G Model

CIRP-1596; No. of Pages 4

CIRP Annals - Manufacturing Technology xxx (2016) xxx–xxx

Contents lists available at ScienceDirect

CIRP Annals - Manufacturing Technology

jou rnal homep age : ht t p: // ees .e lse vi er . com /ci r p/ def a ult . asp

Toward a Digital Twin for real-time geometry assurance in

individualized production
Rikard Söderberg (2)a,*, Kristina Wärmefjord a, Johan S. Carlson b, Lars Lindkvist a
a
Chalmers University of Technology, Department of Product and Production Development, SE 412-96 Gothenburg, Sweden
b
Fraunhofer-Chalmers Centre for Industrial Mathematics, SE-412 88 Gothenburg, Sweden

A R T I C L E I N F O A B S T R A C T

Keywords: Simulations of products and production processes are extensively used in the engineering phase. To
Assembly
secure good geometrical quality in the final product, tolerances, locator positions, clamping strategies,
Simulation
welding sequence, etc. are optimized during design and pre-production. Faster optimization algorithms,
Quality assurance
increased computer power and amount of available data, can leverage the area of simulation toward real-
time control and optimization of products and production systems – a concept often referred to as a
Digital Twin. This paper specifies and highlights functionality and data models necessary for real-time
geometry assurance and how this concept allows moving from mass production to more individualized
production.
ß 2017 Published by Elsevier Ltd on behalf of CIRP.

1. Introduction

A highly automated production factory for complex assembled

products is a huge investment and return on investment requires
high product quality, factory throughput, equipment utilization,
and flexibility as well as low energy consumption. Geometry
related problems, resulting in late changes and delays, usually
constitute a significant part of the total cost for poor quality.

1.1. Geometry assurance Fig. 1. The geometry assurance process.

Geometry assurance can be described as set of activities that

contributes to minimizing the effect of geometrical variation in the Most companies today are fully aware of the fact that a change
final product. Activities take place in all phases of the product is costlier in production than in the design phase. An effective
realization loop, see Fig. 1. digital geometry assurances process has the potential to drastically
The design phase: Here, concepts are analyzed and optimized to reduced costs and adjustments in production [1].
withstand manufacturing variation. Product requirements are
defined and decomposed into locator positions and tolerances on 1.2. Digital Twins
pars and subassemblies.
The pre-production phase: Here, the product and the production The area of virtual/digital development of products and
system are verified physically. Adjustments are made to correct production systems has grown extensively the last 20 years.
initial errors and prepare for full production. Inspection prepara- Simulation and optimization are today used for a variety of
tion and off-line programming of coordinate measurement different products and development tasks. Simulation has been an
machines and scanners are performed and all inspection strategies important tool for shifting expensive product changes, often
and inspection routines are decided. discovered during production start, to earlier design phases where
The production phase: Here, all initial production process cost for change is low. Increased number of model programs with
adjustments are completed and the product is running in full shorter intervals drive the needs for simulation. The ability to
production. Inspection data from parts and subassemblies are used simulate production ramp-up therefore becomes increasingly
to control production and to detect and correct errors. important [2].
Increased computer power, faster algorithms and more efficient
* Corresponding author.
optimization routines have made simulation and optimization an
E-mail address: rikard.soderberg@chalmers.se (R. Söderberg). everyday tool for engineers. Calculation time has gone from weeks

http://dx.doi.org/10.1016/j.cirp.2017.04.038
0007-8506/ß 2017 Published by Elsevier Ltd on behalf of CIRP.

Please cite this article in press as: Söderberg R, et al. Toward a Digital Twin for real-time geometry assurance in individualized
production. CIRP Annals - Manufacturing Technology (2017), http://dx.doi.org/10.1016/j.cirp.2017.04.038
G Model
CIRP-1596; No. of Pages 4

2 R. Söderberg et al. / CIRP Annals - Manufacturing Technology xxx (2016) xxx–xxx

to hours and minutes which have made it possible to, not only 2.1. Functionality: locating scheme optimization
verify solutions, but also to explore the solution space, searching
for the global optimum. Increased focus on sustainability, with Locating scheme optimization is a critical step in geometry
reduced waste, is also a driver for a more global view on assurance. Variation propagates through the physical contacts
optimization of design and manufacture [3]. (locating schemes) between parts and fixtures in an assembly.
Traditionally, simulation has been used in the design phase Locating schemes can be seen as representing the transfer
with estimated or historical data as input. Increased use of functions between input and output variation and controls the
sensors and in-line measuring equipment are making it possible robustness of a mechanical assembly. The stability analysis
to use (and reuse) simulation models created during product evaluates the geometrical robustness of a concept and shows
development also in production, now with real data as input. how variation, introduced by the locators and the additional
This will allow for adjustment of machine settings for the next clamps, propagates and affects critical features and dimensions
product in line based on simulations in the virtual world before [11]. Fig. 2 shows variation for two different locating schemes
the physical changeover, reducing machine setup times and (Concept 1 and Concept 2) used for joining before and after spot
increasing quality. The ability to link large amounts of data to welding and release from the fixture. Only the master location
fast simulation makes it possible to perform real-time optimi- schemes are shown. Additional clamp locations can be included in
zation of products and production processes. The concept of the same way.
using a digital copy of the physical system to perform real-time
optimization is often referred to as a Digital Twin. The concept 2.2. Functionality: statistical variation simulation
of a Digital Twin was adopted by NASA for safety and reliability
optimizations in [4] and [5]. With an aggressive push toward Variation simulation can be performed by utilizing transforma-
‘‘Internet of Things’’, data has become more accessible and tion matrices to calculate how part variation propagates in the
ubiquitous which necessitates the right approach and tools assembly. The method is often combined with Monte Carlo (MC)
to convert data into useful, actionable information [6]. The simulation. For non-rigid variation simulation, finite element
vision of the Digital Twin itself refers to a comprehensive analysis (FEA) is included. The method allows over-constrained
physical and functional description of a component, product or locating schemes that result in bending during assembly due to
system, which includes more or less all information which could variation in parts and fixtures. The method of influence coefficient
be useful in the current and subsequent lifecycle phases (MIC) [12] is used to reduce computational time. Contact modeling
[7,8]. Simulation and seamless transfer of data from one life prevent penetration due to bending and deformation during
cycle phase to the subsequent phase are central for the concept assembly, see [13] and [14]. Fig. 2 shows non-rigid variation
of the Digital Twin. simulation in the software RD&T, where the color coding indicates
The digital development advancements allow sensors, the expected variation.
machines, workpieces, and IT systems to be connected along the Lately, new materials with tougher requirements on bending,
value chain beyond a single enterprise. These connected systems deformation and stress are introduced in aerospace and automo-
(also referred to as cyber-physical systems) can interact with one tive industry. Variation simulation to verify geometrical deforma-
another using standard Internet-based protocols [9] and analyze tion and stress criteria and to support shimming strategies are
data to predict failure, configure themselves, and adapt to changes. treated in [15]. A future challenge is how variation simulation can
Increased availability of data will also open up new possibilities for be performed fast enough for in-line use.
better maintenance and related service systems [10].

1.3. Scope of the paper

This paper proposes the concept of a Digital Twin for geometry

assurance. The paper combines research within variation simula-
tion and quality control to an autonomous self-adjusting system
that optimizes quality and allows for individual production. The
Digital Twin is developed and used for product and production
system design in the concept phase and later on inherited for
inspection preparation and process control. Functionality and
information needed in each phase/step are specified. How the
concept of the Digital Twin allows moving from mass production to
more individualized production is discussed, as well as future
research challenges.

2. The Digital Twin in the design phase

In the design phase, different product concepts are developed

and optimized to withstand the effect of manufacturing variation.
From a geometry assurance perspective, three basic activities are
performed:

 Specification of product requirements/tolerances.

 Specification of locating schemes. Fig. 2. Variation simulation of a two-part assembly.

 Specification of part tolerances.

3. The Digital Twin in the pre-production phase
A Digital Twin, supporting robustness and tolerance analysis in
the design phase, uses geometry representations of the parts, In the pre-production phase, the Digital Twin is used as a basis
kinematic relations (locating schemes and transformation matrix- for inspection preparation and off-line programming (OLP) of
es) in combination with FEA to perform sensitivity and variation coordinate measure machines (CMMs) and scanners. It also
analysis. The Digital Twin is fed with variation data for parts and contains the definition of the final inspection points and a link
fixtures, normally gained from similar earlier projects. to the inspection database.

Please cite this article in press as: Söderberg R, et al. Toward a Digital Twin for real-time geometry assurance in individualized
production. CIRP Annals - Manufacturing Technology (2017), http://dx.doi.org/10.1016/j.cirp.2017.04.038
G Model
CIRP-1596; No. of Pages 4

R. Söderberg et al. / CIRP Annals - Manufacturing Technology xxx (2016) xxx–xxx 3

3.1. Functionality: inspection preparation sequences. As can be seen, one sequence results in quite large
deviation (middle) while the other does not (right). In a sense, the
Based on the Digital Twin geometries and functionality, latter can therefore be seen as the more robust one. Joining
inspection features and inspection programs can be generated. sequence optimization is treated in [21]. The in-plane position
Stability analysis and variation simulation can give valuable variation of the welding gun is treated in [22].
information about where to measure. Automatic collision free path For many non-rigid assemblies, the force needed to close the
planning can determine how to measure. This technology allows gap between the parts is critical. For sheet metal assemblies, this
new inspection programs to be generated, or existing programs to may affect the size of welding gun needed and for assemblies with
be updated, in a time efficient way [16]. plastic parts it affects the size and type of clips used to join the
A fully automatic in-line inspection system was recently parts [23]. For joining of dissimilar material, temperature is also
reported in [17]. The computational time to create a point-cloud quite important to consider [24]. Joining sequence optimization is
from the captured images is about 30 s and the computational time dependent on real-time part geometry data.
for the comparison with the CAD-model is about ten milliseconds.
This technology seems promising for in-line inspection and may 4.3. Functionality: root cause analysis (RCA)
well be combined with real-time process adjustment using the
Digital Twin approach. The assembly process for a complex product such as an
Cluster analysis can be used to reduce the number of inspection automotive body is realized by a large number of fixtures in several
points from an initial large set of points to a smaller set of stations. Fixture related geometrical errors, may in a complex
inspection points used in full production [18,19]. A future assembly be difficult to identify. A number of fixture errors may
challenge is how scanning can be performed fast enough for in- occur that leads to similar deviations in the final assembly. A
line use. general approach and a tool for RCA that allows individual station,
fixture and locator errors to be identified were proposed in
4. The Digital Twin in the production phase [25]. The tool translates variation and deviation in geometry data
to actions for adjustable process parameters. To separate variation
In the production phase all production process adjustments are caused by fixtures and the variation caused by previous and other
completed and the product is in full production. In this phase, the manufacturing processes, a multivariate fixture failure subspace
virtual assembly model (variation simulation model) is used control chart is proposed [26].
together with inspection data to control production and to detect During production, the Digital Twin is fed with inspection data
and correct errors. Future possibilities to fast capture a large from the final product. This is used to analyze if product errors
amount of data allow for in-line inspection, analysis and control of originate from assembly fixtures and decide what fixture and what
batches or even individuals, adding a new dimension to mass locators that have generated the error.
production.
5. The Digital Twin – a sheet metal assembly example
4.1. Functionality: virtual trimming
In a highly automated production factory for complex assem-
During assembly of newly produced components, form errors bled products there could be up to several hundreds of robots
can cause either functional or esthetical problems. Instead of organized into lines and stations for handling and joining
compensating the tools, which is quite expensive, this error is operations. Geometry related problems, resulting in late changes
compensated for by adjusting the locators. This is also known as and delays, usually constitute a significant part of the total cost for
trimming and is traditionally done manually and iteratively. A poor quality.
method for virtual trimming was proposed in [20]. Based on To scan and analyze inspection data of parts and subassemblies
inspection data from initial components and the variation fast and in real-time allow for new possibilities to adjust the
simulation model (the Digital Twin), all trimming activities are process and the equipment to compensate for geometrical
performed in the computer tool presented. After the locators are deviations of incoming parts. In Fig. 4, a concept for an autonomous
adjusted, the result is presented directly, which eliminates the self-optimizing robot welding station is proposed.
need for physical inspection in order to verify the result of the
trimming. The tool also includes optimization of the trimming.
Efficient use is based on real-time data as input.

4.2. Functionality: joining sequence optimization

For non-rigid parts, the joining sequence is crucial for how

variation in the individual parts, fixtures and welding equipment
will affect the final assembly. Fig. 3 shows an example where the
same two parts, with the same fixture (Concept 2 in Fig. 2), are
joined together with seven weld points using two different Fig. 4. Self-compensating assembly line.

The proposed concept is based on the idea of a Digital Twin,

inherited from the design phase, containing all nominal geometries
for the parts (FEA meshes) and the equipment (fixtures, robots and
welding guns). The Digital Twin also contains mating conditions
(and contacts modeling) as well as functionality for locating
scheme optimization (Virtual Trimming), variation simulation, and
welding sequence optimization. This basic functionality is created
and used during the design phase to define tolerances and locator/
clamp positions, based on assumed variation, and is now inherited
and used during production to fine-tune and adjust the process to
the real geometry of incoming parts. The Digital Twin has an
interface to the inspection database that contains information
Fig. 3. The effect of joining sequence. about individual part geometries as well as statistical distributions

Please cite this article in press as: Söderberg R, et al. Toward a Digital Twin for real-time geometry assurance in individualized
production. CIRP Annals - Manufacturing Technology (2017), http://dx.doi.org/10.1016/j.cirp.2017.04.038
G Model
CIRP-1596; No. of Pages 4

4 R. Söderberg et al. / CIRP Annals - Manufacturing Technology xxx (2016) xxx–xxx

for batches of parts. The concept of the self-compensating Acknowledgement

assembly line is built up by the following steps (see also Fig. 4):
The work was carried out in collaboration within Wingquist
1. Individual parts A and B are scanned and data are stored as Laboratory and the Area of Advance Production at Chalmers within
deviations from nominal in all nodes of the meshes. the project Smart Assembly 4.0, financed by The Swedish Founda-
2. Parts are sorted in classes to allow for pairwise matching. tion for Strategic Research. The support is gratefully acknowledged.
3. Based on scan data for part A and part B, pairwise matching is
conducted. The matching criteria is to minimize deviation from
nominal after welding. References
4. Based on scan data for part A and part B, deviation after joining is
[1] Söderberg R, Lindkvist L, Wärmefjord K, Carlson JS (2016) Virtual Geometry
further minimized by adjusting the fixture elements (locating Assurance Process and Toolbox. Procedia CIRP 43:3–12.
schemes for the two parts). [2] Klocke F, Stauder J, Mattfeld P, Müller J (2016) Modeling of Manufacturing
5. Based on scan data for part A and part B, the optimal welding Technologies During Ramp-up. Procedia CIRP 51:122–127.
[3] Le Duigou J, Gulbrandsen-Dahl S, Vallet F, Söderberg R, Eynard B, Perry N
sequence is selected – also here the goal is to minimize deviation (2016) Optimization and Lifecycle Engineering for Design and Manufacture of
after welding. Recycled Aluminium Parts. CIRP Annals – Manufacturing Technology 65(1):149–
6. Subassembly A/B is inspected and compared to simulated 152.
[4] Tuegel EJ, Ingraffea AR, Eason TG, Spottswood SM (2011) Reengineering
geometry (including steps 2–4). Deviations between simulated Aircraft Structural Life Prediction Using a Digital Twin. International Journal
and inspected geometry is related to fixtures and locators using of Aerospace Engineering 2011:14.
Root Cause Analysis (AI and machine learning can also be used [5] Glaessgen EH, Stargel D (2012) The Digital Twin Paradigm for Future NASA and
US Air Force Vehicles. 53rd Struct. Dyn. Mater. Conf. Special Session: Digital Twin,
for capturing non-modeled effects).
Honolulu, HI, US, 1–14.
[6] Lee J, Lapira E, Bagheri B, Kao H-A (2013) Recent Advances and Trends in
Adjustments in the process, to compensate for part deviations, Predictive Manufacturing Systems in Big Data Environment. Manufacturing
Letters 1(1):38–41.
can be made with respect to statistical parameters (mean value
[7] Boschert S, Rosen R (2016) Digital Twin—The Simulation Aspect. in Hehen-
and standard deviation) to optimize the process batch wise, or for berger P, Bradley D, (Eds.) Mechatronic Futures: Challenges and Solutions for
individuals. The latter assumes very fast scanning and optimiza- Mechatronic Systems and Their Designers, Springer International Publishing,
tion routines for optimizing locator positions and welding Cham59–74.
[8] Rosen R, von Wichert G, Lo G, Bettenhausen KD (2015) About the Importance of
sequence which is a research challenge in itself. Due to need for Autonomy and Digital Twins for the Future of Manufacturing. IFAC-Papers
calculation speed, minimizing the gap between the flanges to be OnLine 48(3):567–572.
welded may be a strategy to reduce deviation. Rapid deployment of [9] Khaitan SK, McCalley JD (2015) Design Techniques and Applications of Cyber-
physical Systems: A Survey. IEEE Systems Journal 9(2):350–365.
remote laser welding to secure the quality of the weld joint was [10] Roy R, Stark R, Tracht K, Takata S, Mori M (2016) Continuous Maintenance and
reported in [27]. the Future – Foundations and Technological Challenges. CIRP Annals –
Manufacturing Technology 65(2):667–688.
[11] Söderberg R, Lindkvist L (1999) Computer Aided Assembly Robustness Evalu-
6. Conclusion and discussion ation. Journal of Engineering Design 10(2):165–181.
[12] Liu SC, Hu SJ (1997) Variation Simulation for Deformable Sheet Metal Assem-
A highly automated assembly line is a huge investment. Return blies Using Finite Element Methods. Journal of Manufacturing Science and
Engineering 119:368–374.
on investment requires high product quality, factory throughput,
[13] Wärmefjord K, Söderberg R, Lindkvist L (2008) Tolerance Simulation of
equipment utilization, and flexibility as well as low energy Compliant Sheet Metal Assemblies Using Automatic Node-based Contact
consumption. Today, geometry related problems, resulting in late Detection. Proceedings of ASME IMECE.
[14] Lindau B, Lorin S, Lindkvist L, Söderberg R (2016) Efficient Contact Modeling in
changes and delays usually constitute a significant part of the total
Nonrigid Variation Simulation. Journal of Computing and Information Science in
cost for poor quality. Engineering 16(1):011002.
Therefore, a Digital Twin for geometry assurance is proposed. [15] Söderberg R, Wärmefjord K, Lindkvist L (2015) Variation Simulation of Stress
The Digital Twin contains geometry representation of the During Assembly of Composite Parts. CIRP Annals – Manufacturing Technology
64(1):17–20.
assembly, kinematic relations, FEA functionality, Monte Carlo [16] Salman R, Carlson JS, Ekstedt F, Spensieri D, Torstensson J, Söderberg R (2016)
simulation, material properties and link to inspection data base. An Industrially Validated CMM Inspection Process With Sequence Constraints.
The Digital Twin is created and used in the design phase to develop Procedia CIRP 138–143.
[17] Bergström P, Fergusson M, Folkesson P, Runnemalm A, Ottosson M, Andersson
robust products and to distribute tolerances, but also inherited A, Sjödahl M (2016) Automatic In-line Inspection of Shape Based on Photo-
later on in the production phase to serve as a real-time controller grammetry. 7th International Swedish Production Symposium, SPS16.
for the assembly system. In the production phase the Digital Twin [18] Wärmefjord K, Carlson JS, Söderberg R (2009) A Measure of the Information
Loss for Inspection Point Reduction. Journal of Manufacturing Science and
is fed with scan data from the part geometries and uses this to Engineering 131(5). 051017.
match parts, to adjust locator and clamp positions and to select the [19] Wärmefjord K, Carlson JS, Söderberg R (2010) An Investigation of the Effect of
optimal joining sequence (see Fig. 5). Sample Size on Geometrical Inspection Point Reduction Using Cluster Analysis.
CIRP Journal of Manufacturing Science and Technology 3(3):227–235.
[20] Lindkvist L, Carlson JS, Söderberg R (2005) Virtual Locator Trimming in Pre-
Production: Rigid and Non-Rigid Analysis. ASME IMECE 561–568.
[21] Wärmefjord K, Söderberg R, Lindau B, Lindkvist L, Lorin S (2016) Joining in
Nonrigid Variation Simulation. in Udroiu R, (Ed.) Computer-aided Technologies
– Applications in Engineering and Medicine, InTech, Rijeka, Croatia.
[22] Söderberg R, Wärmefjord K, Lindkvist L, Berlin R (2012) The Influence of Spot
Weld Position Variation on Geometrical Quality. CIRP Annals – Manufacturing
Technology 61(1):13–16.
[23] Wärmefjord K, Söderberg R, Lindkvist L (2013) Simulation of the Effect of
Geometrical Variation on Assembly and Holding Forces. International Journal of
Product Development 18(1):88–108.
[24] Lorin S, Lindkvist L, Söderberg R, Sandboge R (2012) Combining Variation
Fig. 5. Digital Twin input and functionality. Simulation With Thermal Expansion for Geometry Assurance. ASME
2012 IDETC/CIE, American Society of Mechanical Engineers, 477–485.
[25] Carlson JS, Söderberg R (2003) Assembly Root Cause Analysis: A Way to
The Digital Twin can use data from individuals to perform real- Reduce Dimensional Variation in Assembled Products. International Journal
of Flexible Manufacturing Systems 15(2):113–150.
time in-line individual adjustments or data from batches of parts to [26] Wärmefjord K, Carlson JS, Söderberg R (2016) Controlling Geometrical Varia-
make adjustment batch wise. However, individual adjustment tion Caused by Assembly Fixtures. Journal of Computing and Information Science
means new challenges related to scan speed as well as on in Engineering 16(1):011007.
[27] Ceglarek D, Colledani M, Váncza J, Kim D-Y, Marine C, Kogel-Hollacher M,
simulation and analysis speed inside the Digital Twin.
Mistry A, Bolognese L (2015) Rapid Deployment of Remote Laser Welding
An example with a sheet metal assembly station is given but the Processes in Automotive Assembly Systems. CIRP Annals – Manufacturing
concept can be translated to any similar assembly situation. Technology 64(1):389–394.

Please cite this article in press as: Söderberg R, et al. Toward a Digital Twin for real-time geometry assurance in individualized
production. CIRP Annals - Manufacturing Technology (2017), http://dx.doi.org/10.1016/j.cirp.2017.04.038