in temperature it is noted that evaporation of water is energy

expensive. Therefore from an energy perspective it is wished that the feed
has maximum solids content when it is fed to the spray dryer, but still is able
of being completely atomised. However sometimes it is necessary to run the
spray dryer with a low solids content in the feed and low outlet temperature
in order to avoid any degradation in powder quality or to meet a specific
product specification .
Concerning the feed temperature, and increase in the feed temperature reduces the heat required to dry the product, which means the temperature
in the chamber is higher than at default state. Moreover an increased feed
temperature reduces viscosity which makes the feed easier to atomize.
Inlet Air Flow and Temperature
It is from the results table noted that the changes in outlet air temperature
is proportional to the changes in the operation condition for the inlet air.
But the Main inlet air has more effect on the system than the SFB, due
to the higher flow rate and temperature. Increase in any one of the inlet
air flow rates or temperatures signifies an increase in the amount of energy
available for evaporation, and since there is no change in the amount of water
to be evaporated the energy used for the evaporation will be the same. As
a result there is energy in excess which is seen as a rise in the outlet air
temperature. Likewise the temperature falls in the chamber when one of
the operation condition values is reduced. As stated before the absolute
humidity depends on the flow rates ,(4.7) , consequently a fall in the flow
rate causes the absolute humidity to increase.
Generally it can be stated that the lower the outlet air temperature is,
the more effective utilisation of the heat, however the lower temperature
increases the moisture content in the outlet powder. The explanation for

4.4 Steady State Calculations

39

this is that the differences in the outlet air temperature affect the saturation
vapour pressure (4.19), which descends with a falling temperature. So air
with a lower temperature can contain less water. For a constant absolute
humidity, that is the vapour pressure is constant, it means an increase in
relative humidity. The moisture content in the final product will also increase
as seen in the equilibrium moisture contents in table 4.5.
Ambient Humidity
The ambient humidity is viewed to have a minor impact on the temperature
of the air in the chamber. Nevertheless it affects the spray dryers ability to
dry a product. As a consequence of the increased humidity level in the ambient air, the resulting moisture content in the outlet product is increased. It
does not seem to be a problem in this case, but for other products, which are
more hygroscopic, it can be difficult to reach the wanted moisture content
in the powder, if the ambient humidity is too high. In such case, dehumidification of the inlet air is necessary prior to use or the plant capacity can be
reduced.
Ambient temperature
The ambient temperature affects the moisture content of the inlet air. At
constant relative humidity, as in the test, a higher ambient temperature
results in an increase of moisture content of the drying air. Simultaneously
the higher ambient temperature has a positive effect on the heat loss given
that heat transfer through the chamber wall will be less significant when the
temperature difference between wall and ambient temperature is small.
The results of the steady state model have revealed that it is possible to
estimate the outlet air temperature using mass and energy balance equations. However, the model is ideal and describes a complete drying of the
product, such that gas and particles reach an equilibrium state. This is also
noted when the moisture content of the air from the real test is compared
to moisture content from the model. The humidity level in the model is
somewhat higher, due to the complete drying which means more vapour is
released into the drying air.
From having evaluated the steady state results a dynamic model of the spray
dryer is described in the subsequent chapter.

40

Modelling a Spray Dryer

Figure 4.6: MSD-20 Test Station Setup for test of the dynamic process on maltodextrin (21/7-2009). The red boxed marks the manipulated variables. The green boxes are disturbances. The arrows show the direction of flow. Blue line is the air flow. Yellow Lines are feed / product
flow. A cyclone filters the outlet air for fines and returns these into
the chamber

4.4 Steady State Calculations

Figure 4.7: Test Centre at GEA Niro - Multi stage Dryer MSD-20

41

42

Modelling a Spray Dryer

Chapter

Dynamic Modelling of a Mixed

Flow Spray Dryer

In the previous section a steady state model of the spray drying system was
deduced. It is able to estimate the final drying conditions for certain step
changes of the system and its drying environment. However, in an actual
system the drying environment and conditions vary with respect to time.
Under these circumstances a steady state model is insufficient. A dynamic
model, which describes the responses of the system over time for certain
changes in the settings is necessary. This also essential with regards to the
selection and simulation of control systems for the spray dryer.
The main setup of the drying chamber model is similar to the one made
use of in the previous section as seen in figure 4.4. But in the dynamic
model the difference between the input and out flow expresses the rate of
accumulation of a component in the system; this could be mass or energy.
The accumulation rate represents the rate of change in the total mass or
energy of the system with respect to time. At steady state there is no rate
of change and this term (time derivative) will be equal to zero: flow rate in
is equal to flow rate out.
A dynamic model of the temperature changes over time in the spray drying
chamber is derived from the unsteady state energy balance model given for
CSTR process. It is still assumed that the internal energy is the dominant
contribution to the total energy. However it is more convenient to work with
enthalpy to describe energy. Hence the chamber is of constant volume and
the drying air is analysed as an ideal gas, the energy balance equation is

44

Dynamic Modelling of a Mixed Flow Spray Dryer

given in (5.1).
mC

X
X
dT

=
FjIn HjIn
FjOut HjOut + Q + W
dt

(5.1)

The enthalpy equations for the air (Main and SFB) (4.3), feed (4.4), and
final product flow (4.5) known from the steady state model are still valid
and are used in the above written equation. Q denotes the rate of heat
is the rate of work done on system. Yet the
added to the system and W
influence of the variables on the system are ignored given that the chamber
is not externally heated and the pressure is kept constants in the chamber.
Thus there is no additional work done on the chamber. Isolating the time
derivative of the temperature, results in a model specifying the temperature
development over time in the chamber (5.2).
1
dTOutAir
=
(FM ain (CdryAir TM ain + YIn ( + Cvapor TM ain ))
dt
mtotalAir C
+ FSF B (CdryAir TSF B + YIn ( + Cvapor TSF B ))
+ Ms (Csolid Tf eed + XIn Cwater Tf eed )
FOutDry (YOut ( + Cvapor TOutair ) + CdryAir TOutair )
Ms (Csolid Tpowder + Xout Cwater Tpowder ))

(5.2)

Where mtotalAir is the total mass of air in the chamber as a mixture of dry
air and vapour. At the initial state, without any flow entering or leaving the
dryer, this value is calculated from the density of the ambient air pressure
and the volume of the chamber. C is the specific heat capacity of the moist
and dry air in the chamber.

5.1

Total mass of air in the Spray Dryer

The total air density is simplified to be the sum of dry and moist air densities, neglecting the errors due to non-ideal compressibility of gas and using
the ideal gas law ((5.4))(Node (2009)). The density equation is derived in
appendix D. Thus the density is a function of the total pressure P, vapour
pressure Pvapour of the air and temperature T in Kelvin (5.4). From the
equation it is noted that an increase in vapour pressure at constant temperature causes the air density to decrease. This is due the fact that the molar
mass of water is smaller than for dry air.
totalair = dry + vapour
Pv
P Pv
+ Mw
= MdryAir
RT
RT

(5.3)
(5.4)

5.1 Total mass of air in the Spray Dryer

45

The volume of the spray drying chamber is determined by using (5.5) (Refstrup and NIRO),where Dchamber symbolises the diameter of the chamber;
Hchamber , the height of the chamber; ACeiling angle of the chamber ceiling,
and ACone is the angle of cone on chamber.

Dchamber 2 Hchamber
4
1
Dchamber 3
1
)
+
+
(

ACone
(90
ACeiling ))
24
tan(
tan( 180 2 )
180

V =

(5.5)

With the density and the volume of the chamber at hand the mass of air
at the beginning of the process is computed. The mass is separated into
the mass of dry air and mass of moisture in the air, which is determined by
using the same approach as in section 4.2.3 for calculating the inlet dry- and
vapour air flow.
mChamberAirStart = totalair Vchamber
mChamberAirStart = mdryAir Start + mvapour Start
1
mdryAir Start = mChamberAirStart
YIn + 1
mvapour Start = mChamberAirStart mdryAir Start
When the process has begun, the total mass of drying gas in the chamber
is equal to the accumulated change of air mass added to the air mass at the
process start, as in (5.6). The flow rates are, as in the steady state model,
calculated on the dry air basis and assumed to be identical at the inlet and
outlet which means the mass of dry is constant. This implies that a change
in mass of air is due to the increase or decrease of vapour in the air only.
The rate of the change in vapour mass is expressed by vapour entering and
leaving the chamber with the airflows and the air vaporised from the feed
(5.7). All flow rates and the vaporisation rate are described by kg/seconds.
mtotalAir = mChamberAirStart + mvapour

(5.6)
dmvapour

= FM ainDry YIn + FSF BDry YIn + Ms (XIn XOut ) FOut YOut

dt
(5.7)
mvapour Start + mvapour

YOut =
(5.8)
mdryAir Start
By dividing the total amount of vapour mass with the unchanged amount
of dry air the absolute humidity in the chamber is estimated, which is the
absolute humidity of the outlet air YOut as well due to our assumption of the
chamber being well mixed. The effect of this is that the total outlet air flow
rate varies according to the amount of vapour, and thus the vapour pressure,

46

Dynamic Modelling of a Mixed Flow Spray Dryer

which causes a change in chamber air temperature. An increase in outlet

air flow results in less energy accumulated in the chamber and therefore the
temperature in the chamber will decrease. On the contrary there will be an
increase in temperature with a decrease in outlet flow.
With a description of the change of total mass of air in the chamber,(5.2),
can be solved to estimate the temperature development over time. But first
the outlet moisture content XOut has to be determined from the equilibrium
moisture content from (4.10), which is dependent on the relative humidity
of the chamber, which can be found from (4.14). This equation is a function
of the temperature of the chamber and the absolute humidity. This is solved
by an iterative procedure, setting the initial state of the temperature. On
the block diagram in figure 5.1 this basic dynamic model is illustrated.

Figure 5.1: Basic Dynamic Model: The model of the chamber is connected to the
relative humidity equation and the equilibrium moisture content. The
figure shows the main operational variables processed in the spray drying chamber and the output variables that are inputs to the relative
humidity equation and moisture content blocks.

As stated above the flow rates and vaporisation rates in the model are dekg
. In the model it is therefore assumed the feed particles are
scribed by seconds
heated and dried such that vaporisation process is completed within a second from the time the feed enters the dryer (vapour change Ms (XIn XOut )
(5.7)). To validate this hypothesis and estimate the general drying times
and the heat and mass transfer equations between single feed particles and
drying gas are examined.

5.2 Drying Kinetic Mechanism

5.2

47

Drying Kinetic Mechanism

The MSD dryer is the preferred dryer for creating agglomerated powder. But
due to lack of information regarding the statistical properties for agglomeration, like the collision frequency of particles in the spray dryer and how often
these are combined into larger particles or destroyed into smaller pieces, only
the evaporation process of the single droplets entering the chamber before
they get agglomerated are examined. This follows our earlier assumptions
that all droplets are of equal size, homogenous and spherical.

Figure 5.2: Particle Morphology: When a particle is dried it can end up having
different surface forms, which is dependent on the spray dryer design, its setup and the operation conditions.Depending on the spray
dryer the particles can be combined into various structures: single
particles,agglomerates, granules, or Pellets(layered structure)

Solid particles can have different sort of behaviour when reacting with gas.
Some have an unchanged size and some shrinks over time in various ways.
For the drying process for a single particle a shrinking core model is used,
as it is accepted as being the best simple model that describes the reaction
between gas and solids, even though it does not precisely represents the
mechanisms of gas-solid reactions (Levenspiel (1999)) and (Shunji Homma
and Matsumoto). Here it is visualised that the reaction first occurs at the
surface of the particle and then the reaction moves into the particle leaving
behind dried inert solid. Thus the unreacted moist core shrinks while the
shell thickness increases.
The evaporation of the water from the atomized droplets to form dry particles involves simultaneous heat and mass transfer ( (figure 5.3). The heat

48

Dynamic Modelling of a Mixed Flow Spray Dryer

Figure 5.3: Schematic of Drying mechanism

and mass transfer is a function of temperature, humidity, diffusion properties of the surrounding air, relative velocity between droplet and air, and
droplet diameter. The process is described by two drying periods. In the
first period of drying the moisture content in the particles is assumed to be
evenly distributed and the moisture is removed at constant rate. Heat is
transferred by convection from the air to the droplets and converted into
latent heat during moisture evaporation. This heat transfer rate is given by
a heat transfer coefficient hheat and driving force calculated as the difference
between the drying air temperature and the particle temperature , (5.9).
= hheat (TOutair Tf eed )
Qheat

(5.9)

m
= Kmass (Psurf ace Pv )

(5.10)

The vaporised moisture is transferred into the air by convection through the
boundary layer that surrounds each droplet. This vapour flux in this period
is expressed by an external mass transfer coefficient Kmass and a vapour
pressure driving force (vapour pressure difference between the drying air
Pv and pressure at droplet surface Psurf ace , which is saturated). Due to
capillary and diffusion mechanisms moisture migrates to the surface from
the interior of the droplet at a rate sufficient to maintain saturation on the
surface. The shrinkage model is ideal and it is therefore assumed that the
droplets remain perfectly spherical and the droplet solution is homogenous
in this period. The change in particle volume corresponds to the amount of
water evaporated.

5.2 Drying Kinetic Mechanism

49

Figure 5.4: Drying Proces of a Particle with a Shrinking Model.First period the
particle shrinks. Second period the core shrinks and no change in
particle volume.

When the critical moisture content XC is reached within the droplet, the surface wetness cannot be maintained which results in a porous crust formation.
This acts as some resistance to water vapour diffusion and consequently the
drying rate will fall. The drying rate is estimated as vapour diffusion from
a moist core through a dry shell. In this second drying period the particle
can change its formation to any type shown on figure 5.2. However here the
spherical form is kept and when the specified surface condition is reached,
the volume of the particle does not change, but the crust thickness increases
as moisture content decreases. Because the vapour pressure at the surface
descends to the vapour of the surrounding air the moisture content in the
crust is the equilibrium moisture content with the surrounding drying air
humidity.

5.2.1

Mass Transfer Rate

For a spherical particle the mass transfer equation given in (5.11) is developed by (K.H. Clement and Thomsen (1991)). It is a combination of the

50

Dynamic Modelling of a Mixed Flow Spray Dryer

vapour transfer from the moist core to the particle surface and the external
vapour transfer at the surface and describes the mass transfer rate per unit
of the particle surface area (Kg/(s m2 )).
mtransf
er =

P Mw
ToutAir +Tf eed
R
2

2Def f
ddrop (f +

2Def f
kmass ddrop )

ln(

P Pv
) (5.11)
P Psat (Tf eed )

f =0
;
for Xparticle > Xcr 1st drying period
1/3

Xparticle Xeq
1 for Xparticle Xcr 2nd drying period
f=
Xcr Xeq
(5.12)

Ddrop is the diameter of a single droplet.

P is the pressure in the chamber(standard air pressure).
The f factor characterizes the resistance to vapour diffusion through
the crust, which only has an influence in the second drying period
where the average moisture content of the particle Xparticle < Xcr
Until the critical moisture (Xcr ) content is reached, (5.11) is a description
of the external mass transfer resistance only. The fraction in the f factor
is an explanation of the ratio of the core volume to the particle volume by
volumetric averaging of the moisture content (relation between the moisture
content left (Xparticle Xeq ) since crust formation began (Xcr Xeq )). So
the inverse of this illustrates crust thickness. This simplification is possible
due to the assumption of a spherical particle and shrinking core model:
d3
4
Vsphere =
3
2

3
Xparticle Xeq
dcore
=
Xcr Xeq
ddrop

(5.13)

At the commencement of drying the droplet has a diameter of DdropInit .

During the first drying period the diameter of the droplet decreases, equal
to the amount of water evaporated, due to the assumption of ideal shrinking
(5.14)(see appendix D.12). In the second drying period the diameter of the
droplet is constant as given in (5.15).

Ddrop

6Ms (XIn Xparticle )

water


6Ms (XIn Xcr ) 1/3
3
= DdropInit
water

Ddrop =

DdropInit 3

1/3

(5.14)
(5.15)

5.2 Drying Kinetic Mechanism

51

It is noted that the mass transfer model also presents an adjustable parameter, coefficient of effective vapour diffusion through the particle crust Def f ,
which is dependent on the solid material and describes how well moisture is
diffused through this.
The average mositure content of the particle Xparticle is determined by setting up a water balance model as in eq.eq:waterbalance, which
 describes
 the
kgmoist
rate of change in moisture content calculated on dry basis kgdrySolid .
Ms

5.2.2

dXparticle
= Ddrop 2 mtransf
er
dt

(5.16)

Mass and Heat Transfer Coefficients

The mass transfer coefficient Kmass and heat transfer coefficient hheat are
estimated from the Nusselt number, Nu and the Sherwood number Sh (dimensionless). The Nusselt number, Nu, is a measure of the heat transfer
occurring at the droplet surface and is the ratio of convective to conductive
heat transfer defined in (refeq:nusselt). Kair in this equation defines the
thermal conductivity of air (W/(m K)) and by regression analysis of data
given from (Box (2009)) the thermal conductivity as function the temperature is estimated (See appendix D.3).
The Sherwood number is the mass transfer equivalent to the Nusselt number
and characterises the ratio of convective mass transport to diffusive mass
transport and expressed in (5.18). Dair is the diffusion coefficient of water
vapour in air and is obtained by regression curve fit to data from Bolz and
Tuve (Nellis and Klein (2009)). It is a function of the air temperature in
Kelvin and is given in (5.19).
hheat Ddrop
(5.17)
kair
kmass Ddrop
Sh =
(5.18)
Dair
Dair = 2.775 106 + 4.479 108 (Tair + 273) + 1.656 1010 (Tair + 273)2
(5.19)
Nu =

The external transfer coefficients are isolated in the equations above (5.20)
and the Nusselt and the Sherwood number are obtained from the RanzMarshall correlation (5.22), in which they are determined from the Reynolds

52

Dynamic Modelling of a Mixed Flow Spray Dryer

number, Re, Prandtl number, Pr, and the Schmidt number ,Sc.

kair
Ddrop
Dair
= Sh
Ddrop

hheat = N u

(5.20)

kmass

(5.21)

N u = 2 + 0.60Re1/2 P r 1/3
1/2

Sh = 2 + 0.60Re

1/3

Sc

(5.22)
(5.23)

The Reynolds number provides a measure of the ratio of inertial forces (resistant to change or motion) to viscous forces, which for the air and droplet
gives equation (5.24) (Benson (2009)). V is the relative velocity between
air
the air and particle and air is the absolute viscosity of the air. air
, which
is known as the kinematic viscosity is determined in appendix D.5. Due
to lack of information about the relative velocities between air and particle,
it is assumed due to the small size of particles that the relative velocity is
almost zero ( 0.01m/s) (particles follows the air see appendix E) (Shabde
(2006)).
The Schmidt number is a dimensionless number known as is the proportion between the kinematic viscosity and the mass diffusivity (5.25). The
Prandtl number is analogous to the Schmidt number and represents the ratio
of kinematic viscosity to thermal diffusivity air . In table 5.1 the parameters are estimated within the temperature range 300 C 1200 C and with a
constant droplet size Ddrop = 76m. It is noted that the values are small,
which means that their contribution to the Nusselt and Sherwood number
in (5.22) is small. This demonstrates that the viscous forces do not have
a big influence and the main mechanism in the heat and mass transport is
conduction and diffusion.

air V Dd rop
air
air
Sc =
air Dair
nuair
Pr =
air
Re =

(5.24)
(5.25)
(5.26)

5.2 Drying Kinetic Mechanism

53

Ranz-Marshall correlation Parameter values

Parameter
Value
Re
5.22 3.42
Sc
61.82 62.42
Pr
71.22 69.92
Table 5.1: Ranz-Marshall correlation Parameter Values. For droplet Ddrop =
76m and in the temperature range 300 C 1200C. Relative Velocity between particle and air is estimated to 0.01m/s.

It is observed from the equations in (5.20), that the transfer coefficients vary
with the size of the droplet. A decrease in droplet size implies an increase in
the transfer coefficients. This indicates faster heat and mass transfer, which
in the end gives quicker drying times

5.2.3

Droplet size

Droplet and dried particles are by no means of equal size as assumed in

the calculations with a constant droplet diameter, but have a range of sizes
characterized by their size distribution. To portray the size distribution in a
spray dryer, various mathematical functions have been suggested. However
the complex nature of feed atomisation and the subsequent drying makes it
difficult to fit the size distribution to a mathematical function. The most
common distributions are:
Normal distribution (Both rotary and nozzle atomisers)
Log-Normal distribution (rotary atomisers)
Square Root Normal distribution (Nozzle atomiser)
Empirical distribution function
Nukiyama-Tanasawa distribution (Nozzle atomiser)
Rosin Rammler distribution (Nozzle atomiser)
Some distributions give a better fit for nozzle atomisers while others are
better for describing rotary atomisers (Masters (2002)). The MSD-20 operates with a pressure nozzle, but there is no information available about
its particle size distribution. This would have made it possible to identify
the particle sizes that should be used in the examination of the feed drying
time.
Instead the mean particle size is estimated from a function which is based
on the nozzle dimension(Orifice diameter), atomizing pressure, spray angle,

54

Dynamic Modelling of a Mixed Flow Spray Dryer

nozzle capacity, and feed characteristics(viscosity and density), given by

GEA NIRO (OLK and NIRO). The nozzle which is used in the calculation
is of the type : 4931 cof which has an orifice diameter of 2.2mm and the
spray angle is chosen to be the standard 65 degrees. The viscosity is decided
to be 20 centipoise(cp)1 , which is approximately the viscosity of milk. An
increase in viscosity increases the droplet size.
The atomising pressure is determined from the pressure that was used through
the test on MSD-20(sensor 1616. see appendix C.2.5). The feed rate is controlled by setting this pressure. As seen in the plot in appendix the feed
rate increases when the pressure is increased and decreases correspondingly
when pressure is decreased. The droplet size is inversely proportional to
the pressure but at the same time the increase in feed rate has the opposite
effect on the size. The pressure is chosen to 240 bar.
This results in a mean droplet size equal to 76m, which is within the
expected range given that pressure nozzle is known to be able to produce
droplet in the range 10 300m. Moreover the program estimated that 98
percent of the created particles have a particle size less than 298m(3Ddrop ).
The mean droplet size found here is chosen to de be the default particle size
in the following tests of the drying time.

5.3

Matlab Simulink Implementation

The dynamic model has been in implemented in Simulink/Matlab. Two

separate models have been implemented: the model of the spray dryer and
a model of the drying kinetics for a single particle. These two models are
at present time not combined, thus the drying times for a particle and the
evaporation rate is not directly used in the model of the spray drying chamber. Nonetheless the temperature and humidity results from the spray dryer
model are used to estimate the drying time for a single particle. For the spray
dryer model process variables, feed characteristics, ambient air characteristic, spray dryer chamber sizes can be varied. A block diagram illustrating
the model construction in Simulink. The blocks show what they use as input
and give as output.5.5

5.4

Test:Dynamic model

In this section the dynamic model of the spray drying chamber describing
the outlet air temperature ToutAir given in (5.2) is tested and compared with
1
visocosity is dependent on temperature. Viscosity of water at 200 C = 1 cp and olive
oil = 84 cp (Chieh (2009))

FEED SUBSYSTEM

OUTPUT
Feed flow l/s
Ndrop

Surface area (droplet)

Mass solid

INPUT
Ddroplet

More precise model

(not included in this project)

Tout air
Rela!ve humidity
Equilibrium moisture content
Tfeed
OUTPUT
Mass evaporated

OUTPUT
Abs. humidity
Tout air

Moisture content Xpar!cle

MASS
TRANSFER PARAMETERS
INPUT
Ddroplet
Tout air

DRYING
CHAMBER
INPUT
Chamber air (ini!al)
Feed rate
Feed temperature
Main flow
Main temperature
SFB flow
SFB temperature

EQUILIBRIUM
MOISTURE
CONTENT
INPUT
Water ac!vity
Tout air

5.4 Test:Dynamic model

INPUT
Feed Kg/h
Diameter droplet

MASS TRANSFER RATE

SINGLE PARTICLE

RELATIVE
HUMIDITY
INPUT
Abs. humidity
Tout air

Rela!ve humidity chamber

OUTPUT
Kmass

55

Figure 5.5: Block Diagram of the dynamic model.The 3 block to the left describes the particle drying model. The three blocks on the
right hand side describe the spray dryer model. They are at present time implemented separately in matlab. Combining
the two models will give a more precise white box model of the system.The physical description will be exact. But very
difficult the drying of each particle precisely.

56

Dynamic Modelling of a Mixed Flow Spray Dryer

the results from the experiment on the MSD-20 spray dryer. The model is
exposed to the same changes in the operation variables as for the experiment
on the real system, for which the test program can be found in appendix C.
Only one operation variable is manipulated at a time while all the others
are kept constant at default state(B.1). Thus the dynamic response of the
temperature development in the chamber is examined for step changes for
the following operation variables:
1. FM AIN Main inlet air flow rate
2. TM AIN Temperature of main inlet air flow
3. Ff eed Feed flow rate
4. FSF B SFB inlet air flow rate
5. TSF B Temperature of SFB inlet air flow
Generally it is not possible to manipulate the operation variables directly,
but through another device such as a heater or a fan for the air inlets and a
pump for the feeding system. Each one of the elements is normally controlled
by a PI controller. When the set point is exposed to a step change, it takes
some time for the operation variable to settle at the correct value. For that
reason the changes in the operation variables are simplified and modelled as
being a first order system with a time constant .
This value is approximately determined as being the time constant for the
process value in the experiment(see fig G.1 in appendix). The static gain,
which is the relation between the stationary value of the output and the
input is 1 for the systems controlling the operation variables. Both the
original step change and the first order step change have been modelled to
examine the effect, the delay has on the model and its response.
1st order system
1
s + 1

(5.27)

Start-up of a spray drying process and to get the process in a stable state
takes time and requires experienced technicians. However this is kept at
a basic level in the model since the main focus has been on modelling the
spray dryer in a running state and to estimate the effect when it is exposed
to changes. Therefore as an initial condition the moisture content of the air
in the chamber Yout , is set to be equal to the moisture content at steady
state, for the default operation conditions. The steady state value is also
applied as the initial temperature of the outlet air TOutAir .

5.4 Test:Dynamic model

57

In addition to the included heat loss through the chamber cooling air is
included in the model which is used for the air disperser. This supplementary
air is an additional energy term (4.3) in the dynamic model (5.2). Moreover
it contributes to the total amount of vapour in the chamber, exactly like the
other inlet air (5.6) and the total amount of dry air leaving the chamber.
The step change is applied to the system after 1000 seconds, while the development of TOutAir , Relative humidity in the chamber, and equilibrium
moisture content Xeq data is recorded for the analysis. In all tests the step
change has been applied directly on the manipulated variable. The results
are compared and validated against the test results from the MSD-20.

5.4.1

Step: Temperature of main inlet air

The inlet air temperature is at default operation state 1600 C. First a step
change down to TM ain = 1500 C is accomplished, where after a simulation
with a step change up to TM ain = 1700 C is performed. The step change and
the resulting response of the temperature TOutAir for a decrease in main inlet
air temperature is plotted in figure 5.6, top and bottom plot respectively. For
comparison the result from the similar experiment on the MSD-20 is shown
in the same figure (red). It is seems that correlation between the main inlet
air temperature and outlet air temperature is high, since the temperature
response of the outlet follows the changes in the operation variable very
precisely (oscillation). The outlet air temperature is directly proportional
to the inlet air temperature.
For a direct step change in the main inlet temperature (blue) it is observed
that the model responses too fast compared to the results from the true
spray dryer. However, when the manipulated variable is modelled as being
the step response of a first order system with a time constant (green) the
outcome appears to be more realistic. The time constant is here chosen
to be 180 sec, which is approximated by examining the rise time for the
variable in the spray dryer(time it takes the electric heater to increase the air
temperature). Thus after 180 seconds the heater would have reached 63.2%
of its final value. The model reaches the new steady state after 1000 sec. Due
to the oscillations in the real dryer the settle time is longer. The oscillations
is due to an incorrectly tuned PI controller to the heater. Probably the
proportional gain was set too large, which results in the overshoot.
Besides the imperfections in the settling time, which may be solved by a
correction of the time constant in the first order system describing the heater,
it can be concluded that the model gives a good estimate of the outlet air
temperature for a change in the main inlet air temperature. Modelling the
heater as a first order system seems sufficient. The figure shows that control

58

Dynamic Modelling of a Mixed Flow Spray Dryer

signal has a fall before it starts to rise. This is because of a small mistake
by the technician when changing the value of the feed pump. Furthermore
it is observed that the inclusion of the cooling air has reduced the outlet air
steady state temperatures with approximate 0.50 C.
A similar behaviour of the response is observed when the main inlet air
temperature is stepped up from 1500 C to 1700 C The response for a step up
in temperature is found in appendix 5.6.
Main Air Temperature step Tmain=160>150 oC

Temperature [oC]

165

Model: No delay
Model: with delay =180
TEST 6 MSD20

160
155
150
145
140

1000

1200

1400

1600

1800

2000

2200

2400

2600

2800

3000

Time [sec]

Temperature TOut Air for Main Inlet air Temperature change

Model: No delay
Model: with delay =180
TEST 6 MSD20

Temperature [oC]

86
84
82
80
78
1000

1500

2000

2500

3000

3500

4000

4500

5000

Time [sec]

Figure 5.6: Dynamic step response of the TOutAir for decrease in main inlet air
temperature from 1600 C to 1500 C. The top figure shows the applied
step change. Heater is a first order system = 180. The bottom
figure is step response. Model results are compared with data from
MSD-20

5.4.2

Step: Feed Flow rate

The effect of changing the feed flow rate is tested in two cases: increase in
feed flow rate from 65 Lh to 75 Lh and for a decrease in feed flow rate from
75 Lh to 65 Lh . The results from the steady state model proved that the outlet
temperature is inversely proportional to the feed flow rate. The dynamic
response of the outlet air temperature for the step change increasing feed flow
is illustrated in figure 5.7. The feed pump is modelled as a first order system

5.4 Test:Dynamic model

59

with a time constant = 5, which react much faster to changes compared

to the heater. It is seen that the responding feed flow rate builds up in the
same way as in the real system. Correspondingly the model estimates the
temperature response of the outlet air to be fast (150 sec. to reach steady
state). On the contrary, the result from the MSD-20 demonstrates that when
the increased amount of feed enters the dryer the outlet air temperature first
reacts as expect and falls some degrees with the expected rate. But hereafter
the temperature response decreases in speed, thus it takes more than 3000
seconds to reach steady state.
Feed flowrate step Ffeed=65>75 L/h

Feed Flow [L/hour]

80

75

70

65

Model: with delay =5

TEST 2 MSD20

60
900

950

1000

1050

1100

1150

1200

1250

1300

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for a Feed Step Change
87

Model: with delay =5

TEST 2 MSD20

Temperature [ C]

86
85
84
83
82
81
80

1000

1500

2000

2500

3000

3500

4000

4500

5000

Time [sec]

Figure 5.7: Dynamic step response of the TOutAir for a increase in feed flow rate
L
from 65 L
h to 75 h . The top figure shows the applied step change.
Pump is a first order system = 5. The bottom figure is the step
response. Model results are compared with data from MSD-20(red)

The much faster response from the dynamic model can be explained by the
fact that it is assumed the spray dryer behaves completely as a CSTR, which
means the feed particles are evenly distributed in the chamber immediately
after the entry into the dryer. However, in a spray dryer the trajectories
of the particles are much more complex as seen in figure E.1 in appendix.
It is seen the main part of the feed falls straight down in the chamber and
stays down in the SFB for agglomeration. Thus the effect of the increased
amount of particles in the lower part of the chamber might be a reason for
the slower response time.
The increase in feed results in an increase of the moisture content in the
air inside the chamber. This affects the agglomeration rate, which creates
larger particles. A large particle means longer drying times, which implies

60

Dynamic Modelling of a Mixed Flow Spray Dryer

less water is evaporated and less energy is used within a specific time interval.
The slower the evaporation process takes place the longer will it take to reach
steady state, which is what is observed in the MSD-20. Thus the assumption
of drying times of a second and the lack of the dynamics describing the longer
drying times and the change in the amount of particles to be dried are the
main reasons for the deviation between the model and the real spray dryer.
A step change down in feed flow rate shows a similar behaviour with a long
settle time in appendix B.2.
Generally it can be concluded that changes in feed flow rate can be made
very quickly due to the fast response of the feed pump (5 sec) and a change
in temperature will be observed after less than 100 sec. However the time
to reach a stable steady state is longer, compared to a change in main inlet
air temperature for which the heaters response time is 180 sec but the settle
time is only 1000 sec. This has to be considered when the control system is
developed.

5.4.3

Step: Main inlet air flow rate

The main inlet air flow rate is increased to 2000 Kg

h for which the response
of Tout is shown in figure 5.8 and stepped down again back to 1800 Kg
h (result
is found in appendix B.3.1). The fan is modelled as first order system with
the time constant = 40. Hereby the change in flow rate appears to follow
the true change in flow rate.
Nonetheless the dynamics of the recorded responses for the true spray dryer
seems to behave somewhat different than expected by the model. First of
all the explanation for this deviation may be found in the flow pattern of
the air. The model assumes the dryer to be well mixed which is not true
in reality. However the most important factor may be the change in drying
condition and drying time of the particle as for change in feed rate. A change
in air flow inversely affects the moisture content of the air inside chamber,
but the rate of change in humidity condition is too fast compared to the
reality.
The results of the variations in the SFB variables are given in appendixB.3.1.
The responses are the same as for the main inlet air, although their effect on
the total system is much less due to the small amount of energy that enters
the system this way. This is expected since the main purpose of the SFB is
to control the agglomeration process and final product in the fluid bed.

5.4 Test:Dynamic model

61

Main Air inlet flow step Fmain=1800>2000 kg/h

2050

Air Flow [Kg/h]

2000
1950
1900

Model: No delay
Model: with delay =40
TEST 9 MSD20

1850
1800
1750
900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for Main Inlet air Temperature change

Temperature [oC]

92

90

88

86

Model: No delay
Model: with delay =40
TEST 9 MSD20

84
800

1000

1200

1400

1600

1800

2000

2200

2400

2600

2800

3000

Time [sec]

Figure 5.8: Dynamic step response of the TOutAir for increase in Main inlet air
Kg
flow from 1800 Kg
h to 2000 h . The top figure shows the applied step
change. Fan is a first order system = 40. The bottom figure is step
response. Model results are compared with data from MSD-20

62

5.5

Dynamic Modelling of a Mixed Flow Spray Dryer

Test:Drying Time for particle

Figure 5.9: Mass tranfer rate and drying time test setup. The red block is the
Mass tranfer rate system. Orange blocks are mass transfer rate coefficients and droplet characteristic. Blue box are process variables as
step functions.

5.5 Test:Drying Time for particle

63

The drying time is evaluated as being the time from evaporation begins
and until it ends when the droplet reaches equilibrium moisture content.
The examination is accomplished by implementing the mass transfer equation (5.11), with a changing diameter function during evaporation(red) into
Simulink as in figure 5.9. The operation conditions (blue), TOutair , RHchamber ,
and Xeq are the steady state values calculated in the previous chapter for
certain process inputs. These are also applied to calculate the transfer coefficients and the belonging parameters (orange). The simulink models of the
subsystems and function are illustrated in appendix F
The initial droplet size is by default determined to be 76m from the calculation in the previous section. This is, together with the feed flow rate,
used to calculate the mass of solid in a single droplet. The feed flow and the
droplet size is not linked, which means that changes in the feed flow does
not have an effect on the droplet size in the drying tests. The reason for this
is, that the main purpose with this test is to examine how certain process
conditions affect the drying time of a single particle and estimate the limits
for the dynamic drying chamber model.
Four different test cases have been completed. In each test one of the following parameters were varied in order to estimate the drying time for a
particle:
Droplet
Temperature
Feed
Main inlet air
Effective diffusivity Def f
critical moisture content XC
By varying these parameters information about the effect of changing process
conditions and the importance of the product characteristics is obtained.
The process condition for the experiments is the default state condition,
which was also used for the tests at steady state (app.B.1). At default state
the temperature Toutair = 86.10 C, RH = 6%, and the equilibrium moisture
kg
.
content Xeq = 0.0033 kg
2

The effective diffusivity constant Def f is as a standard selected to be 5e9 ms ,

which is an approximate value for the effective diffusivity for milk(empirically
determined in (V. S. Birchal a and Passos (2008))). The critical moisture
content for maltodextrin is in (I. Zbiciski and Prajs (2005)) estimated to be

64

Dynamic Modelling of a Mixed Flow Spray Dryer

0.54 kg
kg . These two parameters characterise the product that is dried and
therefore it seemed essential to examine the importance of the parameters.
With these parameter values used in the calculations, the drying process
for single particle in the default process condition is graphically plotted in
figure 5.10.

Particle Ddrop= 76 m
st

1 drying period

0.8

Moiosture Content (Kg

H O

/Kg

solid

Drying Time for Particle at Default steady state Operation

1

0.6

X critical moisture content

cr

nd

0.4

2 Drying period falling drying

rate

0.2

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Time [sec]

/Kg

solid

Drying Time for Particle at Default steady state Operation(Zoom)

H O

Moiosture Content (Kg

Particle D

0.9

= 76 m

drop

Constant Drying Rate

0.8
0.7

X critical moisture content

cr

0.6
0.5
0.4
0.3
2
st

10

Time [sec]

12

14

16

18
5

x 10

1 drying period

Figure 5.10: Drying time for a particle with a Ddrop = 76m at default operation
state(steady state). Moisture Content vs. time in seconds. The
figure is similar to the top, but zoomed in to show the change in
drying period. The 2 drying periods are illustrated. The first drying
period is at constant drying, almost. The second drying period starts
at Xcr which for maltodextrin is 0.54 kg/kg and has a falling drying
rate. It shows that the crust formation takes place almost instantly.
Drying ends when particle reaches the equilibrium moisture content.

The figure illustrates the moisture content, Xparticle ,in a particle as a function of time. As described the drying process has two periods. The first
period takes place until the particle reaches its critical moisture content and
crust is formed. This happens almost instantly, since there is no resistance
for the moisture transfer and the mass transfer coefficient value is relatively
high (see app F). In the lower figure, which is a zoomed version of the top
figure, the constant drying rate is noted. When moisture content reaches

5.5 Test:Drying Time for particle

65

0.54 kg/kg the second period begins and drying rate falls significantly. It
takes about 0.3 sec for a particle of this size to reach equilibrium moisture
content.(see modification in sec 5.7)

5.5.1

Test 1: particle sizes

The effect of the particle size on the drying time is examined in this section.
This is done by computing the moisture content development over time for
particles of different sizes. The process conditions were kept at default state.
The outcome of the calculation is seen in figure 5.11.
Drying Time for Particles at Default steady state Operation
1

Particle D

= 57 m

drop

Particle Ddrop= 76 m
Particle D

= 114 m

drop

0.8

Particle Ddrop= 152 m

Limit

= 228 m

0.7

Particle D

0.6

Particle Ddrop= 304 m

drop

Moiosture Content (KgH O/Kgsolid)

0.9

0.5
0.4

Increasing
Particle
size

0.3
0.2
0.1
0

0.5

1.5

2.5

3.5

4.5

Time [sec]

Figure 5.11: Drying time for various particle sizes at default operation
state(steady state). Moisture Content vs. time in seconds. Drying times are increased for increasing particles size. Particles larger
than 150 m takes more than one second to dry. This means large
particle will not have completed the drying within 1 second as the
dynamic model takes for granted. This means less vapour in chamber and results in a higher TOutair temperature. It is recalled that
the dynamic model is ideal.

As anticipated the larger the particles are the longer does it take to complete
the drying. The dynamic model of the chamber condition expects the drying
to be completed within one second, in order to be exact. The consequence of

66

Dynamic Modelling of a Mixed Flow Spray Dryer

this is that the dynamic model is only correct for particle sizes below 150 m.
For larger particles for which the drying is incomplete after 1 second in the
chamber, this means less moist is evaporated to the chamber from them than
the model expected. They will naturally continue the drying in the chamber,
which theoretically should end up in the same amount moist evaporated to
the chamber at steady state. However, there is a possibility that some
particles leave the chamber before drying is complete, which means less
vapour in the chamber. Less vapour in the chamber is equal to the fact that
less moisture is evaporated and the consequence is a higher temperature in
the chamber due to less energy has been used on evaporation. In this case
the dynamic model is imperfect, since it models an ideal drying process.

5.5.2

Test 2: Temperature

Two methods has been employed for examination of the drying time for
varying temperature of ToutAir :
Changing the feed flow rate
Changing the Main inlet air flow rate
The feed flow rate has been changed to the standard test values(65 75 Lh )
but also to more extreme values to see the effect on drying times (35105 Lh ).
The results are plotted in a graph and shown in app. F.1.2.
It is known that an increase in feed rate causes the air temperature to
decrease and vice versa. As expected the drying time increases for lower
drying temperatures. For a feed rate of (105 Lh ToutAir = 660 C) the
drying time is now 0.35 second compared to 0.3 sec at default state for the
mean particle size, and seems not to be a problem. Moreover it is noted that
the increase in feed amount increases the equilibrium moisture content.
However the drying model is not fully describing reality since only the drying
of a single particle is examined. The higher feed flow rate will increase the
number of particles and as a consequence the possibility of collision between
the particles increases. This will for the most part give larger particles
and thus larger drying times. Larger drying times could be a problem in
some cases due to the stickiness of the product that could lead to wall
deposits and trouble at powder discharge. This topic has been neglected in
this project but more information into this found in (Masters (2002)) and
(Pilairuk Boonyai and Howes (2004)).
Similar behaviour of the drying times was observed when the drying temperature is changed by varying the main inlet air flow with a constant temperature. The result is shown in app. F.1.2.

5.5 Test:Drying Time for particle

5.5.3

67

Test 3: Effective Diffusivity

The significance of the effective diffusivity for the drying process is examined,
as this variable represents the type of product that is dried. The variable has
been varied by a factor of 10 both up and down. The operation conditions
were kept at default. The results of the simulations is presented in figure
5.12 which shown the moisture content in a particle over time.
Drying Time for Particle by varying the effective diffusivity D

eff
2

Deff= 5.9e9 m /s

0.5

D = 5.9e10 m2/s
eff

Deff= 5e8 m /s

0.45

D = 1e9 m2/s
eff

Deff= 3e9 m /s

/Kg

solid

0.4

D = 2e9 m2/s
eff

Moiosture Content (Kg

H O

0.35

0.3

0.25

0.2

Decreasing
Deff

0.15

0.1

0.05

0.5

1.5

2.5

Time [sec]

Figure 5.12: Drying time for various for a single particle for various effective diffusivity coefficients. Moisture Content vs. time in seconds. Drying
times are increased for decreasing Def f .

It is observed that the effective diffusivity has a great effect on the drying
process, which is expected as this describes the speed the liquid mass can
be transported in the specific product. This means that for a lower value of
Def f gives a longer drying time since it will take more time for the water
to diffuse through the particle, which is also seen in the results. Thus for
products with a low effective diffusivity, for which the particle will take more
time to dry than one second the dynamic model is not completely accurate
for the same reason as explained earlier.

5.5.4

Test 4: Critical Moisture Content

The critical moisture content is the point at which crust formation begins to
takes place. In this test the point of critical moisture content is controlled
for the dried product, thus to examine what effect the point at which the

68

Dynamic Modelling of a Mixed Flow Spray Dryer

critical moisture occurs has on the drying time. For maltodextrin this value
is Xcr = 0.54, this value is both decreased and increased in the test. The
operation conditions are kept at default state. The particle size is the mean
size determined earlier 76m. The result of the simulations is shown in
figure 5.13.
Drying Time for Particle by varying the critical mositure content X

cr

xcr=0.54
xcr=1.00
xcr=0.75

0.8

x =0.25
cr

0.7

x =0.10
cr

Moiosture Content (KgH O/Kgsolid)

0.9

0.6
0.5
0.4

Increasing
X

0.3

cr

0.2
0.1
0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Time [sec]

Figure 5.13: Drying time for various for a single particle for various critical moisture contents. Figure shows Moisture Content vs. time in seconds.
Drying times are increased for products with a higher critical moisture content value.

5.6

Summary: tests

It can be concluded from the results, the before the particle reaches critical
moisture content the longer will it take to complete the drying of the particle.
Apparently the reason for this is the increase in resistance which begins
much earlier for higher critical moisture contents due to crust formation. In
appendix F.5 the development of the resistance during the drying process is
shown. When the critical moisture content is reached the resistance f starts
to increase and converges to a final value when the moisture content goes to
equilibrium.
Modelling the particle as being perfectly spherical is a way of keeping the
model simple as the vapour diffusivity through the crust will be the same
from the boundary of the moist core to the surface in the entire particle and
thus the resistance of the crust in all directions is simplified. In the article
(Stephen R.L. Werner and Paterson), it has been tried to model the drying

5.7 Modifications in the Dynamic Model

69

of a collapsed(shrivelled) particle. An effective diffusivity coefficient and an

appertaining constrictivity factor, which accounts for the structural influence
on the pore system is estimated from experimental data. Compared to the
ideal shrinkage model the model of a shrivelled particle model is slightly
more realistic as expected, but also more complicated because it is difficult
to measure and predict the surface stress on particles and requires more
research in this field.

5.7

Modifications in the Dynamic Model

In the previous sections the equipment model, describing the dynamic changes
in the spray dryer, and a single particle drying model has been evaluated
and compared to data from a real spray dryer. The particle model demonstrated that the drying time of a single particle is dependent on the size of
the particle, but also on it chemicals and physical properties, especially the
effective diffusivity.
The test of the dynamic model of the spray drying chamber illustrated that
changes in drying condition affects the evaporation rate and that it had an
immense effect on the outlet air temperature response. Thus the assumption,
that all the moist is transferred to the surrounding drying air in a second,
does not completely describe the true process. Therefore few modifications
are made to the dynamic model of the temperature condition in chamber,
which will be described in this section.
From the test of the drying times for single particles it was found out that a
decrease in the effective diffusivity increases the drying time of a single particle. It has come to knowledge that the effective diffusivity for maltodextrin
0
of the type DE 10 is in the range 8 1011 m
s at 80 C, and even less for
smaller temperatures (J.G. Baez-Gonz
alez and Vizcarra-Mendoza (2004)).
This is nearly 100 times less than first estimated. The drying time for particles with the new effective diffusivity constant is estimated and illustrated
in figure 5.14.
It is immediately noticed that the total drying time for the particle has
increased, in view of the fact that transportation of liquid through the crust
now is slower. For the default droplet size Ddrop = 76 the drying time is
now 20 seconds compared to a drying time of less than one second before.
Due to agglomeration of the particles the possibility of larger particles is
present. Thus for droplets with a size of 300 the drying size is about
300 seconds. However collision of particles, both dry and wet, can vary the
drying time and the dynamics of the drying process.

70

Dynamic Modelling of a Mixed Flow Spray Dryer

Drying Time for single Particle of different sizes for D =8e11

eff

0.7

Particle D

= 76 m

drop

Particle Ddrop= 150 m

0.6

Particle D

= 300 m

drop

0.5

Moiosture Content (KgH O/Kgsolid)

Particle Ddrop= 400 m

0.4

0.3

0.2

0.1

0
50

100

150

200

250

300

350

400

450

Time [sec]

Figure 5.14: Drying Time for single Particle of different sizes for Def f = 8e
11 at default operation condition. The lower effective diffusivity is
for maltodextrin gives longer drying times. For a particle with a
diameter of Ddrop = 76 is 20 sec. For a particle with a diameter of
Ddrop = 76 is 20 sec.

5.7.1

Implementation of longer drying times and change in

evaporation rate

So far in the spray dryer model the evaporation of water is expressed to

happen in one second, Ms (XIn XOut )(difference between the moisture in
feed in and product out) in (5.7) which describes the addition of vapour from
the feed to the chamber per second. This is correct in steady state and when
the drying time is less than one second, however, variations in the drying
times also changes the time for the evaporation rate to reach steady state.
It is observed from the results (figure 5.7 & C.2.4) for the true spray dryer
that the temperature response for a variation in the feed rate appears to be
a composite of both a slow and fast dynamic. The fast dynamic comes from
the rapid change in the amount of feed entering the chamber, compared to
the slower dynamics that is a consequence of the slow evaporation as seen
for the drying time for a particle in figure 5.14.
A possible way to achieve this is to keep track of the number of particles
that enters the chamber and estimate how much each particle evaporates
per second until a steady state is reached, although it is very complicated.

5.7 Modifications in the Dynamic Model

71

To maintain the simplicity of the model, the amount of moisture evaporated

(Ms (XIn XOut )) is send through a first order dynamic system with a time
constant evap that describes the slow dynamic from the evaporation. Thus
this time constant evap is a function of the drying time of the particle,
agglomeration rate and residence time of particles in the chamber.
The fast dynamic is included by inserting a zero in the transfer function
while keeping the static gain to 1. Thus the transfer function G(s)evap can
be written as:
f ast s + 1
G(s)evap =
evap s + 1
f ast s
1
+
(5.28)
G(s)evap =
evap s + 1 evap s + 1
Now the system is given as the sum of two first order systems. Assessing the rewritten transfer function the effect of the included zero is easily
understood, when the initial and final value theorem is applied. These theorems can be used to estimate the initial and final value of a function in the
time domain by examining its Laplace transform, in the frequency domain.
Though this requires the poles of a transfer function are in the left half plane
so the system is stable. For this to be fulfilled evap has to be positive.
Thus the transfer function G(s)evap is evaluated by applying the theorems
on the step response Y (s) for this system. h0 is the value of the step.
Y (s) = G(s)

h0
s

(5.29)

Initial value theorem:

lim y(t) = lim sY (s) =

t0+

f ast
h0
evap

(5.30)

final value theorem:

lim y(t) = lim sY (s) = 1 h0

s0

(5.31)

It is noticed that in the beginning of the response, for s , the value of

the second fraction in (5.28) will go to 0, so only the first part of the system
has an influence on it, due to the zero in the nominator. Thus the zero has
an effect on the gain of the system and consequently how fast it reacts in the
moments just after the step has occurred f ast /evap . The larger the zero is
the larger the gain will be which means a faster response. However when
the time t the first fraction in (5.28) will go to 0 and the influence
of the second fraction increases. At steady state the gain in the system is
equal to 1(static gain) and the time to reach this only depends on evap .
To demonstrate that this system is able to give an acceptable description
of the change in evaporation rate, this is implemented in the model and

72

Dynamic Modelling of a Mixed Flow Spray Dryer

simulations are to be compared to the real data. By inspection of the temperature response of the outlet air temperature for MSD-20 when feed is
varied the time constant evap is estimated to be around 1000 sec (see fig.
5.7 & appendix C.2.4). f ast is guessed to be approximately half of the value
of evap . If the value is decreased the system will be slower and vice versa.
But a value larger than evap will result in an overshoot, since the gain will
be larger than one in (5.30). The simulation of the outlet temperature for
a step in feed rate from 65 Lh 75 Lh for the modified model and the similar
true data from MSD-20 is shown in figure 5.15.
Feed flowrate Ffeed step 65>75 L/h

Feed Flow [L/hour]

80

Model: with delay =5

TEST 2 MSD20

75

70

65

7000

7100

7200

7300

7400

7500

7600

7700

7800

7900

8000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for a Feed Step Change(Modified model)

Temperature [ C]

=1000 &

=500

=1300 &

=600

=1000 &

=100

=1000 &

=1500

evap

86

fast

evap

84

fast

evap

fast

evap

82

fast

TEST 2 MSD20

80

78
7000

7500

8000

8500

9000

9500

10000

10500

11000

Time [sec]

Figure 5.15: Temperature response of the outlet air for the modified model when
L
a feed step from 65 L
h to 75 h is applied. The top figure shows the
manipulated variable in the model (1st order) and the true dryer.The
figure below is the response for different pole and zero in G(s)evap .
It shows the effect of a zero too high or too low and the resulting
change in the dynamic. Best fit in this case compared to the true
data (purple) is for evap = 1300 and f ast = 600(light green)

The figure shows that the response from the model is reasonably in agreement with the true system for the correct values of the zero and the pole.
For this system the best fit is reached for evap = 1300 and f ast = 600.
The figure also shows how the value of the zero affects the dynamics of the
system. Figure B.12 in the appendix B.4 illustrates the output of the system G(s)evap , which is the evaporation rate per sec and the effect of various
zero- pole combination is seen. The change in the evaporation rate affects
the absolute humidity of the air in the spray dryer. This is also shown in
the appendix (B.11).
Moreover the effect of this modification on the other operation variables has

5.8 Summary: modifications

73

been studied and the result is seen in B.4.2. A change in the main inlet air
temperature is still well estimated. However changes in the air flow rate are
yet less precisely estimated. The dynamics of the air flow is complex and
needs further study to be modelled by a simple approach.

5.8

Summary: modifications

The change in evaporation rate is modelled by applying a first order system

with a zero and a pole. The modification that has been implemented in
the model results in a more exact estimate of the behaviour for a change in
feed rate. However, this part of the model is determined empirically and by
experience like a black box model in chapter 3. The consequence of this is
that the model is only correct for the specific spray dryer(MSD-20) and the
specific product. Thus an alteration of the spray dryer that is wished to be
modelled requires new experiments. In the long term time scale applying the
evaporation times of the particles and evaluate the agglomeration process
will make the model sustainable.

74

Dynamic Modelling of a Mixed Flow Spray Dryer

Chapter

6
Linearisation Analysis

For most control design and model analysis applications a linear time invariant model is needed. In this chapter the dynamic spray dryer model
of environment inside the dryer, described in chapter 5 is linearised and
analysed in order to prepare the model for control design.
Solidcontent
Tfeed
Tamb
RHamb

Feed rate
Feed temperature
Main flow
Main temperature
SFB flow
SFB temperature
Xout

Drying
chamber

Abs. humidity
Tout air

Rela!ve
humidity

Water ac!vity

Equilibrium
moisture
content

Figure 6.1: Dynamic Model of the spray drying chamber, illustrating the manipulated process variable inputs: Feed rate, inlet air flow, inlet air
temperature. Disturbance inputs: Solids contents in feed, ambient air
temperature, relative humidity of ambient air. Output : Toutair

76

Linearisation Analysis

A linearised model is a linear approximation of a nonlinear system. This

approximation is only valid within a specific region around a chosen operating point of the system. The operating point is a set of inputs u, outputs
y, states x, and disturbances v that describe the operating condition of the
system. Thus when the state values and inputs of the linearised system are
close to the operating point, the system will behave approximately linearly.
The system is linearised numerically by applying the linmod function in
Matlab on the Simulink model for a certain operating point. This function
computes the linearised state space model, which is given in (6.1) in terms
of x(t), u(t), v(t), and y(t). The -variables denotes the deviations from
the selected operating point. The matrices A, B, Bv 1 , C, and D are defined
as the Jacobians of the system evaluated at the operating point (6.2).
x (t) = Ax(t) + Bu(t) + Bv v(t)
y(t) = Cx(t) + Du(t)
f
|
, B=
A=
x x0 ,u0 ,v0
g
C=
|
,D =
x x0 ,u0 ,v0

6.1

f
f
|x0 ,u0 ,v0 , Bv =
|
u
v x0 ,u0 ,v0
g
|
u x0 ,u0 ,v0

(6.1)

(6.2)

Operating Point

The operating point is a set of inputs u, outputs y, states x, and disturbances v. The model has 5 inputs to the system, which are used to control
the drying process: Main inlet air, SFB inlet air, their respective temperatures and feed flow. The temperature of the outlet air is the output of
the model. Additionally 2 disturbance inputs are chosen for the model,
which are estimated to have the largest impact on the drying process and
the outputs: The relative humidity of the ambient air and Solids content in
feed. These parameters vary independent of the spray drying operation and
therefore seen as disturbances. The model has in total 8 states: 5 of the
states describe the process operating variable, which are modelled as first
order systems. 2 states describe the evaporation rate of the feed and the
total amount vapour in the chamber, respectively. The last state expresses
the outlet air temperature.
The operating point is selected to be a stationary state for the system, for
which the time derivative of the states are equal to zero. In this state the
system is stable and the Jacobians are constant matrices, hence the model
is time invariant. For various input and mean disturbance values stationary
1

Simulink regards disturbance v as in input

 to the system.Bv is therefore a part of the
u(t)
determined matrix B. 6.1 is then B
v(t)

6.2 Linearised results

77

states can be determined by following the iterative process given in section

4.2.3. The mass and energy balances do not have any constraints for possible
stationary states during the calculations of these, however there are physical
constraints which have to be considered:
Spray Dryer System: The performance limitations on the components
such as heater, fan, and pump. Also the ratio of amount of feed to air
flow and the ratio of Main air flow to SFB air flow is essential for the
drying process and for the system to work properly. Nonetheless the
ratio of air temperature to air flow rate is in particular of interest for
the drying process, since the amount of energy entering the system for
a certain air flow rate and temperature can be obtained, by decreasing
the flow rate and increase the temperature or vice versa. This will
change the properties of the drying air and the drying process.
Air properties: The humidity level of the drying air is vital for the
drying process. The amount of vapour that can be absorbed by the
drying air is limited by the saturation vapour pressure. Hence the
relative humidity cannot be larger than one in reality.
Feed properties: E.g. the amount of solids decides whether the feed is
capable of being atomised, which the model does not consider.
The operating point for the linearization is chosen to be the stationary state
for the default operation condition used for the experiment (B.1). The
performance of the dryer in this operating point is known to be stable and it
is observed that the spray dryer is capable of step up and step down around
this state with regards to air flow, feed flow and temperature, without any
degradation in its performance and stability useful. This state is therefore
estimated to be applicable as the operating point for the linearization. The
operating point for linearization is listed in table 6.1.
The stationary states and outputs are determined by running a simulation
with the chosen process condition. The disturbances input are the mean
values experienced for the process.

6.2

Linearised results

The state space matrices for the linearised model, for the operating point
specified above, is given in app. I. Also the transfer functions from each
input to the output are determined, for which the zero and poles are shown
and their respective frequency responses are illustrated in bode plots(app.
I).

78

Linearisation Analysis
Operating Point for Linearization
Variable Name
Description
Value
Input u0
u0 1
Main air flow IN
1800
u0 2
SFB air flow IN
500
u0 3
Temperature of MAIN
160
u0 4
Temperature of SFB
90
u0 5
Feed flow IN
65
Disturbance input v0 (mean)
v0 1
solids of total feed
50
v0 2
Rel. humidity ambient air
28
State x0
x0 1
Tout
85.44
x0 2
Main air flow IN
1800
x0 3
SFB air flow IN
500
x0 4
Temperature of MAIN
160
x0 5
Temperature of SFB
90
x0 6
Feed flow IN
65
x0 7
evaporation rate
0.0108
x0 8
Vapour in chamber
0.1933
output y0
y0 1
Tout
85.44

Unit
Kg/h
Kg/h
oC
oC
L/h
%
%
oC

Kg/h
Kg/h
oC
oC
L/h
Kg/s
Kg
oC

Table 6.1: Operating Point for linearization: Stationary state

It is observed that the transfer function for the inlet air flow and temperature
process inputs (u0 1 u0 4) and relative humidity disturbance input (v0 2) are
minimum phase systems2 , since all the poles and zeros are in the left half
plane (LHP) (app I.1.3). This means that there is a unique relation between
the gain and the phase for the frequency responses. As none of the transfer
functions have pure integrators the slope of the gain at low frequencies is 0.
The slope of the high frequency gain asymptote depends on the difference
dB
dB
and a pole -20 dec
.
in the number of zeros and poles. A zero adds 20 dec
to the slope. The transfer function for the process inputs (u0 1 u0 4
inlet air characteristics) has 2 poles and 0 zeros which results in a high
dB
. The transfer function for the feed flow input
frequency gain slope of -40 dec
u0 5 to output is a 4 order, 4 poles and 2 zeros, with one in RHP. The high
dB
. The disturbance input v1(solids) have 3
frequency gain slope is -40 dec
dB
poles and 2 zeros and therefore causes a gain slope of -20 dec
. The relative
humidity of the ambient air is a first order system which also descends with
dB
-20 dec
. (The bode plots are found in app. I.1.2).
2

minimum phase lag for the given magnitude respose

6.2 Linearised results

79

A LHP pole and a RHP zero adds a 90o phase shift to the high frequency
phase asymptote and a LHP zero and RHP pole adds 90o phase shift. As
seen in the appendix this results in a larger phase shift for the non minimum
phase transfer functions. The non minimum phase for feed flow and solids is
a result of how the model is build. A change in either the feed rate and solids
content(both non minimum phase) changes the amount of energy(liquid or
vapour conditions) entering system faster than the amount of energy leaving
the system. The amount of energy leaving the system is based on the total
amount of vapour in the system (5.7).
Bode Diagram minimum phase (u1) vs. non minimum phase (u5)
From: In(1)

From: In(5)

To: Out(1)

50

100

150
180
135
90
To: Out(1)

Magnitude (dB) ; Phase (deg)

45
0
45
90
135
180

10

10

10

10

10

10

Frequency (Hz)

Figure 6.2: Bode plot- example for a minimum phase transfer function(process
input main inlet air flow u0 1) (left)and non minimum phase transfer
(process input feed flow u0 5)(right). Both transfer function have 4
poles and 2 zeros. For the minimum phase transfer function the phase
shift= 90o (4 2) = -1800. For the non minimum phase transfer
function the total phase shift = 90o (4) + 90o = 360o

Thus as an example when the feed flow rate increases the amount of energy
entering the system increases, which theoretically means the temperature
rises until the vapour level in the chamber has increased and more energy is
leaving the chamber, which gives the fall in temperature.
To get a more accurate model only the energy of the evaporated amount of
vapour should be added to the system instead of the difference in input feed
and output powder. This is an extension to the modifications made on the
model in section 5.7, which has not been implemented yet.

80

Linearisation Analysis

The poles and zeros for the transfer function for the different input has been
examined. In the next it is examined how these look for the entire system.

6.3

Stability

A linear dynamic system is stable only if all its poles are in the LHP (Skogestad and Postlethwaite (2005)). The poles are determined by calculating the
eigenvalues for the system matrix A and inserted in figure 6.3(app. I.1.1).
The system can be concluded to be stable since all the poles are in the LHP.
The pole in -0.2 denotes the fastest dynamic in the system, which is for the
feed flow. The slowest dynamics in the system is the evaporation, which has
a pole in -0.0008 .
It is denoted that the system does not have any multivariable zeros.

-0.0676

-0.0008

0-0.0563

Poles
-0.0250 -0.0063

-0.0056

-0.0020

-0.2000

PoleZero Map for the Multivariable system :7 inputs 1 output

1
0.8

Imaginary Axis

0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
0.2

0.18

0.16

0.14

0.12

0.1

0.08

0.06

0.04

0.02

Real Axis

Figure 6.3: Zero-Pole plot for the linearized model. The linearized system contains only poles. There are 8 poles and describes the dynamics at each
state.There are no multivariable zeros, which mean there is no common zero for all the transfer functions. The columns in the transfer
function are not linearly dependent

6.4 Comparison of the linear model with the non-linear model

6.4

81

Comparison of the linear model with the nonlinear model

In this section the linear model is compared to the non linear model by applying various steps on the process - and disturbance inputs. The linearised
model describes how far the result is from the linearised operating point for
a certain change. The step change applied is also given as the deviation from
the inputs used for linearization. Thus at the operating point the output
is equal to zero and to compare it with the non-linear model the stationary
output value is added to the output.
The results of the comparison of the models for the various steps on both
process and disturbance inputs are given in app. I.2. Generally it is observed
that the linear model is a reasonably good estimate of the non-linear model
for the same step changes on the process inputs as used in the previous tests
of the non-linear model (table C.2).
The largest discrepancies between the linear and non linear model are observed when the step values are great compared to the stationary inputs.
As seen in figure 6.4 the step change of 10o C in the temperature of the inlet
air, for which the stationary value is 160 o C, gives a close estimate. While
a step change of 10 Lh for the feed flow results in a dissimilarity (figure 6.5).
Also changes in the disturbance input have resulted in good estimates. The
result of a step change in the solids content is seen in figure 6.6. However,
this is only a theoretical test as a change in solids content from 50% to 80
% percent is very unlikely in reality.
The linearised system has made it possible to analyse the system. This is
utilized in the design of a PI controller for the system in chapter 8.

82

Linearisation Analysis

outlet air temperature Linear model compared to Non linear model: main air temperature step T

out

main

86

Linear model: Tmain =10

Nonlinear model
85

Temperature [oC]

84

83

82

81

80

79
0.95

1.05

1.1

Time [sec]

x 10

Figure 6.4: Comparison of linear and Non-linear model: Main inlet air temperature. Step change= 10 from linearised input. No change difference is
observed

Tout outlet air temperature Linear model compared to Non linear Feed flowrate step Ffeed
86
Linear model: F

=10

Nonlinear model
Linear model: F

=2

feed

feed

85

Nonlinear model

Temperature [oC]

84

83

82

81

80

1.1

1.2

1.3

Time [sec]

1.4

1.5

1.6

1.7
4

x 10

Figure 6.5: Comparison of linear and Non-linear model: Feed step : 2 & 10. For
the small step no difference is observed. For the larger step a small
deviation is noted.

6.4 Comparison of the linear model with the non-linear model

out

83

outlet air temperature Linear model compared to Non linear model: Solids content step S

cont

Temperature [oC]

105
100
95

Linear model: S

90

cont

=0.3

Nonlinear model
85

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

Time [sec]

2
4

x 10

Temperature [ C]

Zoomed in
105
104.5
104

Linear model: Scont =0.3

103.5

Nonlinear model
1.3

1.4

1.5

1.6

1.7

1.8

1.9

Time [sec]

Figure 6.6: Comparison of linear and Non linear model: Solids content step from
50 % to 80 % Such a large change is not possible in reality. Small
difference between the non- linear and linear model.

x 10

84

Linearisation Analysis

Chapter

7
System Identification of
Residual Moisture Content

In the previous chapter a grey box model describing the temperature conditions of the drying air (environment) in the chamber using a combination
of first principle equations and observations, from the experiment on the
MSD-20 spray dryer, has been developed. The purpose was to estimate the
surrounding conditions the particles are experiencing during the drying process as an indirect measure for the final product quality. A model that is
able to estimate the operations condition in the dryer, and is able to estimate
the moisture content of the final product would have a distinct preference
when developing new control systems. However, as declared earlier due the
complexity of the product behaviour and drying parameters it is difficultto
express the moisture content of the product leaving the dryer by a white
box model.
In this chapter a black box model, as described in chapter 3, of the final
moisture content in outlet powder is modelled by applying system identification principles. This is another way of getting a linear model of the system
as it was completed for temperature model of the spray dryer in chapter 6.
The black box model makes it possible to create a mathematical model of
the dynamic process based on the measured data from the experiment on
the MSD-20, and then be able to estimate the changes in the moisture content when the process condition are changed. Nonetheless this subject is
only handled superficially in this project, to prove that black box modelling
is a proficient tool to model parts of a system which are problematical to
model, such as the moisture content of the product leaving the chamber.
Thus only linear models are evaluated and only parametric and state space

86

System Identification of Residual Moisture Content

identification methods are used.

7.1

Applied Identification Methods

Parametric identification is a matter of estimating numerically the values

of the parameters in a given model structure that gives the best agreement between the output of the model and the measured data. The system
identification toolbox in Matlab is able handle a variety of diverse linear
polynomial model structures, which can be given in the general form as in
(7.1).
The model structure shown is for discrete time, due to the fact that it is
straightforward to estimate the model in discrete time and then convert it
to continuous time. A, B, C, D, and F are polynomials that contains the
time shift operator q(1 sample). The coefficients in each polynomial are the
parameters that are to be found. ui is the ith input to the model and the
total number of inputs is nu. nki is the input delay for the ith input which
characterise the number of samples it takes for the output to respond to the
input. y(t) is the output and e(t) is white noise. (Ljung (2001), Mathworks
(2009))
A(q)y(t) =

nu
X
Bi (q)
i=1

Fi (q)

ui (t nki ) +

C(q)
e(t)
D(q)

(7.1)

Thus the output at the time t depends on the input and output values at
previous time instants. The order of the model is determined by the number of coefficients in the polynomials. Herafter when the model order is
mentioned it is thought to be the number of coefficients in the A polynomial. The model structures vary by how many of the polynomials that are
included in the structure or whether they are set to 1. Hereby they provide some flexibility for modelling the dynamics and noise characteristics.
In this project the main emphasis is on modelling the dynamic changes in
the moisture content when the operation conditions change.
Moreover due to the very few real samples ( 10 min) of the moisture
content in the powder, compared to the number of samples in the process
data from the spray dryer (1-10 sec), which is used as input data, it is
difficult to evaluate the direct influence of the noise at the inputs. Therefore
it has been decided that a detailed model of the noise is not necessary and
consequently the F and D polynomials are not included in the models that
are estimated here.
The model structures that will be examined is an ARX
1

Auto Regressive with eXogenous input

model, which is

7.2 Estimation Data and Validation Data

87

described by the A and B polynomial, and an ARMAX 2 model, which

provides some flexibility for modelling the disturbance dynamic with the
inclusion of the C polynomial. The parametric identification method minimises a performance function which is based on the sum of squared errors.
It can therefore experience many local minima in the performance function
and thus not converge to the global minima and problems with instability
can occur. For this reason state space identification method is examined,
which gives a model that is well conditioned. The state space model (7.2)
describes the same linear difference relationship between the input and output as the ARX model, but on the contrary only one delay is used. Instead
state variables are introduced corresponding to the model order and a Knoise
matrix that determines the noise properties
x(t + 1) = Ax(t) + Bu(t) + Knoise e(t)

(7.2)

y(t) = Cx(t) + Du(t) + e(t)

The focus of the estimated models is on simulation purposes (dynamic properties) rather than prediction purposes (stochastical properties).

7.2

Estimation Data and Validation Data

For estimation of a linear model, time domain data is used. The data is
collected from the experiment on the MSD-20 spray dryer (section 4.3). As
described in the test description, most of the tests on an operation variable
were repeated. In this way the first part of a test can be used as estimation
data, which is the data set that is used to train the model to fit given
data. The repeated test data is appropriate to be used as validation data to
validate the estimated model. Thus the model is simulated using the input
data from this data set and the result is compared to the output data in
this data set.
The moisture content is the parameter that is wished to be modelled based
on essential inputs (single output multiple inputs system). The data set,
for both estimation and validation data, contains data from 9 inputs and 1
output and is listed in table 7.1. The variables that are chosen to be inputs
are first of all the most significant parameters for the drying process. The
second criteria is that it should be able to model the input parameter values,
thus the moisture content can be estimated without the need for real data
once the model is up running. The input parameters used can therefore be
obtained from the dynamic model of the chamber developed in this project.
2

Auto Regressive Moving Average with eXternal input

88

System Identification of Residual Moisture Content

Sensor Name
MAINKGH
SFBKGH
T1702
T1704
F1626
P1706
T1624
P1618
T1709
T1616
P1614
INHUMABS
OUTHUMABS

Data Set
Description
Input Data
Main air flow into chamber
SFB air flow into chamber
Temperature of MAINKGH
Temperature of SFBKGH
Feed flow into chamber
Pressure difference of powder layer in SFB
Temperature of ambient temperature
Relative humidity of ambient air
Temperature of air flow out of chamber
Temperature of air leaving system
Relative humidity of air leaving system
Absolute humidity air IN
based on sensor values (T1624 & P1618)
Absolute humidity air OUT
based on sensor values (T1616 & P1614)
Output Data
Moisture content of Powder (SFB outlet)

Value

Unit

PV
PV
PV
PV
PV
PV
PV
PV
PV
PV
PV
PV

Kg/h
Kg/h
oC
oC
L/h
mmH2 O
oC
%
oC
oC
%
Kg/Kg

PV

Kg/h

%water /kg

Table 7.1: PV:process value, Out: Controller Output value

Accordingly five of the nine input parameter are the feed and inlet air parameters that have been extensively used during this project. Moreover the
outlet temperature of the chamber and the absolute humidity of the ambient
air at air intake are chosen, as these affects the evaporation. The absolute
humidity of the air leaving the chamber gives a description of the amount
of water evaporated. The last parameter is the pressure difference above
and below the powder in the SFB, which is an estimate of the height of
the powder layer in the SFB. This parameter is indirectly a measure of the
residence time of the powder in the chamber and included in the data set,
despite the fact that is not found in the model of the chamber. If the amount
of powder discharged the varied according to feed flow rate this parameter
can be modelled as a constant.
Two data sets are available. In the first one data is logged every second
and in the other one data is logged every tenth second. The noise in the
one second data set is more apparent than for the 10 sec. data set, which
due to the lower sampling frequency does not capture the high frequency
variations. For that reason the 10 sec data set is used in the identification
process. The moisture content of the powder at the outlet of the SFB is
only sampled every fifth (feed flow tests) or tenth minute (the rest of tests)
for offline measurement. These data are linearly interpolated to be able

7.3 System identification Results

89

estimate the moisture content with a 10 sec interval and combined with the
10 sec data from the system.(see figure C.2.3 in appendix C)
The estimation data set is the combination of all data from the tests shown
in the logbook in app. C.3. The validation data set is decided to be data
for the entire test, including the part which has been used to train the
model. This way the model is tested both with known and unknown data.
Furthermore the point of reference for the step change will be similar to the
one which has been used to train the model.This will make it easier for the
model to recognise the change and respond to it in a correct way.

7.3

System identification Results

The best model between the selected model structures according to the given
data is wished to be determined. However, the quality of the model can be
defined and measured by various parameters (Mathworks (2009)):
Loss function - value of the identification criterion at the estimate,
thus it indicates how well the model is fitted to the estimation data.
Best fit - sum of the squared error between the validation data output
and the model output
Final Prediction Error (FPE)- Akaikes critierion is another way to
describe the difference between the model and validation data.
Model Order
All the parameters are aimed to be as low as possible. But concerning the
loss function, a value too low can also indicate over fitting of the model to the
estimated data, which can have a negative effect when the model is used on
new data. Also an increased model order in normal cases gives a better fit,
but this increases the possibility for zero pole cancellations and over fitting
to the noise. Hence these parameters can be used as guiding indicator about
the quality of the estimated model, nonetheless it is essential that the model
has captured the dynamic and therefore a graphical view of the simulated
output and measured out will be used as well. Since there is no previous
knowledge about the dynamics of the final moisture content, the trial-anderror approach is used to determine the model order and delays necessary
to get the best model.

90

System Identification of Residual Moisture Content

7.3.1

ARX model

The ARX model is the simplest polynomial model and fast to be calculated.
A function in the System Identification Toolbox in Matlab allows one to
model a range of orders and delays simultaneously and compare the resulting
models. This was used to estimate model orders from 1 to 10 and likewise for
the delay. The result showed that a model of the order 10 with one coefficient
for each input (B polynomial coefficient) and a time delay of 8, gives the
best fit(F P E = 0.00105) (see appendix H.1.1). The model is simulated for
the validation data set and compared to measured output in figure 7.1. For
comparison a similar ARX model of order 15 is simulated and included in
the same figure.
Measured and simulated ARX model output of the moisture content

Residual moisture Content [% water/kgpowder]

2.5

1.5

Model order 10 & 1 input coeffcients

Measured
Model order15 & 1 input coeffcients
0.5

Time [sec]

10
4

x 10

Figure 7.1: Simulated ARX model output and measured output(moisture content
(%water/kgpowder ) for a 10th order model and a 15th order model.
Both describe the input with one coefficient and time delays for these
are 8 sec.

It is noticed that the simulated model has captured some of the trends but
the result is far from acceptable. This is mainly due to the fact that the
system dynamic and the stochastic dynamics are coupled by having the same
set of poles. Thus the model is deficient when it comes to distinguishing noise
from the system dynamics. Increasing the model order could give a better
fit. The zero-pole plot for this model illustrates a zero-pole cancellation and
increasing the model order is not desired, which increases the possibility of

7.3 System identification Results

91

more zero-pole cancellations (app. H.1.1).

7.3.2

ARMAX Model

The ARX model has given a good starting point for estimating ARMAX
models, which unlike the ARX model describes the stochastic dynamics with
additional polynomial C (7.1). Starting with a 10th order ARMAX model,
the estimated model was successfully decreased to a 6th order ARMAX
model. It has 7 input coefficients (nb) and 8 time delays (nk). At the same
time it gives a better fit according to quality parameters. The simulated
outputs and the coefficients for both models is illustrated in figure 7.2. A
Measured and simulated ARMAX model output

Residual moisture Content [% water/kgpowder]

2.5

1.5

0.5

Feed Flow
1626

SFBKGH &
1704

MAINKGH

1702

0.5

END of
TEST

na=6 nb=7 nc=2 nk=8

na=10 nb=5 nc=8 nk=8
Measured data
0

Time [sec]

10
4

x 10

Figure 7.2: ARMAX model of order 6 and 10 with the coefficients given in the
na, nb, nc, nk). 6th order: F P E = 0.00017, Loss = 0.000169. 10th
order: F P E = 0.0010, Loss = 0.00109. The tests periods are shown
as well. In the last part of the main inlet air flow test (MAINKGH)
the PI controller on the air temperature was turned off.

few of the estimated ARMAX models and their quality parameter is listed
in table H.1 the simulated outputs are given in app. H.1.1.
The first part of the tests is used for estimation of the model. As it is
seen the models fits well to this part. The second part is the new data for

92

System Identification of Residual Moisture Content

validation, which at first sight may not look excellent, but acceptable and
the deviations may be explained.
In the feed flow test it observed that the moisture contents decreases in
spite of the fact that the same step changes are applied and the outlet air
temperature is almost constant (app. C.2.4). The explanation could not be
found in any of parameters used for the identification of the system. But
this may be due to the changes in the amount of fines collected and the
changes in the agglomeration process when the feed flow is changed. These
are unmeasured parameters. it notes that the model has estimated the step
change similar to the behaviour of the training data.
The deviation between the measured and simulated value in the temperature
test (sensor 1702. see app. C) for main air inlet is due to the fact that
different steps are applied in the estimation data set and the validation data
set. In the estimation data the steps applied are from 1600 C 1500 C
and 1500 C 1700 C. In the validation data set the step used is from
1700 C 1600 C. (C.3). This illustrates the fact that a black box model is
only useful in the operating region which data is extracted from and used in
the identification process (see modelling section 3).
The effect of changing the main inlet air on the final moisture content is not
distinct. Moreover the validation data set for this test cannot be used to
evaluate the model, because in this part the controller to 1702 was turned
off. This means the effect from the electric heater is constant. Consequently
an increase in air flow decreases the inlet temperature. The model therefore
misinterprets the situation and estimates the fall in temperature to be equal
to a rise in the moisture content. The outcome of changing the SFB inlet air
flow is indistinguishable as for the Main inlet air. So the model estimates
noise more than the actual dynamic. But the temperature of the SFB is
observed to have a great impact on the moisture content and model estimates
it well.
It is observed that the 6th and the 10th order model behave identically.
The main difference is how the noise is described. The 10th order model
includes more noise, while for the 6th order model only 2 coefficients are
used to describe the noise. This seems satisfactory in a situation where the
system dynamics is of main interest. The zero-pole plot of the 6th order
model is illustrated in figure H.8 in appendix. The model is stable since the
poles of the system lies within the unit circle and no zero-pole cancellation
is observed. In the same appendix results from other ARMAX models are
shown and what effect the number of polynomial coefficients has on the
estimated model.

7.4 Summary: System identification

7.3.3

93

State Space model

For the state space model structure the best model is estimated to be of 4th
order. The results from simulating this model is shown in figure 7.3. For
comparison the ARMAX model is also given in the figure. It is observed
that the state space model experience the same problem as the ARMAX
model. Both models are able to capture the most dominant dynamics.

Residual moisture Content [% water/kgpowder]

Measured and simulated model output

1.5

0.5

ARMAX model 6th order

State Space model 4th order
measured data
0

Time

9
4

x 10

Figure 7.3: State space model of 4th is identified as being the best for this model
structureF P E = 0.00030, Loss = 0.0003.. Here the model is compared to the ARMAX model of 6th order and the measured data.

7.4

Summary: System identification

ARX, ARMAX, and state space model structures has been examined. The
ARMAX model and state space model gave the best simulation results, but
since the estimation data and validation data are not based on the same step
changes for all tests, it is difficult to estimate the quality of the estimated
model. Nonetheless this confirmed that a black box model only is acceptable
within the operation region that has been used to train the model.
Another problem appears to be, that the test periods are too short with only
a few measurements. Longer test periods would give time for the moisture
content to settle around a point, which might help the model to distinguish

94

System Identification of Residual Moisture Content

between a noise and the system dynamics. For a short test period with only
a few measurements the observed changes may be assumed as being a part
of the noise dynamics than a part of the system dynamics.
In the models investigated in this section the number of coefficients (nb)
for the input has been chosen to be the same for all of them. This is not
necessarily the best, since each input can go through different orders of
dynamic in reality. Further studies of the inputs are necessary to optimise
the model. This same applies for the time delay of the input coefficient.
The system identification is only superficially handled in this project. This
also means that the model order has been reduced by the trial-and -error
method. Model reduction can also be done analytically for example by using
the Hankel norm approximation.
A black box model is a fast and straightforward way to model a system.
It needs a great number of data for different operation regions both for
estimation and validation. Otherwise the developed model will be too simple
and ineffective for modelling. The disadvantage of this type model is the
lack of flexibility as it cannot be used for other systems. The conclusion
is that a complete black box model is also difficult to estimate without the
appropriate data. However, it might be possible to estimate some simpler
parameters, that can be combined with the white box model. For example
the time it takes for the system to settle after a step change, which gave
difficulties for the white box model in chapter 5.4.

Chapter

8
Control of spray dryers

In this chapter it is demonstrated that the developed model of the spray

dryer is relevant for examining control strategies. This is realised by implementing the main feed rate control strategy with a PI controller which is
used on most of the spray dryers at present time. The input, output, and
disturbance variables associated with the spray dryer for process control is
shown in figure 8.1. Though, only control on a single operation variable is
described in this section.

8.1

Control Strategy

As described in the introduction in chapter 1, the control system of a spray

dryer has many objectives:
Maintenance of a desired dried product quality, irrespective of the
disturbances in the drying operation.
Stable drying process
Optimisation of the performance of the drying process- maximisation
of throughput at optimal energy efficiency and at minimum cost
safety system- in case of system failure.
The target of the inspected controller in this project is only to keep the
process stable and disturbance rejection. Thus the dried product quality,

96

Control of spray dryers

Solidcontent
Tfeed
Tamb
RHamb

Feed rate
Main flow
Main temperature
SFB flow
SFB temperature

Drying
chamber

Tout air

Figure 8.1: Illustration of the process input and outputs to the spray drying system and disturbances which can be used in the controller design process

measured as the moisture content in the final powder, is maintained and

uniform. Since the moisture content is not directly measurable, it is indirectly controlled by maintaining the outlet air temperature at a set value by
varying either the feed rate to the dryer or the main inlet air drying temperature (feedback control strategy). Maintaining a constant outlet drying
air temperature through feed rate control is mostly used, in which the feed
pump/nozzle pressure is varied to counteract any temperature deviations.
The time constant for a change in the manipulated variables is smallest
for the feed rate, as observed from the test results from the MSD-20(app.
C.2.4), which is the reason to use this as the main control loop. While the
main control loop is used to control the drying process the other input variables are kept constant by the relation between air flow and the heated air.
However for some systems it is required to keep a constant nozzle pressure,
due to the product characteristics of the feed. In such cases the outlet air
temperature is controlled by the inlet air temperature.

8.1 Control Strategy

97

Disturbance:
Solides content

Tamb

Feedstock

PUMP

PI

RHamb
Drying air

HEATER

Variable
speed drive
Tin air

DRYING
CHAMBER

Tout air

Powder
PI

Figure 8.2: Illustration of feed rate control by using a PI controller. Inlet air
temperature is kept constant by a PI controller, which measures the
temperature of the inlet air and controls the heater.

8.1.1

PI controller

The control strategy used in the spray dryer is feedback control with a PI
controller, in which an error signal e(t) is used to generate the proportional
and integral actions. The resulting signals are weighted and summed to
shape the control signal u(t) to the plant model. The algorithm for the PI
control is given in (8.1). Kp is the proportional gain and i is the integral/reset time which weights the influence of the integral term. When the feed
rate is controlled the error signal is defined as e(t) = rtemp Toutair . The
laplace transformed transfer function of the PI controller is given in (8.2).


Z
1 t
e( )d
u(t) = Kp e(t) +
i 0


i s + 1
U (s)
= Kp
Gc (s) =
E(s)
i s

(8.1)
(8.2)

The controller has a pole in origo and a zero in s = 1i , which results in

the controller to have its cutoff frequency at b = 1i .

98

Control of spray dryers

8.2

PI controller for disturbance rejection of solids

content variation

A PI controller for feed rate control is determined, for which the objectives
are given as:
Reference tracking: The controller should be able to hold the outlet air
temperature value at a set point. For a certain change in the reference
the controller should manipulate the feed rate to obtain the desired
effect.
Disturbance rejection: The disturbances shown in figure 8.1 should be
rejected. The disturbance from the ambient air characteristics have a
slow dynamics, as it takes time for the temperature and humidity level
to change. However the disturbances from the feed is somewhat faster,
as this depends on the preheater/evaporator. The solids content has
a major impact on the drying process and the final moisture content,
which is the reason for examining this disturbance.
The general requirements for the control system:
Stability and Accuracy: In order to obtain powder with a constant
level of moisture content, no oscillation in the system is desired, as
this result in fluctuations in the output moisture content. A constant
outlet air temperature is wanted.
Response speed: The response speed is of less importance, since the
spray dryer in general is a slow system, due to the long evaporation
time of the feed and the time for the evaporation rate to settle as
noticed in chapter 5.7. Hence it is not necessary for the controller to
be as fast for some servo system.

8.2.1

PI controller design

The temperature control system by feed rate variation is shown in figure

8.3. The transfer function Gf eed (s) from the feed input to the outlet air
temperature output was found in chapter 6 and given in (8.3). The transfer
function from disturbance input to the output, Gsolid (s), is given in (8.4).
0.00042s2 0.0002213s 3.736 107
s4 + 0.3247s3 + 0.02885s2 + 0.0007841s + 6.095 107
0.4735s2 + 0.09684s + 0.0001852
Gsolid (s) = 3
s + 0.1247s2 + 0.003905s + 3.047 106
Gf eed (s) =

(8.3)
(8.4)

8.2 PI controller for disturbance rejection of solids content variation 99

Gsolids

+
r

CONTROLLER
Gc(s)

+
u

Gfeeds

Tout air

Figure 8.3: The temperature control system by feed rate variation. Gf eed (s) is the
transfer from feed input to output. Gc - controller transfer function.
Gsolid (s) is the transfer function of the disturbance.

It is noticed that the static gain for the disturbance is Gsolid (0) 60, which
means that the response to a unit disturbance will be 60 times larger than
acceptable. The Bode plot of this transfer function shows that this decreases
for larger frequencies, but the gain remains larger than 1 up to d = 0.5 rad
sec
at which frequency Gsolid (jd ) = 1 (see app. I.3).
In order to be able to reject the disturbances the cross over frequency, c , for
the controller is chosen to be equal to d . It is known that the PI controller
includes a phase shift of 900 at low frequencies and 00 at higher frequencies.
To reduce the impact of the introduced negative phase shift on the closed
loop systems phase margin, the cutoff frequency b of the controller is chosen
to be four times lower than the determined cross over frequency c (8.5).
Hence the phase shift of the controller will be close to zero at this frequency
(Jannerup and Srensen (2004)).

1
1
=
i
c
1
= 8 sec
i =
0.25 0.5 rad
sec

k =

(8.5)
(8.6)

Inserting the integral time in (8.2) and the open loop transfer function can

100

Control of spray dryers

be written as:

Gopen (s) = Gc (s)Gf eed (s)



8s + 1
0.00042s2 0.0002213s 3.736 107
Gopen (s) == Kp
8s
s4 + 0.3247s3 + 0.02885s2 + 0.0007841s + 6.095 107
(8.7)

The bodeplot for the open loop system Gopen for proportional gains Kp =
1 10 is given in figure 8.4. The phase margin, which is a description of how
much phase lag can be added to the system before the phase at the gain
crossover frequency becomes 1800 , is for Kp = 1 is 1400 . Phase margin
for Kp = 10 is 700 . Normally a phase margin larger than 300 is required,
which is seen to be satisfied for the examined proportional gains.
Furthermore the system fulfils Bodes stability condition for closed loop system, since |Gopen (j180 )| < 1. (Skogestad and Postlethwaite (2005))

Bode Diagram Open loop transfer function with PI control, (Kp =110)
150

Magnitude (dB)

100
50

System: G_open
Frequency (rad/sec): 0.0238
Magnitude (dB): 0.745

K =10
p

0
System: G_open
Frequency (rad/sec): 0.103
Magnitude (dB): 0.203

50

K =1
p

100
150
90

Phase (deg)

45

System: G_open
Frequency (rad/sec): 0.0238
Phase (deg): 49.6

for K = 1

45

c for Kp =10

System: G_open
Frequency (rad/sec): 0.103
Phase (deg): 20.8

90
135
180
5
10

10

10

10

10

10

10

Frequency (rad/sec)

Figure 8.4: Bode plot for the open loop transfer function for various proportional
gains Kp . It seen the frequency response is close to each other for the
different gains. Phase margin for Kp = 1 is 1400 . Phase margin for
Kp = 10 is 700 .

10

8.3 Results from PI controller implementation in dynamic model

8.3

101

Results from PI controller implementation in

dynamic model

In this section a PI controller with the previous determined integral time

has been implemented in the dynamic model. Step tests on the reference
temperature and in solids contents has been conducted for the lowest and
the highest proportional gain (Kp = 1 and Kp = 10) has been conducted.

8.3.1

Step on reference temperature

In this test the reference temperature is stepped up from steady state temperature 84.440 C 870 C. The resulting control signal u, which is the
output of the PI controller, and the response of the outlet air temperature
is shown in figure 8.5.
It is seen that the rise time is shorter for Kp = 10 is approximately 50 sec,
while it takes 100 sec for Kp = 1. It is noticed that the overshoot of the
outlet air temperature is lower than for the smaller gain, nonetheless more
oscillations are observed and it requires more of the feed pump due to the
large variations in the control signal. This is not desired, since changes in
feed rate also affects the structure of the particle and the agglomeration
process, which can result in a non uniform product. The settle time is
equal for both gains. The outlet air temperature is within 0.10 C of the final
temperature after 200 sec. Temperature is completely settled after 1500 sec.
This is due to the change in evaporation rate, with a very slow pole (see
section 5.7). It has to be emphasized that an increase in the control signal
is a decrease in feed rate, since Ff eed u is the input to the system.

8.3.2

Step on Solids content

The solids content can vary 2% from the mean solids content in the feed.
(Westegaard (2004)). In this project, the mean solids content has been
calculated to 50%. At first a step change is applied on the solids content
value. This is only of theoretical interest as this does not occur in reality.
The step applied is from the mean value of 50% to 52% solids in feed. The
target is to keep the default outlet air temperature at 85.440 C. The result
of the temperature control and the disturbance rejection is shown in figure
8.6.
Again it is observed that the controller with the larger proportional gain
has the fastest response. Due to the slower response from the controller
with Kp = 1 the outlet air temperature increases more than for the other
controller. This controller is also noted to have larger oscillations, but the

102

Control of spray dryers

control signal u

Step on reference temperature r= 85.44 >87 0C

PI Control signal K =1

15

PI Control signal K =10

p

10
5
0

7.05

7.1

7.15

7.2

7.25

Time [sec]
Temperature T

x 10

out air

Temperature [ C]

87.5

Tout air Kp =1

87

Tout air Kp =10

86.5
86
85.5
85

7.02

7.04

7.06

7.08

7.1

7.12

7.14

7.16

7.18

Time [sec]

Figure 8.5: PI controller: reference temperature is stepped up from steady state

temperature 84.440C 870 C. Controller is examined for Kp = 1
and Kp = 10. Top figure shows the control signal. The figure below
illustrates the outlet air temperature response. The settle time is
equal for both gains. The rise time is shorter for Kp = 10 is ca. 50
sec, while it takes 100 sec for Kp = 1. Outlet air temperature is
within 0.10 C of the final temperature after 200 sec.

settle time is similar for both controllers: 400 sec.

Continuous Disturbance in solids content The effect of the controller
is examined by simulating a continuous disturbance in solids content. This
is completed by using a uniform random block in Simulink, which generates uniformly distributed random numbers over a specifiable interval. The
interval in this test is specified to be from 48% to 52%. In figure 8.7 the temperature response of a process with a continuous disturbance in which the
PI controller is turned off and then switched on is illustrated. For the process running without the PI controller the temperature variation is 0.70 C.
When the PI controller is turned on the temperature variation decreases to
less than 0.30 C.

8.4

Possible control strategies

In this chapter a PI controller has been designed, implemented in the dynamic model of the spray dryer and tested for reference tracking and disturbance rejection to demonstrate that the model can be used to evaluate the
performance of a controller for a spray drying process.

7.2
4

x 10

8.4 Possible control strategies

103

Step on solids content S

= 50% >52%

control signal u

conc

control signal u Kp =1

0.5

control signal u Kp =10

1
1.5
2
2.5
3
7

7.005

7.01

7.015

7.02

7.025

7.03

7.035

7.04

7.045

7.05

Time [sec]

x 10

Temperature T

out air

Temperature [ C]

86
85.9

Tout air Kp =1

85.8

out air

K =10
p

85.7
85.6
85.5
85.4
85.3
85.2

7.005

7.01

7.015

7.02

7.025

7.03

7.035

7.04

7.045

Time [sec]

Figure 8.6: PI controller:step on solids content is stepped up from the mean value
of 50% to 52% solids in feed . Top figure shows the control signal. The
figure below illustrates the outlet air temperature response. Kp = 1
gives a slower response and therefore larger temperature oscillations.
settle time is similar for both controllers: 400 sec

The PI controller is a good basic controller with no steady state error, but
to optimise the performance of the spray drying process it is expected more
variables has to be controlled. During the development of the dynamic
model, it was experienced that the humidity level in the dryer is an essential
parameter for the drying process and the final moisture content in the product. Applying the air humidity as a control variable in the control system
gives some possibilities.
At present time single input single output control is used. But multivariable
control is estimated to have a great potential for spray drying systems. It
has been observed that the air inlet temperatures has relatively fast response
time and has a great effect on the moisture content. Moreover it is known
that the relative humidity is dependent on the temperature and the difference in vapour pressure between the feed and air determines the drying rate.
Manipulating the inlet air temperature will therefore give a better control
of the drying condition in the chamber.
Especially the SFB inlet air temperature was observed, during the system
identification process, to have a great effect on the moisture content. In a
multistage dryer the SFB part gives, during the fluidisation, the powder its
specific characteristics. Using temperature from this part of the spray dryer

7.05
4

x 10

104

Control of spray dryers

Solids content [kgsolids/kgfeed]

Continuous Disturbance in solids content S

= 48% 52%

conc

0.53

Solids content
0.52
0.51
0.5
0.49
0.48
0.47
1.9

1.92

1.94

1.96

1.98

2.02

2.04

2.06

2.08

Time [sec]

2.1
4

x 10

Temperature T

out air

85.55

Temperature [oC]

Tout air Kp =10

PI controller ON

PI controller OFF
85.5

85.45

85.4

85.35
1.9

1.95

2.05

2.1

2.15

2.2

2.25

Time [sec]

Figure 8.7: PI controller:Continuous Solids content Disturbance. The figure

above is the simulated variation solids content. The figure below
shows the temperature response. Before the time reaches 20000 sec.
the PI controller is turned off and the outlet air temperature varies
0.70 C. After 20000 sec PI controller is turned on. The temperature
variation decreases to less than 0.30 C.

to control the SFB air temperature is assumed to result in a more uniform

powder quality.
The linearised model provides the possibility for LQG control and the Kalman
filter. Otherwise more advanced H2 optimal control can be used, in which
the output error of the system is minimized. Another possibility is to
shape the sensitivity function, as this is a good indicator for closed loop
performance(H control). Further description of multivariable control can
be found in (Skogestad and Postlethwaite (2005))

2.3
4

x 10

Chapter

9
Conclusion

The objective of the spray drying process is to produce dried product of a

desired quality regardless of the disturbances in the drying operation and
variations in feed supply. In order to improve the dryer operation and its efficiency, alternative control strategies than the classical PI control are wished
to be examined. However, the design of a controller has not been the focus
of the project. The aim of this project was to construct a dynamic model of
the drying process in a multistage spray dryer with a mixed air flow, that
can be utilised in the future study of new control strategies for the spray
dryer.
In this project a dynamic model describing the drying environment, with
regards to the air temperature and humidity, inside the chamber has been
developed based on white box modelling methods. The underlying basis for
this dynamic model is mass and energy balance equations. It is assumed
that the air and particles in the drying chamber are well mixed, due to the
mixed air flow, and the spray dryer is therefore modelled as a continuously
stirred reactor tank. The outlet air temperature is for that reason equal
to temperature in the chamber. The calculated steady state temperatures,
when heat loss through the chamber is included, is very close to the temperatures measured from the experiment on the Multi Stage Dryer-20 at GEA
Niros test station(deviation 1 2o C).
The absolute humidity of the air in the chamber is determined from the
amount of vapour evaporated from the feed. In the model it is assumed
that the final product leaving the spray dryer has reached its equilibrium
moisture content. This is a simplification, since some of the powder leaves

106

Conclusion

the dryer before this state is reached in reality, which gives a lower humidity
level than the model estimate.
The dynamic model has been implemented in MATLAB/Simulink, in which
the input process variables, size of the spray drying chamber, disturbance
and product characteristics can be varied and the process simulated. The
dynamic model has been validated by comparing the simulation results with
the experimental results from MSD-20.
The temperature response for step changes in the input process variables
dryer has been examined. For a step change in main inlet air temperature
the model behaved similar to what was observed for the real spray dryer.
The time for the outlet air temperature to settle for a step in main inlet temperature is 1500 sec. Modelling the temperature response for a step change
in air flow appears to be more complicated due to the complex behaviour
of the air flow around the inside of the chamber. The model presents a
response which is twice as fast the MSD-20 test results. For a change in
feed flow rate into the chamber, the temperature response is sensitive to
the changes in evaporation rates. In this project the change in evaporation
rate is modelled as a first order system by inspecting the drying times for
a single particle and the experimental results. The outcome of using this
method has shown good results for the temperature response when the feed
rate changes. The time for the temperature to settle for step in feed rate is
approximately 2500 sec.
The model has been linearised in order to analyse the model. It is seen
that the behaviour of the linear model is close to the non-linear model for
small steps. Moreover the frequency response was examined and used in the
design of a simple PI controller.
An estimate of a black box model, which is another way of examining a linear model, of the moisture content as a function of input process variables,
ambient air disturbances and outlet air temperature, has been attempted
by using the system identification toolbox in Matlab. However lack of estimation and validation data for moisture content assessment resulted in less
optimal models. However it is expected that these methods can be used to
determine some parameters for the white box mode, which will result in a
gray box model.
In the end it was demonstrated that the model can be used with a PI controller. The PI controller used here was for reference tracking of the outlet
air temperature and rejection of disturbances from the variation in solids
content, which it accomplished successfully.

Nomenclature

mtransf
er mass transfer rate, page 50

(Kg/(s m2 ))

latent heat of vaporization, page 25

air

absolute viscosity, page 52

kinematic viscosity, page 52

break frequency, page 99

Cross over frequency, page 99

Cross over frequency for disturbance-for rejection, page 99

relative humidty of vapour gas mixture, page 27

integral time for PI control, page 97

evap

time constant for the slow dynamics in evaporation rate, page 71

f ast

time constant fast response in evaporation , page 71

ACeiling angle of the chamber ceiling, page 45

AChamber surface area, page 34

Kj/Kg

%
-

m2

ACone angle of cone on chamber, page 45

CdryAir heat capacity of dry air, page 23

Kj/Kg K

CdryAir specific heat capacity of dryAir, page 25

Ceq

parameter for Xeq calculations, page 26

Csolid specific heat capacity of dry solid, page 23

Kj/(Kg K)
Kj/(Kg K)

108

Conclusion

Cvapour heat capacity for water vapour, page 23

Kj/Kg K

Cwater specific heat capacity of water, page 23

Kj/(Kg K)

CST R Continuously Stirred Reactor Tank, page 20

Dair

diffusion coefficient of water vapour in air, page 51

Dchamber diameter of the chamber, page 45

ddrop

diameter of a single droplet, page 50

Def f

effective diffusivity, page 50

e(t)

error signal to controller, page 97

m2 /s
-

F 1626 Sensor:Feed flow into chamber, page 94

L/h

FF eed feed flow rate, page 21

L/s

FM aindry Main dry air flow into chamber, page 21

Kg/s

FOutdry outlet dry air flow , page 21

Kg/s

Fpowder powder flow out of system, page 21

Kg/s

FSF Bdry SFB dry air flow into chamber, page 21

Kg/s

Gc

Controller transfer function, page 100

Gf eed Transfer function- feed input to outlet temperature, page 99

Gsolid Transfer function- solid input to outlet temperature, page 99

Hair

enthalpy of air, page 21

Hchamber height of the chamber, page 45

J/Kg
m

HdryAir enthalpy of dry air, page 23

J/Kg

Hf eed enthalpy of feed, page 21

J/Kg

heat transfer coefficient, page 48

W/(m2 K)

HhumAir enthalpy of humid air, page 23

J/Kg

Hpowder enthalpy of powder, page 21

J/Kg

hheat

Kp

proportional gain, page 97

Keq

parameter for Xeq calculations, page 26

Kmass mass transfer coefficient, page 48

m/s

109
mv

mass of vapour, page 28

Mw

molar mass of water, page 28

mw

mass of water, page 28

Kg
g/mol
Pa

MdryAir molar mass of dry air, page 28

g/mol

mdryAir mass of dry air, page 28

Ms

Kg

dry solid flow, page 24

Kg/s

M AIN KGH Main air flow into chamber, page 94

Nu

Kg/h

Nusselt number, page 51

P 1614 Sensor:Relative humidity of air leaving system, page 94

P 1618 Sensor:Relative humidity of ambient air, page 94

P 1706 Sensor:Pressure difference of powder layer in SFB, page 94 mmH2 O

Pv

partial vapour pressure, page 27

Pa

Patm

standard atmospheric pressure, page 28

Pa

PdryAir partiel pressure of dry air, page 28

Pa

Psat

Pa

Saturated vapour pressure, page 27

Psurf ace pressure at surface, page 48

Pa

Ptotal Total Pressure, page 28

Pa

Pr

Prandtl number, page 52

Qheat heat transfer rate, page 48

Qloss heat loss through chamber, page 35
R

Universal gas constant, page 28

Re

Reynolds number, page 52

J/(s m2 )
KW
8.314J/(mol K)
-

rhodry density for dry air, page 44

Kg/m3

rhof eed density of feed, page 33

Kg/m3

rhototalair density for air, page 44

Kg/m3

rhovapour density for vapour, page 44

Kg/m3

Solids in, page 33

110

Conclusion

Sconc amount of solid in feed, page 33

Sc

Kgsolids /Kgf eed

Schmidt number, page 52

SF BKGH SFB air flow into chamber, page 94

Sh

Sherwood number, page 51

Temperature, page 28

Kg/h
oC

T 1616 Sensor:Temperature of air leaving system, page 94

oC

T 1624 Sensor:Temperature of ambient temperature, page 94

oC

T 1702 Sensor:Temperature of MAINKGH, page 94

oC

T 1704 Sensor:Temperature of SFBKGH, page 94

oC

T 1709 Sensor:Temperature of air flow out of chamber, page 94

oC

Tair

oC

air temperature, page 23

TM ain temperature of Main air flow, page 24

oC

TOutair temperature of outlet air, page 25

oC

Tpowder temperature of powder, page 25

oC

Tref

reference temperature, page 23

oC

TSF B temperature of SFB air flow, page 24

oC

KJ/m2 /h

UChamber heat transfer coefficient, page 34

m3

volumen, page 28

Weq

parameter for Xeq calculations, page 26

X0

initial moisture content, page 24

Xeq

equilibrium moisture content, page 26

Kgw ater/Kgs olid

XIn/out moisture content of particle, page 23

Kgwater /Kgsolids

XIn

moisture contents of feed, page 24

Kgwater /Kgsolids

Xout

final product, page 24

Kgwater /Kgsolids



Kgwater /Kgsolids

Xparticle average mositure content of the particle, page 51

kgmoist
kgdrySolid

YIn

absolute humidty inlet air, page 24

Kgwater /KgdryAir

Yout

absolute humidity outlet air, page 24

Kgwater /KgdryAir

Y Hvapour moisture content in air, page 23

Kgvapour /KgdryAir

Bibliography

Tom Benson. Reynolds number, 2009. URL http://www.grc.nasa.gov/

WWW/BGH/reynolds.html.
Engineering Tool Box. Air properties, 10 2009. URL http://www.
engineeringtoolbox.com/air-properties-d_156.html.
Chung (Peter) Chieh. The liquid state, 11 2009. URL http://www.science.
uwaterloo.ca/~cchieh/cact/c123/liquid.html.
W.H. Gauvin and S. Katta. Basic concepts of spray dryer design. American
Institute of Chemical Engineers Journal, 22,4:713723, 1976.
M. Piatkowski I. Zbiciski and W. Prajs. Determination of spray-drying
kinetics in a small scale. Drying Technology, 23,8:17511759, 2005.
Czeslaw Strumillo Ireneusz Zbicinski and Agnieszka Delag. Drying kinetics
and particle residense time in spray drying. Drying Technology, 90,4:
17511768, 2002.
Ole Jannerup and Paul Haase Srensen. Reguleringsteknik. Polyteknisk,
third edition, 2004.
Jorge C. Olivera Jes
us M. Fras and Klaus Schittkowski. Modeling and
parameter identification of a maltodextrin de 12 drying process in a convection oven. Applied Mathematical Modelling, 25:449462, 2001.
C.I. Beristain E.J. Vernon-Carter J.G. Baez-Gonz
alez, C. Perez-Alonso and
M.G. Vizcarra-Mendoza. Effective moisture diffusivity in biopolymer
drops by regular regime theory. Food Hydrocolloids, 18:325333, 2004.
H.C Dich C.M.Le J. Mortensen K.H. Clement, A. Hallstrom and H.A Thomsen. On the dynamic behaviour of spray dryers. Trans IChemE, 69, Part
A, 1991.

112

BIBLIOGRAPHY

Frank Geert Kieviet. Modelling Quality in Spray Drying. Ph.d, Eindhoven

University of Technology, Chemical Engineering, January 1997.
Labspace. Modelling: Choosing a model, 9 2009. URL http://labspace.
open.ac.uk/mod/resource/view.php?id=361363&direct=1.
T.A.G Langrish. Multi-scale mathematical modelling of spray dryers. Journal of Food Engineering, 93:218228, 2008.
T.A.G Langrish and T.K Kockel. The assessment of a characteristic drying
curve for milk powder for use in computational fluid dynamics modelling.
Chemical Engineering Journal, 84:6974, 2001.
Octave Levenspiel. Chemical Reaction Engineering. John Wiley and Sons,
third edition, 1999.
Kurichi Kumar Lixin Huang and Arun.S. Mujumdar. Computationalfluid
dynamic simulation og droplet drying in a spray dryer. Drying 2004 Proceedings of the 14th International Drying Symposium IDS 2004, A:326
332, 2005.
Lennart Ljung. System identification Toolbox users guide. The Mathworks,
5th edition, 2001.
Keith Masters. Spray drying in Practice. SprayDryConsult, 2002.
Mathworks. System identification toolbox, 11 2009. URL http://
www.mathworks.com/access/helpdesk/help/toolbox/ident/ident_
product_page.html.
Arun S. Mujumdar. Handbook of Industrial Drying. CRC Press, third edition, 2007.
Gregory Nellis and Sanford Klein. chapter EXAMPLE 9.2-1. Cambridge
University Press, 2009. URL http://cambridge.org/us/engineering/
author/nellisandklein/downloads/examples/EXAMPLE_9.2-1.pdf.
Diffusion Coefficient for Air-Water Vapor Mixtures.
PDS: The Planetary Atmospheres Data Node. Humidity, 9 2009. URL
http://atmos.nmsu.edu/education_and_outreach/encyclopedia/
humidity.htm.
Donal OCallagan and Peter Cunningham. Modern process control techniques in the production of dried milk products a review. EDP Sciences,
2005.
OLK and GEA NIRO. Nozzle program.

BIBLIOGRAPHY

113

Spyridon E. Papadakis. Air temperature and humidity profiles in spray

drying. 1. features predicted by the particle source in cell model. Industrial
Engineering Chemistry Research, 27,11:21112116, 1988.
Bhesh Bhandari Pilairuk Boonyai and Tony Howes. Stickiness measurement
techniques for food powders: a review. Powder Technology, 145:3446,
2004.
Ejnar Refstrup. Density of milk concentrate.
Ejnar Refstrup and GEA NIRO. Gea niro data.
Shelquist Engineering Richard Shelquist. Air density and density altitude calculations, 9 2009. URL http://wahiduddin.net/calc/density_
altitude.htm.
Maykel VERSCHUEREN Jan J. VAN HAREN Erik SMIT Gerrald BARGEMAN Ruud E.M. Verdurmen, Han STRAATSMA and Peter DE JONG.
Modelling spray drying processes for dairy products. EDP Sciences, 82:
453463, 2002.
Vikram Shabde. Optimal design and control of a spray drying process that
manufactures hollow micro particles. Ph.d, Texas Tech University, Chemical Engineering, December 2006.
Jiro Koga Shunji Homma, Shinji Ogata and Shiro Matsumoto. Gas solid
reaction model for a shrinking spherical particle with unreacted shrinking
core.
Sigurd Skogestad and Ian Postlethwaite. Multivariable Feedback ControlAnalysis and Design. Wiley, 2nd edition, 2005.
Jim R. Jones . John E. Bronlund Stephen R.L. Werner, Richard L. Edmonds
and Anthony H.J. Paterson. Single droplet drying: Transition from the
effective diffusion model to a modified receding interface model.
A. S. Mujumdar b V. S. Birchal a, L. Huang bc and M. L. Passos. Spray
dryers: Modeling and simulation. Drying Technology, 24,3:359371, 2008.
Vagn Westegaard. Milk powder technology- Evaporation and Sp
aray drying.
Niro A/s, 5th edition, 2004.
Pieter R. Wiederholt. Water Vapor Measurement: Methods and Instrumentation. Marcel Dekker INC, 1997.

114

BIBLIOGRAPHY

Appendix

A
Appendix A

116

Appendix A

Desorption Isotherm at low and high humidity level

/Kg

solid

A.1

0.28

4C

0.26

25 C

0.24

37 C

0.22

50 C

0.2

65 C

Increasing Temperature

85 C

0.18

100 C

0.16

115 C

0.14
0.12

0.79

0.8

0.81

0.83

0.84

Desorption Isotherm for Maltodextrin DE12 at various Temperatures

0.82

Water activity a
x 10

H O

Equilibrium Moiosture Content (Kg

X: 0.83
Y: 0.2023

X: 0.82
Y: 0.1893

solid

Moiosture Content (Kg

/Kg Equilibrium
)
H

Desorption Isotherm for Maltodextrin DE12 at various Temperatures

o

4C
o

25 C

2.5

Increasing Temperature

37 C
o

50 C

65 C
o

1.5
1

85 C

X: 0.03
Y: 0.001541

X: 0.02
Y: 0.0008069

100 C
o

115 C

0.5
0.02

0.022

0.024

0.026

0.028

0.03

0.032

Water activity a

Figure A.1: Desorption Isotherm Maltodextrin DE12: Equilibrium moisture content as function of water activity for temperatures between 40 C and
1150 C. First figure at low water activity level and the second figure
is for high water activity level

117

A.2

General moisture Characteristic and food microbiology

Figure A.2: General moisture Characteristic and Food microbiology.

content vs. Water activity is the product characteristic.
reaction are slow for water activities below 0.5. Micro
growth does not take place for water activity levels below

Mositure
Chemical
biological
0.75.

118

Appendix A

Appendix

B
Appendix B

120

B.1

Appendix B

Modelling Variables

Variable Name
FM AIN
Tmain
FSF B
TSF B
Fcool
Tcool
Ff eed
f eed
Tf eed
Tamb
RHamb
Solid
XIn
Patm
YIn
Cdryair
Cvapour
Cdryair
Cwater

Uloss
AChamber

Default Operation Values

Description
Value
Main air flow IN
1800
Temperature of MAIN
160
SFB air flow IN
500
Temperature of SFB
90
Cooling air flow IN
80
Temperature of cooling air
60
Feed flow IN
65
Density of feed
1.208
Temperature of feed
50
Temperature of ambient air
30
Rel. humidity ambient air
28
solids of total feed
50
Initial Moisture Content
1
Standard Atmospheric pressure
101325
Absolute Humidity air IN
0.0079
Specific Heat Capacity
Dry air
1
vapour
1.8
Maltodextrin
1.5
water
4.2
latent heat of vaporisation
2.5
Qloss
heat transfer coefficient (chamber) 16.75
Surface area(chamber)
26.2

Unit
Kg/h
oC
Kg/h
oC
Kg/h
oC
L/h
Kg/L
oC
oC
%
%
Kg/Kgsolid
Pa
Kg/Kgdryair
KJ/(Kg K)
KJ/(Kg K)
KJ/(Kg K)
KJ/(Kg K)
KJ/(Kg)
KJ/m2 /h
m2

Table B.1: The manipulated variables default operation values for test of the
models

121

B.2

Steady State Calculation

Steady State Results
Loss Included
TEST
Test
TOutAir
TOutAir
Description
Temperature Temperature
Default System
86.1o C
86o C
o
Ff eed = 75 L/h
80.6 C
81o C
Tmain = 150 o C
79.9o C
79o C
o
o
Tmain = 170 C
92.4 C
90o C
FM AIN = 2000 kg/h
90.9o C
89o C
o
FM AIN = 1600 kg/h
80.6 C
FSF B = 600 kg/h
86.2o C
83o C
FSF B = 350 kg/h
85.9o C
84o C
o
o
TSF B = 80 C
84.4 C
82.5o C
TSF B = 100 o C
87.9o C
85.6o C
o
RHamb = 75 %
102 C
86.7o C
o
RHamb = 10 %
101.4 C
85.9o C

Table B.2: Steady State Results for the drying air temperature TOutAir calculated
with the variables and values as used in the real test on MSD-20. Calculated with a energi loss funtion included and without a loss function.
These are compared with the results from the Test on MSD-20

122

Steady State Results for humidity

Loss Included Loss Included
Test
Eq Moist.
YOut
Description
Content
Abs. Hum
Default System
0.0033
0.0251
Ff eed = 75 L/h
0.0047
0.0277
o
Tmain = 150 C
0.0045
0.0251
Tmain = 170 o C
0.0024
0.0251
FM AIN = 2000 kg/h
0.0025
0.0237
FM AIN = 1600 kg/h
0.0047
0.0267
FSF B = 600 kg/h
0.0032
0.0244
FSF B = 350 kg/h
0.0035
0.0263
TSF B = 80 o C
0.0036
0.0251
TSF B = 100 o C
0.0030
0.0251
o
Tamb = 50 C
0.0051
0.0397
Tamb = 10 o C
0.0026
0.0198
RHamb = 75 %
0.0050
0.0386
RHamb = 10 %
0.0026
0.0199

Appendix B

TEST
Abs. Hum
Outlet Air
0.010
0.012
0.011
0.010
0.010
0.0115
0.012
82.5o C
85.6o C

Table B.3: Steady State Results of absolute humidity and equlibrium moisture
content calculated with the variables and values as used in the real test
on MSD-20. Calculated with a energi loss funtion included. These are
compared with the results from the Test on MSD-20. The absolute
humidity is calculated from sensor 1616 and 1614.

123

B.3

Results from the Dynamic Model before modification

B.3.1

Main inlet air temperature step up

Main Air Temperature step Tmain=150>170 oC
180

Temperature [oC]

175
170
165
160

Model: No delay
Model: with delay =180
TEST 7 MSD20

155
150
145

1000

1500

2000

2500

3000

3500

4000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for Main Inlet air Temperature change
94

Temperature [oC]

92
90
88
86
84

Model: No delay
Model: with delay =180
TEST 7 MSD20

82
80
78

1000

1500

2000

2500

3000

3500

4000

Time [sec]

Figure B.1: Dynamic step response of the TOutAir for increase in main inlet air
temperature from 1500 C to 1700C . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

124

Appendix B

B.3.2

Feed Flow step down

Feed flowrate step F

=75>65 L/h

feed

Feed Flow [L/hour]

85

Model: No delay
TEST 4 MSD20

80
75
70
65
60
900

950

1000

1050

1100

1150

1200

1250

1300

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for a Feed Step Change
87

Temperature [oC]

86
85
84
83
82

Model: No delay
TEST 5 MSD20

81
80

1000

1500

2000

2500

3000

3500

4000

4500

5000

Time [sec]

Figure B.2: Dynamic step response of the TOutAir for a decrease in feed flow rate
from 75L/h to 65L/h. The top figure shows the applied step change.
The bottom figure is the step response. Model results are compared
with data from MSD-20(red)

125

B.3.3

Temperature SFB step down

SFB Air Temperature step T

=90>80 C

main

Temperature [oC]

105
Model: No delay
Model: with delay =500
TEST 16 MSD20

100
95
90
85
80
75

1000

1500

2000

2500

3000

3500

4000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for SFB Inlet air Temperature change

Temperature [oC]

88
Model: No delay
Model: with delay =500
TEST 16 MSD20

86

84

82

80

1000

1500

2000

2500

3000

3500

4000

Time [sec]

Figure B.3: Dynamic step response of the TOutAir for decrease in SFB inlet air
temperature from 900 C to 800 C . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

126

Appendix B

B.3.4

Temperature SFB step up

SFB Air Temperature step T

=80>100 C

main

Temperature [oC]

100
95
90

Model: No delay
Model: with delay =500
TEST 17 MSD20

85
80
1000

1500

2000

2500

3000

3500

4000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for SFB Inlet air Temperature change

Temperature [oC]

88

86

84

Model: No delay
Model: with delay =500
TEST 17 MSD20

82

80

1000

1500

2000

2500

3000

3500

4000

Time [sec]

Figure B.4: Dynamic step response of the TOutAir for increase in SFB inlet air
temperature from 800 C to 1000 C . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

127

B.3.5

SFB air flow step down

SFB Air inlet flow step Fmain=500>350 kg/h

Temperature [ C]

620
600
580
560
540
520

Model: No delay
Model: with delay =160
TEST 13 MSD20

500
480
1000

1500

2000

2500

3000

3500

4000

4500

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for SFB Inlet air Temperature change

Temperature [ C]

86

Model: No delay
Model: with delay =160
TEST 13 MSD20

85

84

83

82
1000

1500

2000

2500

3000

3500

Time [sec]

Figure B.5: Dynamic step response of the TOutAir for increase in SFB inlet air
flow from 500kg/h to 600kg/h . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

128

Appendix B

B.3.6

SFB air flow step up

SFB Air inlet flow step F

=350>500 kg/h

main

650

Model: No delay
Model: with delay =160
TEST 14 MSD20

Temperature [oC]

600
550
500
450
400
350
300
900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for SFB Inlet air Temperature change

Temperature [oC]

87

Model: No delay
Model: with delay =160
TEST 14 MSD20

86
85
84
83
82

800

1000

1200

1400

1600

1800

2000

2200

2400

2600

2800

3000

Time [sec]

Figure B.6: Dynamic step response of the TOutAir for decrease in SFB inlet air
flow from 600kg/h to 350kg/h . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

129

B.3.7

Main air flow step down

Main Air inlet flow step F

=2000>1800 kg/h

main

Model: No delay
Model: with delay =40
TEST 10 MSD20

2000

Temperature [ C]

2050

1950
1900
1850
1800
1750
900

1000

1100

1200

1300

1400

1500

1600

1700

1800

1900

2000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for Main Inlet air Temperature change
Model: No delay
Model: with delay =40
TEST 10 MSD20

90

Temperature [ C]

92

88

86

84
800

1000

1200

1400

1600

1800

2000

2200

2400

2600

2800

3000

Time [sec]

Figure B.7: Dynamic step response of the TOutAir for decrease in Main inlet air
flow from 2000kg/h to 1800kg/h . The top figure shows the applied
step change. Fan is a first order system = 40. The bottom figure is
step response. Model results are compared with data from MSD-20

130

Appendix B

B.4

Results from the Dynamic Model after modification

B.4.1

Feed Flow step down with modefication

Feed flowrate step Ffeed=75>65 L/h (modified)

Feed Flow [L/hour]

80

Model: with delay =5

TEST 2 MSD20
75

70

65

60
6900

7000

7100

7200

7300

7400

7500

7600

7700

7800

7900

8000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for a Feed Step Change(Modified model)

Temperature [oC]

86

84

82

80

evap=1300 & fast=600

TEST 2 MSD20

78
7000

7500

8000

8500

9000

9500

10000

10500

11000

Time [sec]

Figure B.8: Dynamic step response of the TOutAir for a decrease in feed flow rate
from 75L/h to 65L/h. The top figure shows the applied step change.
The bottom figure is the step response. Model results are compared
with data from MSD-20(red)

B.4.2

Temperature MAIN step down with modification

Main Air Temperature step T

=160>150 C

main

Model: No delay
Model: with delay =180

160

Temperature [ C]

165

155
150
145
140
6800

7000

7200

7400

7600

7800

8000

8200

8400

8600

8800

9000

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for a T

Main

Step Change(Modified model)

evap=1300 & fast=600

Temperature [oC]

86

TEST 2 MSD20
84
82
80
78
7000

7500

8000

8500

9000

9500

10000

10500

11000

Time [sec]

Figure B.9: Dynamic step response of the TOutAir for decrease in MAIN inlet air
temperature from 1600 C to 1500 C . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

131

B.4.3

Air flow MAIN step up with modification

Main Air inlet flow step F

=1800>2000 kg/h (modified)

main

Air flow [Kg/h]

2000
1950
1900
1850

Model: with delay =40

MSD20

1800
6900

7000

7100

7200

7300

7400

7500

7600

Time [sec]

Temperature in Spray Dryer Chamber Spray Dryer for Main Inlet air flow change

Temperature [oC]

90
89
88
87
86

Model:with delay =40

MSD20

85
84
7000

7500

8000

8500

9000

9500

10000

10500

11000

Time [sec]

Figure B.10: Dynamic step response of the TOutAir for increase in MAIN inlet air
flow from 1800kg/h to 2000kg/h . The top figure shows the applied
step change. The bottom figure is step response. Model results are
compared with data from MSD-20

132

Appendix B

B.4.4

Absolute Humidity in Dryer For feed step up with

modification
Absolute humidity in Spray Dryer Chamber for a Feed Step Change (65>75 l/h)(Modified model)
0.03

=1000 &

evap

=500

fast

evap=1300 & fast=600

Moisture content [kgwater/kgdry air]

0.029

evap=1000 & fast=100

=1000 &

evap

=1500

fast

0.028

0.027

0.026

0.025

0.024

0.023

7000

7500

8000

8500

9000

9500

10000

10500

11000

Time [sec]

Figure B.11: Absolute Humidity in Dryer For feed step up with model modification. For various zero and pole in dry transfer function

B.4.5

Response for the system G(s)evap

Output of G(s)evap for various zero and pole combination for feed step (65 >75 l/h)

=1000 &

evap

=500

fast

evap=1300 & fast=600

evap=1000 & fast=100

0.013

=1000 &

Moisture evaporation rate (kg /s)

evap

=1500

fast

0.0125

0.012

0.0115

0.011

0.0105

7000

7500

8000

8500

9000

9500

10000

10500

11000

Time [sec]

Figure B.12: Response for the system G(s)evap . Evaporation rate (kg/s) as a
function of time

133

B.4.6

Simulink model

Figure B.13: Simulink implementation of the dynamic model. Left hand side:
process inputs. Right hand Side: Drying chamber block, relative
humidity block and equilibrium moisture content block. red block
are subsystems

134

Appendix B

Appendix

C
Appendix C

C.1

MSD-20 test 24-7-2009

Sensor Name
MAINKGH
SFBKGH
T1702
T1704
F1626
P1706
T1624
P1618
T1709
T1616
P1614
INHUMABS
OUTHUMABS

Sensor Description1
Description
Input Sensors
Main air flow into chamber
SFB air flow into chamber
Temperature of MAINKGH
Temperature of SFBKGH
Feed flow into chamber
Pressure difference of powder layer in SFB
Temperature of ambient temperature
Relative humidity of ambient air
Output Sensors
Temperature of air flow out of chamber
Temperature of air leaving system
Relative humidity of air leaving system
Calculated parameters
Absolute humidity air IN
based on sensor values (T1624 & P1618)
Absolute humidity air OUT
based on sensor values (T1616 & P1614)

Value

Unit

PV
PV
PV
PV
PV
PV
PV
PV

Kg/h
Kg/h
oC
oC
L/h
mmH2 O
oC
%

PV
PV
PV

oC

PV

Kg/Kg

PV

Kg/h

Table C.1: PV:process value, Out: Controller Output value

oC

136

C.2
C.2.1

Time
8.00
9.00
10.00
11.00
12.30
14.00
15.30
17.00
18.30
20.00
21.30
23.00
00.30
2.00

3.30
4.30
5.30
6.30
7.30
8.30
9.30
10.30
11.30
12.30
13.30

Appendix C

Test Step & Responses

Test Program MSD-20 week 30 2009
Test Program MSD-20
Test
Step
Powder Sample
Description
Value
Interval(min)
System Start Up
Set Default System
Feed Flow 1626
1
Stable system:Test start
65 L/h
5
2
Feed Flow Step Up
75 L/h
5
3
Feed Flow Step Down
65 L/h
5
4
Feed Flow Step Up
75 L/h
5
5
Feed Flow Step down
65 L/h
5
Main Inlet Air Temperature 1702 (Default 160)
6
Temperature Step down
150 o C
10
7
Temperature Step Up
170 o C
10
o
8
Temperature Step down
160 C
10
Main Inlet Air Flow MAINKGH(Default 1800 Kg/h)
9
MAINKGH Step Up
2000 Kg/h
10
10
MAINKGH Step Down 1800 Kg/h
10
11
MAINKGH Step Up
2000 Kg/h
10
(NO PI 1702)
12
MAINKGH Step Down 1800 Kg/h
10
(NO PI 1702)
SFB Inlet Air Flow SFBKGH(Default 500 Kg/h)
13
SFGKGH Step Up
600 Kg/h
10
14
SFBKGH Step Down
350 Kg/h
10
15
SFBKGH Step Up
500 Kg/h
10
SFB Inlet Air Flow Temperature(Default 90)
16
Temperature Step down
80 o C
10
17
Temperature Step Up
100 o C
10
18
Temperature Step down
90 o C
10
VFB
19
VFB Flow Step Up
400 kg/h
10
20
VFB Flow Step Down
300 Kg/h
10
21
VFB Temp Step Up
70 o C
10
22
VFB Temp Step Down
60 o C
10
END

Test step

Table C.2: Test Program for Test On MSD-20 week 30 2009. In total 22 Test
Steps. Results are found on CD

C.2 Test Step & Responses

137

Test Step & Results for the entire test on MSD-20

TEST STEPS & RESULTS FROM MSD20 (21/72009)COMPARED
3000

Kg/hour

Appendix C

C.2.2

MAINKGH
SFBKGH

2000

1000

Time [s]

10
4

x 10

T1702
F1626
T1704
P1706
T1709

150

100

50
1

Kg water/kg dry air

Time [s]
0.015

InHumAbs
ExHumAbs

0.01

0.005

NIGHT

DAY
0

DAY
5

Time [s]

138

x 10

10
4

x 10

Figure C.1: Test Step & Results for the entire test on MSD-20 to get a overview and quickly compare the results. For more detailed
plots of test steps and responses see the following sections

Moisture content of the particle from the SFB discharge for the entire test on MSD-20
3

Powder RM VFB
Powder RM SFB

RM %H20/kg powder

2.5

C.2 Test Step & Responses

C.2.3

1.5

0.5

Time [s]

10
4

x 10

Figure C.2: Moisture content of the particle from the SFB discharge for the entire test on MSD-20.

139

Test Step & Results for change in feed rate on MSD-20

Test: Feed rate 1626
100

Feed Flow 1626

Liter/hour

Appendix C

C.2.4

80
60
40

0.5

1.5

2.5

/Kg

3
4

x 10

Time Respons Outlet Air Temperature 1709

88

Outlet Air Temp. 1709

86
84
82
80

solid

)
Temperature (celcius)

Time [s]

0.5

1.5

2.5

Time [s]

3
4

x 10

140

Moiosture Content (Kg

H O

Outlet Powder Moisture Content

2.5

Powder Moisture Content

2
1.5
1
0.5

Test1

0.5

Test2

1.5

Time [s]
Test3

2.5

3
4

Test5

x 10

Test4

Figure C.3: Test Step & Results for change in feed rate on MSD-20. Outlet air Temperature respons and mositure constent of the
powder taken from the SFB discharge.

Feed flow rate and Nozzle pressure results from test on MSD-20

C.2 Test Step & Responses

C.2.5

Feed rate from MSD20 test

Feed rate L/h

Liter/hour

80
75
70
65
60
1

Time [s]

x 10

Feed Nozzle Pressure (bar)

320

Nozzle pressure
pressure [bar]

300
280
260
240
220
200
180

Time [s]

10
4

x 10

141

Figure C.4: Feed flow rate and Nozzle pressure results from test on MSD-20. Feed flow is controlled by controlling the nozzle pressure.
Increase in nozzle pressure increases feed rate and similarly for decreasing the pressure the feed rate decreases.

Test Step & Results for change in Main inlet air temperature on MSD-20
Temperature (celcius)

Appendix C

C.2.6

Test: Main Inlet Air Temperature

180

Main Air Temp. 1702

170
160
150
140

2000

4000

6000

8000

10000

12000

14000

16000

18000

/Kg

solid

)
Temperature (celcius)

Time [s]

Time Respons Outlet Air Temperature 1709

95

Outlet Air Temp. 1709

90
85
80
75

2000

4000

6000

8000

10000

12000

14000

16000

18000

Time [s]

142

Moiosture Content (Kg

H O

Time Respons Outlet Powder Moisture Content

2

Powder Moisture Content

1.5
1
0.5

2000

4000

6000

8000

10000

12000

14000

16000

18000

Time [s]

Test 6
Test 7

Test 8

Figure C.5: Test Step & Results for change in Main inlet air temperature on MSD-20. Outlet air Temperature respons and mositure
constent of the powder taken from the SFB discharge.

Test Step & Results for change in Main inlet air flow on MSD-20
Temperature (celcius)

Test: Main Inlet Air MAINKGH

2100

MainKGH
2000
1900
1800
1700

0.5

1.5

/Kg

2.5
4

x 10

Time Respons Outlet Air Temperature 1709

90

Outlet Air Temp. 1709

85
80
75

solid

)
Temperature (celcius)

Time [s]

C.2 Test Step & Responses

C.2.7

0.5

1.5

Time [s]

2.5
4

x 10

Moiosture Content (Kg

H O

Time Respons Outlet Powder Moisture Content

Powder Moisture Content
1.8
1.6
1.4

Test 9

0.5

Test 10

1.5

Time [s]

2.5
4

Test 12

x 10

Test 11

143

Figure C.6: Test Step & Results for change in Main inlet air flow on MSD-20. Outlet air Temperature respons and mositure constent
of the powder taken from the SFB discharge.

Test Step & Results for change in SFB inlet air temperature on MSD-20
Temperature (celcius)

Appendix C

C.2.8

Test: SFB Inlet Air Temperature

110

SFB Air Temp. 1704

100
90
80
70

2000

4000

6000

8000

10000

12000

Time Respons Outlet Air Temperature 1709

86

Outlet Air Temp. 1709

84
82
80

/Kg

solid

)
Temperature (celcius)

Time [s]

2000

4000

6000

8000

10000

12000

Time [s]

144

Moiosture Content (Kg

H O

Time Respons Outlet Powder Moisture Content

2.5

Powder Moisture Content

2
1.5
1
0.5

2000

4000

6000

8000

10000

12000

Time [s]

Figure C.7: Test Step & Results for change in SFB inlet air temperature on MSD-20. Outlet air Temperature respons and mositure
constent of the powder taken from the SFB discharge.

Temperature (celcius)

Test Step & Results for change in SFB inlet air flow on MSD-20
Test: SFB Inlet Air SFBKGH
700

SFBKGH
600
500
400
300

2000

4000

6000

8000

10000

12000

Time Respons Outlet Air Temperature 1709

85

Outlet Air Temp. 1709

84
83
82

/Kg

solid

)
Temperature (celcius)

Time [s]

C.2 Test Step & Responses

C.2.9

2000

4000

6000

8000

10000

12000

Time [s]

Moiosture Content (Kg

H O

Time Respons Outlet Powder Moisture Content

2.5

Powder Moisture Content

2
1.5
1

2000

4000

6000

8000

10000

12000

Time [s]

145

Figure C.8: Test Step & Results for change in SFB inlet air flow on MSD-20. Outlet air Temperature respons and mositure constent of
the powder taken from the SFB discharge.

Ambient Air Conditons At AIR intake (21/7-2009)

Relative Humidity at AIR intake (Sensor 1618)
40

Rel. Humidity
Relative humidity%

Appendix C

C.2.10

35

30

25

20

15

Time [s]

10
4

x 10

Temperature of AIR at Intake (sensor 1624)

34
o

Temperature C
o

Temperature C

32

30

28

26

DAY

Night

DAY
24

146

Time [s]

Figure C.9: Ambient Air Condition at Intake during the test

10
4

x 10

C.3 Logbook for MSD-20 test 24-7-2009

C.3

Logbook for MSD-20 test 24-7-2009

Figure C.10: Logbook From Test on MSD-20 week 30 2009

147

148

Appendix C

Figure C.11: Logbook From Test on MSD-20 week 30 2009

C.3 Logbook for MSD-20 test 24-7-2009

149

150

Appendix C

Appendix

D
Appendix D

D.1

Humidity Calculation

Ideal Gas Law

P V = nRT

(D.1)

P = pressure(P a)
V = V olume
T = T emperature(kelvin)
R = Universal Gas Constant(J/K mol)
n = number of moles(mole)n

= m/M

M = Molar mas(Kg/Kmol)
m = amount of substance(Kg)
Molar Mass Water

Mw = 18.01(g/mol)

(D.2)
(D.3)

Molar Mass dry air

MdryAir = 28.96(g/mol)

(D.4)

152

Appendix D

Density
n
m/M
=
V
V
m/M
m 1
=
V
V M
m Kg
= ( 3)
V m

(D.5)

Pressure
Ptotal = PDry + PV apour
P
n
=
V
RT
P
=M
RT
total = Dry + V apour
Pdry
Pvapour
= MdryAir
+ Mw
RT
RT
Pvapour
P Pvapour
+ Mw
= MdryAir
RT
RT

(D.6)

(D.7)
(D.8)

Volume of Chamber

Dchamber 2 Hchamber
(D.9)
4
Dchamber 3
1
1
+
+
(
)

Cone
24
T AN ( 180 2 ) T AN ( 180 (90 ACeiling ))

V =

Dchamber = Diameter of Chamber

Hchamber = Height of Chamber
Acone = Chamber cone angle
Aceiling = Chamber ceiling angle

D.2

Droplet calculations

Surface Area of Sphere

Adrop = 4r 2 = d2
r = radius of sphere
d = diameterof droplet

(D.10)

D.3 Thermal Conductivity Air

153

Volume of Sphere

Vdrop = 4/3 r 3

(D.11)

r = radius of sphere
Diameter change of particle during evaporation (Xparticle > Xc )

VdropChange = VdropInit Vdrop

(D.12)
3

DdropInit
Ddrop
VdropChange = 4/3
4/3
2
 2

Ms (X0 Xparticle )
4 1
=
(DdropInit 3 Ddrop 3 )
water
3 23


6Ms (X0 Xparticle ) 1/3
3
Ddrop = DdropInit
water

(D.13)
(D.14)

Diameter change of particle during evaporation (Xparticle < Xc )

Ddrop =

DdropInit

6Ms (X0 XC )

water

1/3

(D.15)

Number of drops

Ndrops =

Ff eed
Vdrop

(D.16)

Mass of a single drop

M assdrop =

D.3

Ff eed f eed
M asssolid = M assdrop Sconc Sconc = Solid concentration in feed
Vdrop
(D.17)

Thermal Conductivity Air

Air Prperties table http://www.engineeringtoolbox.com/air-properties-d_

156.html Coefficients:

154

Appendix D

Figure D.1: Thermal Conductivity of Air vs. Temperature in kelvin. http://

users.wpi.edu/~ierardi/PDF/air_k_plot.PDF

4
3
2
kair = p1kair Tgas
+ p2kair Tgas
+ p3kair Tgas
+ p4kair Tgas + p5kair
(D.18)

p1k air = 2.3572e 013

p2kair = 1.3447e 010
p3kair = 3.3402e 008
p4kair = 7.5159e 005
p5kair = 0.024159

D.4

Thermal Diffusivity Air

D.5

Kinematic Viscosity Air

Kinematic viscosity from absolute viscosity

=

(D.19)

D.6 Mean Residense Time for the particle in the Spray dryer

155

Figure D.2: Thermal Diffusivity of Air vs. Temperature in kelvin. http://

users.wpi.edu/~ierardi/PDF/air_alpha_plot.PDF

Reynolds number Inertial forces are characterized by the product of the

density r times the velocity V times the gradient of the velocity dV/dx.
The viscous forces are characterized by the dynamic viscosity coefficient mu
times the second gradient of the velocity d2 V /dx2 . http://www.grc.nasa.
gov/WWW/BGH/reynolds.html
Re =
Re =

Inertia force
Viscous force
V dV
dx
2

ddxV2
V L
Re =

D.6

(D.20)
(D.21)
(D.22)

Mean Residense Time for the particle in the

Spray dryer

Mean particle residence time The mean residense time for the powder
in the spray dryer can be calculated by the amount of powder in the SFB

156

Appendix D

Figure D.3: Kinematic Viscosity of Air vs. Temperature in Kelvin. http://

users.wpi.edu/~ierardi/PDF/air_nu_plot.PDF

divided by the amount of powder leaving(either per second or per hour).

The mass of powder in the SFB is calculated from the size of the SFB, the
height of the powder layer in SFB and the density of the of the agglomerated
powder(V olume density). The time from the powder enter the dryer and
until it ends in the SFB is very small compared to the residence time in the
SFB and the therefore neglected in the calculations.
SF Barea = 0.1886m2
powderlayerSF B = 0.35 ( Height of the powder layer in SFB)
kg
powder = 300 3 (agglomerated Maltodextrin powder density)
m
kgwater
(mositure content in particle at equilibrium)
Xout = 0.03
kgpowder
powderSF B = SF Barea pulverl ag densp owderkg powder in SFB
Fs olid = F eedf lows 1 Solida mountDry solids feed flow per sec
mp owder = Fs olid (1 + Xo ut) 3600mass powder leaving the chamber per hour
residenst M EAN = powderS F B/mp owder 3600Mean residense time in hours
residenst M EAN = 0.59hours 36min

(D.23)

The mean particle residense time in MSD-20 spray dryer during the experiment was 36 min.

D.6 Mean Residense Time for the particle in the Spray dryer
Bv

157

Simulink regards disturbance v as in input

 to the system.Bv is therefore a part of the
u(t)
determined matrix B. 6.1 is then B
v(t)

158

Appendix D

Appendix

E
Appendix E

160

E.1

Appendix E

Air and Particle Trajectory in Chamber

Figure E.1: Air Stream and particle trajectory in chamber. Particles follows the
air stream.CFD simulation GEA NIRO

Appendix

F
Appendix F

F.1

Mass transfer and Drying time appendix

F.1.1

Default state operation condtion

Mass evaporated at default steady state Operation for particles of different size

12

x 10

= 57 m

Particle D

drop

Particle Ddrop= 76 m

Particle Ddrop= 114 m

= 152 m

Particle D

drop

Particle Ddrop= 228 m

= 304 m

Particle D

Mass evaporated(Kg

H O

drop

0.5

1.5

2.5

3.5

4.5

Time [sec]

Figure F.1: Mass evaporated for various particles sizes at default state operation
conditions vs. time

F.1.2

TOut varied

162

Appendix F

Mass Transfer coefficients at default operation for different particle sizes

6.5

Particle D

= 57 m

Particle D

= 76 m

Particle D

= 114 m

Particle D

= 152 m

Particle D

= 228 m

drop
drop

drop
drop

5.5

drop

Particle Ddrop= 304 m

(m/s)

mass

4.5

3.5

2.5

0.5

1.5

2.5

3.5

4.5

Time [sec]

Figure F.2: Mass Transfer coefficient for various particles sizes at default state

Resistance f in crust in mass tranfer rate equation

150

Particle D

= 57 m

Particle D

= 76 m

drop
drop

Particle Ddrop= 114 m

100

Particle D

= 152 m

Particle D

= 228 m

Particle D

= 304 m

drop
drop
drop

f resistance

50

50

100

150

0.5

1.5

2.5

3.5

4.5

Time [sec]

Figure F.3: Crust resistance f in mass transfer rate equation for various particle
sizes

F.1 Mass transfer and Drying time appendix

163

Drying Time for Particle at various Drying Temperature by changing Feed rate
Feed rate = 65 L/h
Feed rate = 75 L/h
Feed rate = 105 L/h
Feed rate = 35 L/h

solid

0.25

Moiosture Content (Kg

H O

/Kg

0.2

0.15

0.1

0.05

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0.55

Time [sec]

Figure F.4: Drying time for particle at various feed flow rate, thus varying outlet
air temperature. Moisture content vs. time

f resistance in crust in mass tranfer rate equation at changing feed rate

70

Feed rate = 65 L/h

Feed rate = 75 L/h
Feed rate = 105 L/h
Feed rate = 35 L/h

60

f resistance

50

40

30

20

10

0.1

0.2

0.3

0.4

0.5

Time [sec]

Figure F.5: Crust resistance f for various feed flow rates, thus varying outlet air
temperature. Moisture content vs. time. f resistance in crust takes
longer to increase for higher feed rates.

164

Appendix F

Drying Time for Particle at various Drying Temperature by MAINinlet

Main Inlet air= 1800 Kg/h
Main Inlet air= 2500 Kg/h
Main Inlet air= 1000 Kg/h

0.2

Moiosture Content (Kg

H O

/Kg

solid

0.25

0.15

0.1

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

Time [sec]

Figure F.6: Drying time for particle at various Main inlet air flow, thus varying
outlet air temperature. Moisture content vs. time

Appendix

G
Appendix G

G.1

1st order system

1. Order system
1
G(s) =
s + 1
Step response

(G.1)

y(t) = (1 et/ )hstep + h0

(G.2)

Figure G.1: Control variable as a first order system. The gain is equal to 1/

166

Appendix G

Appendix

H
Appendix H

168

H.1

Appendix H

System Identification

Linear difference equation for the BEST ARX model(10th order)

A(q)y(t) = B(q)u(t) + e(t)
A(q) = 1 1.899(+ 0.01379)q 1 + 0.8909(+ 0.02959)q 2
+ 0.0004731(+ 0.03204)q 3 + 0.0001774(+ 0.03204)q 4
0.0003988(+ 0.03204)q 5 + 0.01811(+ 0.03204)q 6
0.01516(+ 0.03204)q 7 0.0005041(+ 0.03204)q 8
+ 0.02059(+ 0.02961)q 9 0.01521(+ 0.01382)q 10
B1(q) = 1.939e 006(+ 1.846e 006)q 8
B2(q) = 8.505e 005(+ 4.977e 005)q 8
B3(q) = 1.792e 006(+ 1.318e 006)q 8
B4(q) = 3.227e 005(+ 1.647e 005)q 8
B5(q) = 4.07e 005(+ 3.612e 005)q 8
B6(q) = 0.1174(+ 0.1623)q 8
B7(q) = 7.449e 006(+ 1.87e 005)q 8
B8(q) = 0.0001645(+ 0.000102)q 8
B9(q) = 0.3597(+ 0.3024)q 8
Estimated using ARX with focus from data set data
Loss function 0.00105043 and FPE 0.00105798
Sampling interval: 10

H.2

ARMAX models
na
10
8
6
5
2
10
2
10

nb
5
7
7
7
7
2
2
5

nc
8
8
2
8
8
8
2
8

ARMAX models
nk
Loss
FPE
8
0.0010
0.0010
8
0.000171 0.000176
8
0.000169 0.000174
8
0.000168 0.000178
8
0.00022
0.00023
8
0.00025
0.00025
1
0.00242
0.00224
1
0.00020 0.000213

Best
29
24
36
33
-49
-48
-39
-45

Table H.1: ARMAX model with number of coeffiecients in polynomial A (na), B

(nb), C (nc), and delays (nk). The quality parameters Loss function,
Final Prediction Error, and Best fit for the given model.

Plots of the ARMAX model listed in H.1. The figures show the simulated

H.1 System Identification

H.1.1

169

Model Misfit Vs Number parameters for ARX model

3

x 10

Model Misfit vs number of parameters in the ARX model (9 inputs 1 output)

Unexplained output variance (in %)

0.9

best fit 19 parameters

na =10, nb=1, nk=8

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

20

40

60

80

100

120

Number of parameters

Figure H.1: Model Misfit Vs Number parameters for ARX model. Test range
1:10 for all coefficients. Model order of 10 with a total number of
parameters of 19 seems to be the best with lowest Unexplained output variance (in %), which is the ratio between the prediction error
variance and the output variance in percent. The final prediction
error is also the lowest for this model.

output compared to the measured output (moisture content) from the validation data.The coefficient of the model is given in the title of the figure.

170

H.1.2

Appendix H

Zero Pole plot for the ARX model

Poles (x) and Zeros (o) for the ARX model of 10th order
1

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

0.8

0.6

0.4

0.2

0.2

0.4

0.6

0.8

Figure H.2: SZero-Pole plot for the 10th order ARX model. There is a zero pole
cancellation in 0.

H.1 System Identification

H.2.1

171

ARMAX simulations

Plots of the ARMAX model listed in H.1. The figures show the simulated
output compared to the measured output (moisture content) from the
validation data.The coefficient of the model is given in the title of the
figure. Low values for the A and B polynomial and small delay results in
bad estimations.
Measured and simulated model output (10,5,8,1)
4
3.5
3
2.5
2
1.5
1
0.5
0
0.5
1

Time

10
4

x 10

Figure H.3: ARMAX model with the coefficients given in the title.(na, nb, nc,
nk)

172

Appendix H

Measured and simulated model output (2,2,2,1)

4

3.5

2.5

1.5

0.5

0.5

10

Time

x 10

Figure H.4: ARMAX model with the coefficients given in the title.(na, nb, nc,
nk)

Measured and simulated model output (8,7,8,8)

1.5

0.5

Time

9
4

x 10

Figure H.5: ARMAX model with the coefficients given in the title.(na, nb, nc,
nk)

H.1 System Identification

173

Measured and simulated model output(5,7,8,8)

2.5

1.5

0.5

Time

10
4

x 10

Figure H.6: ARMAX model with the coefficients given in the title.(na, nb, nc,
nk)

Measured and simulated model output(2,7,8,8)

4

3.5

2.5

1.5

0.5

0.5

Time

10
4

x 10

Figure H.7: ARMAX model with the coefficients given in the title.(na, nb, nc,
nk)

174

Appendix H

H.2.2

ARMAX Zero-pole plot for 6th order model

Poles (x) and Zeros (o) for the 6th order ARMAX model

0.5

0.5

0.5

0.5

Figure H.8: Zero-Pole plot for ARMAX 6th model decribed in the report. The
model is table since all the poles lies within the unit circle.

H.1 System Identification

175

Poles (x) and Zeros (o) for the 6th order ARMAX model

0.01

0.005

0.005

0.01

0.015
0.01

0.005

0.005

0.01

Figure H.9: Zero-Pole plot for ARMAX 6th model decribed in the report. The
model is table since all the poles lies within the unit circle. Zoomed
in origo

176

Appendix H

Poles (x) and Zeros (o) for the 6th order ARMAX model
0.02

0.015

0.01

0.005

0.005

0.01

0.015

0.02
0.99

0.992

0.994

0.996

0.998

1.002

1.004

Figure H.10: Zero-Pole plot for ARMAX 6th model decribed in the report. The
model is table since all the poles lies within the unit circle. Zoomed
in around +1 on the x axis

H.1 System Identification

177

Poles (x) and Zeros (o)

2
1.5
1
0.5
0
0.5
1
1.5
2
1

0.5

0.5

1.5

Figure H.11: Zero-Pole plot for State space model 4th order.Discrete time. plotted with confidence interval. Thus there is no zero pole cancellation.

Appendix H

H.2.3

State Space- continous time zero-pole plot

From u1

0.1

To y1

From u3

From u2

0.1

From u4

0.1

0.1

0.05

0.05

0.05

0.05

0.05

0.05

0.05

0.1

0.1

0.1

0.1

0.05

0.05
0.1
0.1 0.05 From
0 u5 0.05

0.05

0.1 0.05
From0u6 0.05 0.1

0.1

0.2

0.05

0.1

From u9

0.1 0.05 0 0.05

From u8

0.1

0.2

0.05

0.15

0.1
0.1

0.1

0.1

0.1 0.05

0
0.1

0.1

0.05

0.2

0.1

0.15
0.2

0.2
0.1

0.1

0.2

0.05
0.1

0.1

0.05

0.05
0

0.1

0.1

0.2

0.1

0.1

0.1 0.05

0.05 0.1

178

Figure H.12: Zero-Pole plot for State space model 4th order.Continuous time. There are no zero-pole cancellation.

0.05 0.1

Appendix

I
Appendix I

I.1.1

Linearized model- State Space(Jacobians)

A
-0.0250
0.0000
0.0000
0.0000
0.0000
0.0007
-0.0000
0.0000

0.0000
-0.0063
0.0000
0.0000
0.0000
-0.0009
-0.0000
0.0000

0.0000
0.0000
-0.0056
0.0000
0.0000
0.0413
0.0000
0.0000

0.0000
0.0000
0.0000
-0.0020
0.0000
0.0115
0.0000
0.0000

0.0000
0.0000
0.0000
0.0000
-0.2000
0.0021
0.0001
0.0002

0.0000
0.0000
0.0000
0.0000
0.0000
-0.0676
0.0000
0.0000

0.0000
0.0000
0.0000
0.0000
0.0000
-12.2654
-0.0563
-0.0001

0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0004
-0.0008

B
0.0250
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000

0.0000
0.0063
0.0000
0.0000
0.0000
0.0000
0.0000
0.0000

C 0.0000

0.0000
0.0000
0.0056
0.0000
0.0000
0.0000
0.0000
0.0000

0.0000

0.0000
0.0000
0.0000
0.0020
0.0000
0.0000
0.0000
0.0000

0.0000

0.0000
0.0000
0.0000
0.0000
0.2000
0.0000
0.0000
0.0000

0.0000

0.0000
0.0000
0.0000
0.0000
0.0000
-0.4735
-0.0101
-0.0219

0.0000

1.0000

0.0000
0.0000
0.0000
0.0000
0.0000
0.0672
-0.0000
-0.0000
0.0000

0.0000

180

Appendix I

I.1

Linearization of the dynamic Model - open

loop
Operating Point for Linearization
Variable Name
Description
Value
Input u0
u0 1
Main air flow IN
1800
u0 2
SFB air flow IN
500
u0 3
Temperature of MAIN
160
u0 4
Temperature of SFB
90
u0 5
Feed flow IN
65
Disturbance input v0 (mean)
v0 1
solids of total feed
50
v0 2
Temperature of ambient air
30
v0 3
Rel. humidity ambient air
28
v0 4
Temperature of feed
50
State x0
x0 1
Tout
85.44
x0 2
Main air flow IN
1800
x0 3
SFB air flow IN
500
x0 4
Temperature of MAIN
160
x0 5
Temperature of SFB
90
x0 6
Feed flow IN
65
x0 7
evaporation rate
0.0108
x0 8
Vapour in chamber
0.1933
output y0
y0 1
Tout
85.44

Unit
Kg/h
Kg/h
oC
oC
L/h
%
oC

%
oC
oC

Kg/h
Kg/h
oC
oC
L/h
Kg/s
Kg
oC

Table I.1: Operating Point for linearization: Stationary state

D
0.0000

0.0000

0.0000

0.0000

0.0000

0.0000

0.0000

Transfer function from input 1 to output:

s2

1.75e 005
+ 0.0926s + 0.00169

(I.1)

Transfer function from input 2 to output:

s2

5.67e 006
+ 0.0739s + 0.0004259

(I.2)

Transfer function from input 3 to output:

s2

0.0002313
+ 0.0732s + 0.0003786

(I.3)

Frequency response - from inputs to output

Bode Diagram process inputs (u) to output
From: In(1)

From: In(2)

From: In(3)

From: In(4)

From: In(5)

To: Out(1)

50

100

150

200
180

90
To: Out(1)

Magnitude (dB) ; Phase (deg)

I.1.2

90

10

10

10

10

10

10

Frequency (Hz)

Figure I.1: Frequency response of the linearized model from each process input u to output.

10

10

10

181

180
5
10

Bode Diagram feed flow input to output

Appendix I

From: In(5)
0

To: Out(1)

40
60
80
180
135
90

To: Out(1)

Magnitude (dB) ; Phase (deg)

20

45
0
45
90
135
180

182

10

10

10

10
10
Frequency (rad/sec)

10

Figure I.2: Frequency response of the linearized model from each feed flow input to output.

10

Bode Diagram solids content disturbance input to output

From: In(6)

40

To: Out(1)

System: sys
I/O: In(6) to Out(1)
Frequency (rad/sec): 0.442
Magnitude (dB): 1.28

20

40

60
360

To: Out(1)

Magnitude (dB) ; Phase (deg)

20

270

180

90
5
10

10

10

10

10

10

Figure I.3: Frequency response of the linearized model from each disturbance input (Solids content) v1 to output.

10

183

Frequency (rad/sec)

10

Bode Diagram disturbance inputs Ambient Relative Humidity (v) to output

From: In(7)

Appendix I

To: Out(1)

20
30
40
50

90
45
To: Out(1)

Magnitude (dB) ; Phase (deg)

10

0
45
90
135

184

180
5

10

10

10

10

10

10

Frequency (Hz)
Figure I.4: Frequency response of the linearized model from each disturbance input(Ambient Relative humidity) v2 to output.

I.1.3

Zero-Pole plot for the transfer functions - from each inputs to output
PoleZero Map For the transfer function from each process input u to output
From: In(1)

From: In(2)

From: In(3)

From: In(4)

From: In(5)

1
0.8

Imaginary Axis

0.6
0.4
0.2
0
0.2
0.4
0.6
0.8
1
0.08 0.06 0.04 0.02

0.08
0
0.06 0.04 0.02

0.08
0
0.06 0.04 0.02

0.08
0
0.06 0.04 0.02

0
0.2

0.2

0.4

Real Axis
Figure I.5: Zero-Pole plot for the transfer functions - from each process inputs to output. All poles and zeros are in left half plane for
the first 4 process inputs. feed flow transfer function has a zero in RHP

185

PoleZero Map: for the transfer functions from each disturbance input v to output
From: In(6)

From: In(7)

Appendix I

1
0.8

Imaginary Axis

0.6
0.4
0.2
0
0.2
0.4
0.6
0.8

186

1
0.1

0.1

0.2

0.3
0.08

0.06

0.04

0.02

Real Axis
Figure I.6: Zero-Pole plot for the transfer functions - from each disturbance inputs to output. All poles are in left half plane . Disturbances (Solids content, Ambient temperature, and Relative humidity amibient air) inputs have zeros in Right half plane.

I.2 Comparison of linear and Non linear model

187

Transfer function from input 4 to output:

s2

2.3e 005
+ 0.0696s + 0.0001352

(I.4)

Transfer function from input 5 to output:

0.00042s2 0.0002213s 3.736e 007
s4 + 0.3247s3 + 0.02885s2 + 0.0007841s + 6.095e 007

(I.5)

Transfer function from input 6 to output:

0.4735s2 + 0.09684s + 0.0001852
s3 + 0.1247s2 + 0.003905s + 3.047e 006

(I.6)

Transfer function from input 7 to output:

0.0672
s + 0.0676

(I.7)

Eigenvalues
for system matrix A- poles for

-0.0676 0.0000 0.0000 0.0000 0.0000

0.0000 -0.0008 0.0000 0.0000 0.0000

0.0000 0.0000 -0.0563 0.0000 0.0000

0.0000 0.0000 0.0000 -0.0250 0.0000

0.0000 0.0000 0.0000 0.0000 -0.0063

0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000
Eigenvector for the system matrix A

0.0000 0.0000 0.0000 0.9999 0.0000

0.0000 0.0000 0.0000 0.0000 0.9999

0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000

0.0000 0.0000 0.0000 0.0000 0.0000

1.0000 -0.7978 -1.0000 0.0164 -0.0147

0.0000 0.0043 0.0009 0.0000 0.0000

0.0000 0.6029 0.0000 0.0000 0.0000

I.2

the linearized system

0.0000 0.0000 0.0000
0.0000 0.0000 0.0000
0.0000 0.0000 0.0000
0.0000 0.0000 0.0000
0.0000 0.0000 0.0000
-0.0056 0.0000 0.0000
0.0000 -0.0020 0.0000
0.0000 0.0000 -0.2000

0.0000
0.0000
0.8323
0.0000
0.0000
0.5544
0.0000
0.0000

0.0000
0.0000
0.0000
0.9850
0.0000
0.1727
0.0000
0.0000

0.0000
0.0000
0.0000
0.0000
0.9968
-0.0798
-0.0007
-0.0010

Comparison of linear and Non linear model

Linear and nonlinear model is compared by applying a various steps on the

process and disturbance input. The step value shown in the plots are is
deviation from the linearised operating point.

188

Appendix I

Process inputs
Feed
Main inlet air flow
Main inlet air temperature
Disturbance inputs
solids content
Relative humidity of ambient air

Feed flow
T

out
85.8

outlet air temperature Linear model compared to Non linear model Feed flowrate step Ffeed

Linear model: F

feed

85.6

=2

Nonlinear model

85.4

Temperature [oC]

I.2.1

85.2

85

84.8

84.6

84.4

1.1

1.2

1.3

1.4

1.5

Time [sec]

Figure I.7: Comparison of linear and Non linear model

1.6

1.7
4

x 10

I.2 Comparison of linear and Non linear model

out

189

outlet air temperature Linear model compared to Non linear model: Feed flowrate step F

feed

Linear model: Ffeed =25

Nonlinear model
84

Temperature [ C]

82

80

78

76

74

72

1.1

1.2

1.3

1.4

1.5

1.6

Time [sec]

1.7
4

x 10

Figure I.8: Comparison of linear and Non linear model

Tout outlet air temperature Linear model compared to Non linear Feed flowrate step Ffeed
86

Linear model: Ffeed =10

Nonlinear model

84

Temperature [ C]

85

83

82

81

80

1.1

1.2

1.3

1.4

1.5

Time [sec]

Figure I.9: Comparison of linear and Non linear model

1.6

1.7
4

x 10

190

Main inlet air flow

Tout outlet air temperature Linear model compared to Non linear model: main air flowrate step Fmain
91

Linear model: F

main

=200

Nonlinear model
90

89

Temperature [oC]

I.2.2

Appendix I

88

87

86

85

1.1

1.2

1.3

1.4

1.5

Time [sec]

Figure I.10: Comparison of linear and Non linear model

1.6

1.7
4

x 10

I.2 Comparison of linear and Non linear model

out

191

outlet air temperature Linear model compared to Non linear model: main air flowrate step F

main

85.9

Temperature [oC]

85.8

85.7

85.6

85.5

85.4

Linear model: Fmain =20

Nonlinear model
85.3

1.1

1.2

1.3

1.4

1.5

Time [sec]

Figure I.11: Comparison of linear and Non linear model

1.6

1.7
4

x 10

192

Appendix I

outlet air temperature Linear model compared to Non linear model: main air flowrate step F

out
87

main

86.8

Temperature [oC]

86.6

86.4

86.2

86

85.8

85.6

Linear model: F

main

85.4

=50

Nonlinear model
1

1.1

1.2

1.3

1.4

1.5

Time [sec]

Figure I.12: Comparison of linear and Non linear model

1.6

1.7
4

x 10

I.2 Comparison of linear and Non linear model

I.2.3

193

Main inlet air temperature

Tout outlet air temperature Linear model compared to Non linear model: main air temperature step Tmain
86

Linear model: T

main

=10

Nonlinear model
85

Temperature [ C]

84

83

82

81

80

79
0.95

1.05

Time [sec]

Figure I.13: Comparison of linear and Non linear model: Main inlet air temperature

1.1
4

x 10

194

Appendix I

out

outlet air temperature Linear model compared to Non linear model: main air temperature step T

main

Linear model: Tmain =10

Linear model: Tmain =30
140

Linear model: T

=50

Linear model: T

=100

main
main

Temperature [ C]

130

120

110

100

90

10

12

14

Time [sec]

x 10

Figure I.14: Comparison of linear and Non linear model: Main inlet air temperature
Tout outlet air temperature Linear model compared to Non linear model: main air temperature step Tmain
98

96

Temperature [oC]

94

92

90

88

Linear model: Tmain =20

86

Nonlinear model
0.95

1.05

1.1

1.15

1.2

1.25

1.3

1.35

1.4

1.45

Time [sec]

Figure I.15: Comparison of linear and Non linear model: Main inlet air temperature

1.5
4

x 10

I.2 Comparison of linear and Non linear model

out

195

outlet air temperature Linear model compared to Non linear model: main air temperature step T
110

Temperature [oC]

105

100

95

90

Linear model: T

main

=40

Nonlinear model
85
0.95

1.05

1.1

1.15

1.2

1.25

1.3

1.35

1.4

1.45

Time [sec]

Figure I.16: Comparison of linear and Non linear model: Main inlet air temperature

1.5
4

x 10

196

Appendix I

I.2.4

Solids Content
Tout outlet air temperature Linear model compared to Non linear model: Solids content step Scont
92

91

Temperature [oC]

90

89

88

87

86

Linear model: Scont =0.1

Nonlinear model

85

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

Time [sec]

x 10

Figure I.17: Comparison of linear and Non linear model: Solids content

out

outlet air temperature Linear model compared to Non linear model: Solids content step S

cont

Temperature [ C]

105
100
95

Linear model: S

90

cont

=0.3

Nonlinear model
85

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

Time [sec]

2
4

x 10

Temperature [ C]

Zoomed in
105
104.5
104

Linear model: S

cont

103.5

=0.3

Nonlinear model
1.3

1.4

1.5

1.6

1.7

1.8

Time [sec]

Figure I.18: Comparison of linear and Non linear model: Solids content

1.9
4

x 10

I.2 Comparison of linear and Non linear model

I.2.5

197

Relative Humidity of Ambient air

T

out

outlet air temperature Linear model compared to Non linear model:ambient Relative humidity RH

amb

85.56

Temperature [oC]

85.54

85.52

85.5

85.48

85.46

Linear model: RHamb =0.1

Nonlinear model
85.44
1

1.1

1.2

1.3

1.4

1.5

1.6

1.7

1.8

1.9

Time [sec]

x 10

Figure I.19: Comparison of linear and Non linear model: Relative humidity of
ambient air

Tout outlet air temperature Linear model compared to Non linear model:ambient Relative humidity RHamb
Temperature [oC]

86
85.9
85.8
85.7
85.6

Linear model: RH

85.5

Nonlinear model

amb

1.1

1.2

1.3

1.4

1.5

1.6

1.7

=0.5

1.8

Time [sec]

x 10

Zoomed in
Temperature [oC]

86
85.9
85.8
85.7

Linear model: RHamb =0.5

85.6

Nonlinear model

85.5
1

1.002

1.004

1.006

1.008

1.01

1.012

Time [sec]

Figure I.20: Comparison of linear and Non linear model: Relative humidity of
ambient air

1.014
4

x 10

198

Appendix I

www.elektro.dtu.dk
Department of Electrical Engineering
Automation and Control
Technical University of Denmark
Elektrovej
Building 326
DK-2800 Kgs. Lyngby
Denmark
Tel:
(+45) 45 25 35 50
Fax:
(+45) 45 93 12 95