Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Cyber-Physical MBD For Multi-Physics Automotive Systems: Sujit S. Phatak DJ Mccune George Saikalis, Ph.D. Yasuo Sugure

Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

Hitachi Review Vol. 61 (2012), No.

183

Cyber-physical MBD for Multi-physics Automotive Systems


Sujit S. Phatak
DJ McCune
George Saikalis, Ph.D.
Yasuo Sugure

OVERVIEW: The need for CPSs derives from the complexity of modern
automotive embedded systems, which can contain more than 100 ECUs,
a large ROM capacity, and a large amount of software code. Another
issue is that traditional design methodologies find it difficult to cope when
fundamental hardware problems are identified late in the development
process, during the validation phase. The MBD approach is one possible
solution to these issues and it lays the foundation for CPSs.

INTRODUCTION
THE defining feature of cyber-physical systems (CPSs)
is their close integration between physical processes
and systems on one hand and computational systems
on the other(1). CPSs integrate the dynamics of physical
processes with those of the software and network,
providing abstractions for modeling and design as well
as analysis techniques suitable for integrated systems.
In contrast to traditional embedded systems, where the
emphasis is more on operation as a standalone device,
CPSs emphasize the network of interacting systems.
Similarly, while traditional embedded systems focus on
the computational components, CPSs primarily focus
on the interfaces between the computational elements
of the system. Their scope of application covers all
sensor-based control systems, embedded systems,
and autonomous systems. These cover a very wide
range, from device-based systems such as automotive
systems, entertainment and home appliances to
integration systems such as social infrastructure,
energy, freight and transportation, aeronautical and
space applications, and healthcare, and also technical
platforms such as manufacturing systems.
The mechatronic control systems that are typically
implemented in automotive applications, such as
engine control, transmission control, throttle control,
and braking, typically involve multiple complex
physical systems with dedicated embedded controllers
that communicate with each other via a vehicle
network, such as Controller Area Network (CAN)
or FlexRay. Model-based design (MBD) is adopted
as a way of making the design process for these
complex system more efficient(2). The system design
stage integrates models of physical system behavior
(also called plant models) with controller models
to produce an abstracted system implementation.
*1 MATLAB and Simulink are registered trademarks of The MathWorks, Inc.

The controller models can be implemented either


at the algorithm level, using popular tools such as
MATLAB*1/Simulink*1, or they can be implemented
at a lower abstraction level using virtual central
processing unit (CPU) modeling techniques.
Virtual CPU modeling(2) involves development of a
software model of the microcontroller hardware itself.
This microcontroller model can then be integrated with
the behavioral models of the plant (physical system)
so that realistic system performance measurement and
validation can be performed. This approach allows
concurrent development of the plant models and
control software applications, and also their validation,
including the realtime operating system (RTOS) and
device drivers.
Sometimes the system may consist of multiple
plant models (physical systems) representing different
components of the mechatronic system that need to be
implemented in different domains, and that need to
be connected to a controller (computational system)
via some interface. Together, these elements form a
mechatronic CPS.
This article reviews the basic approach to
implementing a CPS adopted at the Automotive
Products Research Laboratory of Hitachi America,
Ltd., and describes a CPS case study in the form of a
gasoline fuel pump that was implemented as a multidomain co-simulation platform(3).
NEED FOR CPS IN AUTOMOTIVE
APPLICATIONS
Challenges for Automotive Embedded Systems
The automotive embedded systems used in
different parts of modern vehicles (including the
powertrain, transmission, vehicle dynamics, and
infotainment) can be very complex, consisting of
more than 100 electronic control units (ECUs). These
ECUs are microcontroller-based embedded systems.

184

Hitachi Review Vol. 61 (2012), No. 5

Abstraction
level

Design

Concept

Component

Validation
Car
maker

Concept
design
System
design
Component
Implementation

Product

Abstraction
level
Concept

System
validation

Tier 1

Design
Car
maker

Behavioral/
mathematical model
MATLAB

Detail model
Component

Tier 2

Unit validation
Time

Fig. 1Traditional V-cycle Development Process.


The typical development cycle for automotive electronic control
units (ECUs), including the role of the respective contributors.

Typically, each sub-component of the physical system


has a dedicated ECU. Automotive network protocols
such as CAN and FlexRay are used to link these ECUs
in a network.
Modern automotive applications often require
more than 4Mbyte of read-only memory (ROM)
(program memory), which stores a dedicated
application program. The size of software code
is also increasing exponentially. It is estimated to
grow more than 1,000 times in the next 20 years. In
contrast, the capacity of the microcontroller hardware
(computational system) typically grows by only 20
times over 10 years. To cope with this gap between
increasing code size and hardware capacity, and
to improve cycle performance and reliability, it is
anticipated that multi-CPU or multi-core architectures
will be required in the near future.
Inefficiency in Traditional Design Methodology
Traditionally, automotive embedded controller
development has often followed the V-cycle shown
in Fig.1.
The V-cycle is broadly divided into two phases, the
design phase and the validation phase. The vertical
axis shows the level of abstraction while the horizontal
axis represents time. The design phase begins with
the concept design at a high level of abstraction. This
takes place at the car maker or original equipment
manufacturer (OEM). Typically, the process then
proceeds through progressively lower levels of
abstraction, to system design at a tier-one supplier and
then to the actual components and hardware supplied
by a tier-two supplier. This is the implementation
phase, and is followed by the validation phase. This
begins with unit validation, which happens at the same
level of abstraction as the design implementation. Next

Tier 1

Plant: MATLAB
Tier 2
Ctrl.: Micro./ECU model

Algorithm design
Behavior simulation
Rapid prototyping
ECU hardware
(CPU, ASIC, board)
BIOS software coding
RTOS configuration
Code generation

Computational Simulink based


system
Virtual CPU based

Time

MBD: model-based design CPS: cyber-physical system


ECU: electronic control unit CPU: central processing unit
ASIC: application-specific integrated circuit
BIOS: basic input/output system RTOS: realtime operating system

Fig. 2V-cycle Incorporating MBD and CPS Approach.


The role of MBD and the CPS approach in V-cycle development,
including the different activities involved.

is system validation, which represents a higher level


of abstraction, with the final product as the end result.
A problem with this traditional V-cycle is that,
if a fundamental hardware problem is not detected
until the validation phase, going back to the design
phase to identify its cause is not always time- and
cost-effective. It is important to identify any inherent
hardware problems early in the design phase, before
investing in hardware development. Also important
for business expansion and innovation is the research,
development, and testing of future advanced system
architectures (hardware and software).
MBD is one possible solution to these problems.
The MBD approach is also called model in the loop
simulation (MILS). It lays the foundation for CPS
development. Fig.2 shows the V-cycle when MBD
and CPS are incorporated.
The figure shows how use of MBD concentrates
activity in the design phase. This includes algorithm
design, behavior simulation, and rapid prototyping,
which are carried out during concept design. Initially,
signal flow or conserved system simulators like
MATLAB/Simulink or Synopsys*2 Saber*2 are used to
develop behavioral or mathematical models.
This is followed by component simulation at
a lower abstraction level. The main development
activities here are ECU hardware specification, basic
input/output system (BIOS) software coding, RTOS
configuration, and autocode generation. In terms of
MBD, this phase involves developing detailed models
*2 Synopsys is a registered trademark of Synopsys, Inc. and Saber is a
registered trademark of SabreMark Limited Partnership and is used
under license.

Cyber-physical MBD for Multi-physics Automotive Systems

of both the physical system (plant) and the ECUs and


control system (computational system).
The computational system can be built in Simulink
or using a virtual CPU. This gives two different
configurations for the CPS. Typically, when using
MBD, the next step after developing a Simulinkbased computational system is to perform autocodegeneration and implement the generated software on
a real hardware ECU. This is called processor in the
loop simulation (PILS). The work described here,
however, adopts a new approach whereby a simulation
of the processor (CPU) is also included in the model.
This approach could be called virtual PILS, although
the term virtual CPU modeling is used here for
consistency. The following section gives a detailed
description of CPS using both the Simulink and virtual
CPU methods.
CPS FOR AUTOMOTIVE APPLICATIONS
Simulink-based CPS
Fig.3 shows a simplified block diagram of a
Simulink-based CPS.
The system consists of two main parts: the plant
model (physical system) and the control / soft ECU
model (computational system). The plant models
model the physical systems of the vehicle, such as
the engine for a gasoline car or the electric motor of
a hybrid electric vehicle (HEV) or electric vehicle
(EV). The control models implement the logic used
to control the plant models in such a way that they
achieve the desired functionality. These models also
provide visualization of the control parameters.
Simulink is used for simulation, and communication
between the plant and the control models is in terms
of physical quantities or parameters represented in
engineering units (Pascals for pressure or Amperes for
current). The system uses feedback control whereby
the user/driver inputs are applied to the plant model
together with the actuator inputs determined by the
control algorithm running on the control model. An
important point to note here is that the link between
the plant and control models is handled within the
simulation system via a direct connection between
the models.
In the case when different models running on
different simulators are used, a co-simulation needs
to be implemented to provide the connection between
the simulators. Co-simulation can be achieved by
a direct connection between the models if their
respective simulators support such a function, or it
can be achieved using a co-simulation bus(4). The

185

Feedback

User/
driver
inputs

Plant models
from OEMs
Gasoline engine
HEV/EV system
Other physical systems

Engineering
units

Control/
soft ECU
model

MATLAB/
Simulink
Visualization/analysis

OEM: original equipment manufacturer HEV: hybrid electric vehicle


EV: electric vehicle

Fig. 3Simulink-based CPS.


This simplified block diagram shows the CPS simulation model.

physical location of models may or may not be on


the same personal computer (PC) in the case of a cosimulation bus implementation. Co-simulation also
requires that the simulators support its use, or that it
can be implemented using a high level programming
language such as C/C++.
Use of a co-simulation bus offers several other
benefits. (1) It enables a multi-domain/multi-physics
system simulation where models stay in their native
domain but work together. (2) The multi-domain
interaction yields a robust design methodology. (3)The
co-simulation bus also provides a framework for
implementing the interface between the computational
and physical systems of the CPS.
Virtual-CPU-based CPS
Fig.4 shows a simplified block diagram of a
virtual-CPU-based CPS.
The system architecture for a CPS based on a
virtual CPU is an extension of the Simulink-based
CPS architecture. In this case, the control model is
implemented using a virtual CPU simulator, which
can emulate a microcontroller model or virtual CPU
model for the specific microcontroller specified by
the user. Microcontroller models are available for the
main automotive semiconductor suppliers, including
Freescale Semiconductor, Inc., Renesas Electronics
Corporation, and Infineon Technologies AG.
These models emulate the microcontroller
hardware and can execute the same object code as
the actual hardware. This simulator also gives access
to internal information about the microcontroller
hardware, such as the program counter, instruction
cycles, interrupts, and software states.
As in a real microcontroller, these models process
analog or digital voltage signals and cannot work
with engineering units directly. Hence an additional
sensor model (sensor models in Fig.4) needs to be

Hitachi Review Vol. 61 (2012), No. 5

User/
driver
inputs

Engineering
units

Plant models
from OEMs
Gasoline engine
HEV/EV system
Other physical systems

Sensor models
Analog I/O
Digital I/O
Signal
conditioning

MATLAB/Simulink

Co-simulation

Feedback

Virtual CPU
simulator
Virtual CPU model
MPC/SH-2A/...

Voltage
signals Debugging Visualization of CPU
capability Program counter
Instruction cycles
Interrupts
Other...

Co-simulation
Direct one-to-one
Co-simulation bus

I/O: input/output

implemented to act as an interface and perform the


required signal conditioning between the plant model
and the virtual CPU model. In other words, this block
handles analog and digital input and output (I/O) and
converts engineering units into voltage signals, which
can be processed by the virtual CPU model. The rest
of the system architecture remains the same, including
the feedback loop and the user/driver inputs.
Another important point to note is the use of cosimulation. The plant models and sensor models are
typically implemented using MATLAB/Simulink
while the virtual CPU is implemented using a
dedicated virtual CPU simulator as described above.
Therefore, a co-simulation is required between these
simulators and can be achieved either via a direct oneto-one connection or by using the co-simulation bus.
While a direct one-to-one connection is available for
systems that run entirely within Simulink, this is not
supported for some other simulators, in which case
the co-simulation bus approach must be used. This is
described in detail in the case study in the next section.
AUTOMOTIVE CPS CASE STUDY:
GASOLINE FUEL PUMP SIMULATION
A gasoline fuel pump simulation model developed
by Hitachi was used to investigate CPS implementation
by comparing the results obtained using the Simulinkbased and virtual-CPU-based approaches respectively.
Simulink-based CPS for Gasoline Fuel Pump
A physical system model (plant model) was
implemented using a co-simulation bus (see Fig.5).
The physical system of the gasoline fuel pump
consists of three parts: a driver circuit simulated
using an electromechanical simulator, and separate
pump and inlet valve models, which are simulated
using the hydraulics simulator. Using a co-simulation
bus, the driver circuit was implemented in the
electromechanical simulator running on one PC and

186

Fig. 4Virtual-CPU-based CPS.


The simplified block diagram
of the CPS shows how it now
includes a virtual CPU model
instead of the traditional
Simulink-based control model.

Implementation

System
Gasoline fuel
pump
PWM

Simulator
Pump
model

Simulator
Driver
circuit

Gasoline fuel pump

I/O
signals

Response
pressure

Simulator

Inlet
valve

Simulator

I/O signals

Co-simulation bus
PWM: pulse width modulation

Fig. 5Gasoline Fuel Pump Physical System.


This block diagram of the gasoline fuel pump system and its
simulation model represents a multi-PC implementation with a
co-simulation bus.

Battery voltage
Target pressure

MATLAB/Simulink
I/O signals
Feedback

Co-simulation bus

Fig. 6Gasoline Fuel Pump Computational System.


The simulation model of the gasoline fuel pump control system
interfaces to a co-simulation bus.

the pump and inlet valve models of the actual pump


were implemented in a hydraulic simulator running
on a second PC.
The computational system (control model) was
implemented using Simulink (see Fig.6). Fig.7 shows
the complete integrated Simulink-based CPS for the
gasoline fuel pump.
Proportional-integral (PI) control was used to
generate the controller output based on the difference
between the target pressure and the pressure feedback

Cyber-physical MBD for Multi-physics Automotive Systems

Gasoline fuel pump


Plant
model

Control model

Simulator

MATLAB/
Simulink

Simulator

I/O signals

Target
pressure

I/O signals
Feedback

Virtual-CPUbased implementation

Simulink
implementation

Battery
voltage

Battery
voltage
Target
pressure

Virtual CPU
Device drivers,
RTOS

MATLAB/
Simulink

C auto
code

I/O signals

at a constant battery voltage input level. Since the


final goal was to implement this control algorithm
on hardware (virtual in this case), and as triggering
the solenoid in response to the controller output costs
energy, pulse width modulation (PWM) duty-cycle
control was added to the original control. While
the solenoid had to be fully triggered to open the
inlet valve, less force was required for holding it
open and therefore the current through the solenoid
could be reduced. The percentage reduction in this
solenoid current was determined by the PWM duty
factor. Based on several tests and on experience, data
were gathered on the relationship between the duty
factor for various start angles and the revolutions per
minute (RPM) and battery voltage. From these were
produced a look-up table that could be used to obtain
the duty-cycle value(1). The PWM based control also
enabled the use of the virtual CPU based approach for
implementing this CPS.
Virtual-CPU-based CPS for Gasoline Fuel Pump
The first requirement for the virtual-CPU-based
CPS was to obtain the object code for the software
control algorithm used in the Simulink CPS. This was
achieved using auto-code generation to generate the
application layer of the software from the Simulink
control model. The virtual CPU used was a Renesas
Electronics Corporations SH-2A*3 microcontroller.
Other development included hand coding the device
driver software for the SH-2A and configuring the
RTOS to be used for this application (see Fig.8).
The virtual CPU simulation itself combined two
simulators: the virtual CPU simulator and MATLAB/
Simulink. Accordingly, the virtual CPU simulation
used a direct one-to-one co-simulation between these
two simulators. In terms of the overall system, this cosimulation bus approach involved using the MATLAB/
*3 SH-2A is a trademark or registered trademark of Renesas Electronics
Corporation.

Target
pressure

Virtual CPU
simulator
MATLAB/
Simulink

Feedback

Co-simulation bus
Fig. 7Simulink-based CPS for Gasoline Fuel Pump.
This Simulink-based CPS uses a co-simulation bus.

SH-2A

I/O signals

Feedback

Co-simulation bus

187

Co-simulation bus

Fig. 8Auto-code Generation for SH-2A CPU.


The Simulink-based control model was converted into C code,
compiled with the device drivers and RTOS, and executed on the
Renesas Electronics Corporations SH-2A CPU.

Virtual CPU
Gasoline fuel pump
SH-2A
Target
pressure

Virtual CPU
simulator
MATLAB/
Simulink

I/O signals

Simulator

Simulator

I/O signals

Feedback

Co-simulation bus
Fig. 9Virtual CPU-based CPS for Gasoline Fuel Pump.
The virtual SH-2A control model integrates with the plant model
of the gasoline fuel pump via a co-simulation bus.

Simulink portion of the virtual CPU as an interface to


the co-simulation bus since the virtual CPU simulator
did not directly support the co-simulation bus. Fig.9
shows the complete system using the virtual-CPUbased CPS for the gasoline fuel pump.
This system enabled the virtual tuning of the PI
control algorithm, which will help accelerate overall
pump controller development in the future.
Correlation between Simulink-based CPS and
Virtual-CPU-based CPS
After adopting both the Simulink-based and virtualCPU-based approaches for CPS implementation, the
next step was to observe the relationship between the
behaviors of the pump simulation using these two
approaches under identical simulation conditions and
configuration (see Fig.10).
The comparison shows a very good correlation
between the Simulink-based and virtual-CPU-based
approaches for the gasoline fuel pump control system
under identical simulation configurations (same target
pressure at different engine RPMs).

250

250

200

200

150
100

Simulink
Virtual CPU simulator

50

Pressure (MPa)

Pressure (MPa)

Hitachi Review Vol. 61 (2012), No. 5

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5

Simulink
Virtual CPU simulator

150
100
50

0
0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5

Simulation time (s)

Simulation time (s)

Target pressure: 20 MPa


750 r/m (idle)

Target pressure: 20 MPa


1,000 r/m

FURTHER ACHIEVEMENTS USING CPS


After the successful implementation of the CPS for
a gasoline fuel pump control system at the Automotive
Products Research Laboratory of Hitachi America,
Ltd., a decision was made to experiment with the
flexibility offered by the co-simulation bus approach
for a multi-location co-simulation. The trade-off
when a multi-location CPS implementation was
tested was that the analysis took longer due to the
slower data transfer time. The problem in this case
was slower speed due to latency on the Ethernet-based
Transmission Control Protocol/Internet Protocol
(TCP/IP) network. At a sampling rate of 100 s, for
example, the co-simulation across offices in Japan and
the USA was three times slower than the equivalent
co-simulation run in the USA (see Fig.11).
As shown in the figure, the gasoline fuel pump
physical model was implemented on two PCs, one at
the Japan office and one at the USA office, while the
computational system model was implemented on a
PC at the USA office. Apart from the problem with the
data transfer time, this demonstration showed that a
co-simulation could span two offices, one in Japan and
the other in the USA, and proved that it was possible to
achieve a flexible CPS implementation across offices
in different countries.
CONCLUSIONs
This article has reviewed the basic approach to
implementing a CPS adopted at the Automotive
Products Research Laboratory of Hitachi America,
Ltd., and described a CPS case study in the form of a
gasoline fuel pump that was implemented as a multidomain co-simulation platform.
An automotive CPS involving a gasoline fuel
pump was implemented to enable virtual tuning of
the PI control algorithm and accelerate the overall
development process. The system behavior was first

188

Fig. 10Correlation between


Simulink-based and VirtualCPU-based CPS for Gasoline
Fuel Pump.
This comparison shows the
proportional integral (PI)
control profiles for the gasoline
fuel pump for two target
pressure settings.

Dual-location CPS implementation at sites in Japan and the USA


Co-simulation bus

USA office

Japan office

PC-2 @ Japan office


PC-3 @
USA Gasoline fuel pump
office
Battery
voltage
MATLAB/
Simulator
Simulator
Simulink
Target
pressure
I/O signals

PC-1 @ USA office

Sampling rate: 100 s


Speed: 3 times slower
than local co-simulation

I/O signals
Feedback

Co-simulation bus
Multiple PC/Multiple simulator/
Multiple location implementation

PC-1

PC-2

PC-3

Co-simulation bus

PC-1: MATLAB/Simulink @ USA office


PC-2: Hydraulic simulator @ Japan office
PC-3: Electromechanical simulator @ USA office

Fig. 11International Multi-location CPS.


This distributed implementation of the gasoline fuel pump
control operated a co-simulation bus between two Hitachi
offices in USA and Japan.

simulated using Simulink and later using a virtualCPU-based approach. Both approaches achieved a
good correlation. An international multi-location CPS
implementation was also tested and validated to prove
the flexible multi-location implementation of these
systems as a multi-physics robust design methodology.

REFERENCES
(1) S. Phatak et al., Cyber Physical System: A Virtual CPU
Based Mechatronic Simulation, IFAC (Sep. 2010).
(2) G. Saikalis et al., Virtual Embedded Mechatronics System,
SAE Technical Paper # 2006-01-0861.
(3) J. Sinnamon et al., Co-simulation Analysis of Transient
Response and Control of Engines with Variable Valvetrains,
SAE Technical Paper, 2007-01-1283.
(4) ChiasTek, 2009, http://www.chiastek.com

Cyber-physical MBD for Multi-physics Automotive Systems

189

ABOUT THE AUTHORS


Sujit S. Phatak

DJ McCune

Joined Hitachi America, Ltd. in 2007, and is a


Researcher at the Research & Development Division,
Automotive Products Research Laboratory. He is
currently involved in model based development for
embedded systems and virtual prototyping systems.
Mr. Phatak is a member of the IEEE and the Society of
Automotive Engineers (SAE).

Joined Hitachi America, Ltd. in 2003, and now works


at the Automotive Products Research Laboratory. He
is currently engaged in modeling and simulation of
cyber physical systems. Mr. McCune is a member of
the SAE.

Yasuo Sugure
George Saikalis, Ph.D.
Joined Hitachi America, Ltd. in 1990, and now works
at the Research & Development Division, Automotive
Products Research Laboratory. He is currently Vice
President engaged in many aspects of automotive
R&D. Dr. Saikalis is a member of the IEEE and SAE.

Joined Hitachi, Ltd. in 1999, and now works at the


Central Research Laboratory. He is currently engaged
in virtual prototyping systems using microcontroller
models for embedded control systems. Mr. Sugure
is a member of the SAE and Institute of Electronics,
Information and Communication Engineers (IEICE).

You might also like