Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
ScienceDirect
Available online at www.sciencedirect.com
Procedia Engineering 207 (2017) 42–47
1877-7058 © 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientific committee of the International Conference on the Technology of Plasticity.
10.1016/j.proeng.2017.10.735
10.1016/j.proeng.2017.10.735 1877-7058
Available online at www.sciencedirect.com
ScienceDirect
Procedia Engineering 00 (2017) 000–000
www.elsevier.com/locate/procedia
1877-7058 © 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientific committee of the International Conference on the Technology of
Plasticity.
International Conference on the Technology of Plasticity, ICTP 2017, 17-22 September 2017,
Cambridge, United Kingdom
A knowledge-based control system for the robust manufacturing of
deep drawn parts
P. Fischera*, D. Harschb, J. Heingärtnerb, Y. Renkcic, P. Horaa
aETH Zurich, Institut of Virtual Manufacturing, Technoparkstrasse 1, 8005 Zurich, Switzerland
bInspire-ivp, Technoparkstrasse 1, 8005 Zurich, Switzerland
cFranke Technology and Trademark Ltd, Franke-Strasse 2, 4663 Aarburg, Switzerland
Abstract
Throughout a single batch of deep drawing parts the settings of the press have to be adjusted to account for several
influences. These can be divided in influences originating through the process, like heating of the tools or
aggregation of the lubricant in the tool, and influences originating in the manufacturing of the blanks, like scattering
material properties within a coil or between different coils. In the present paper, a method is shown to minimize the
effects of both types of influences. The first step in building up a knowledge based control is the quantification of
the influences. This is done by running a virtual process tryout based on FEM simulations in order to predict the
influence of the scattering material and process properties on the process outcome. For an effective feed forward
control based on the variant system, the blank properties are measured during the cutting stage and every part is
labeled with a unique identification. The yield strength and ultimate tensile strength are measured by an eddy-
current system, while the blank thickness is measured via laser triangulation. As the knowledge of the blank
properties alone is not sufficient, a feedback loop is introduced to compensate for the non-blank related influences.
For the feedback control, an optical measurement system is proposed, which is able to calculate the draw-in at pre-
defined points. The relevant measuring points are defined by evaluation of the correla tion between draw-in and
changing properties in the virtual process tryout. Both control mechanisms are solely using the usual available and
adjustable press settings. In the presented case, the position of the blank as well as the different blankholder forces
were chosen. Finally the applicability of the proposed method is evaluated virtually.
© 2017 The Authors. Published by Elsevier Ltd.
* Corresponding author. Tel.: +41 79 902 09 79; fax: +41 44 633 15 96.
E-mail address: fischer@ivp.mavt.ethz.ch
Available online at www.sciencedirect.com
ScienceDirect
Procedia Engineering 00 (2017) 000–000
www.elsevier.com/locate/procedia
1877-7058 © 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientific committee of the International Conference on the Technology of
Plasticity.
International Conference on the Technology of Plasticity, ICTP 2017, 17-22 September 2017,
Cambridge, United Kingdom
A knowledge-based control system for the robust manufacturing of
deep drawn parts
P. Fischera*, D. Harschb, J. Heingärtnerb, Y. Renkcic, P. Horaa
aETH Zurich, Institut of Virtual Manufacturing, Technoparkstrasse 1, 8005 Zurich, Switzerland
bInspire-ivp, Technoparkstrasse 1, 8005 Zurich, Switzerland
cFranke Technology and Trademark Ltd, Franke-Strasse 2, 4663 Aarburg, Switzerland
Abstract
Throughout a single batch of deep drawing parts the settings of the press have to be adjusted to account for several
influences. These can be divided in influences originating through the process, like heating of the tools or
aggregation of the lubricant in the tool, and influences originating in the manufacturing of the blanks, like scattering
material properties within a coil or between different coils. In the present paper, a method is shown to minimize the
effects of both types of influences. The first step in building up a knowledge based control is the quantification of
the influences. This is done by running a virtual process tryout based on FEM simulations in order to predict the
influence of the scattering material and process properties on the process outcome. For an effective feed forward
control based on the variant system, the blank properties are measured during the cutting stage and every part is
labeled with a unique identification. The yield strength and ultimate tensile strength are measured by an eddy-
current system, while the blank thickness is measured via laser triangulation. As the knowledge of the blank
properties alone is not sufficient, a feedback loop is introduced to compensate for the non-blank related influences.
For the feedback control, an optical measurement system is proposed, which is able to calculate the draw-in at pre-
defined points. The relevant measuring points are defined by evaluation of the correla tion between draw-in and
changing properties in the virtual process tryout. Both control mechanisms are solely using the usual available and
adjustable press settings. In the presented case, the position of the blank as well as the different blankholder forces
were chosen. Finally the applicability of the proposed method is evaluated virtually.
© 2017 The Authors. Published by Elsevier Ltd.
* Corresponding author. Tel.: +41 79 902 09 79; fax: +41 44 633 15 96.
E-mail address: fischer@ivp.mavt.ethz.ch
2 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
Peer-review under responsibility of the scientific committee of the International Conference on the Technology
of Plasticity.
Keywords: Deep drawing; control
1. Introduction
In production of deep drawing parts, scattering material properties as well as process related changes like increased
temperatures in the tools through the forming process, have a significant impact on the scrap rate and therefore the
costs. In an automotive press shop up to 89% of the part costs are material cost [1]. To realize a zero defect production,
a combination of measurement systems and knowledge based control is required. For a knowledge based control it is
necessary to measure all relevant process data. In the past different approaches based on in-line measurement systems
and data mining have been tested. While feedback solutions are covered well in literature, the implementation of
feedforward control in deep-drawing based on material properties is only proposed by Mork, Neumann and
Heingärtner [2][3][4]. In all systems for deep-drawing the state variable (draw-in) is, either measured in the forming
tool as done by Bräunlich [5][2][3][6][7][8] or it is acquired after the drawing operation [9]. Also the location of the
intervention to adapt the process can change. The actuators can be integrated in the forming tool [2][7][8] or the
blankholder forces of the press line can be adjusted [9][3]. For strip bending a possible solution for a model based
control is shown by van den Boogaard [10]. In deep-drawing, Mork [2] gathered data and trained a neural network
using these data. He proposed to directly control the process through neural networks. The approach of Neumann [3]
focused on data collection and evaluation, as well as proposing a possible system for process control. The success of
all these approaches heavily depends on the evaluation of the acquired data. As Neumann showed in her work, not all
influences on the process can be measured directly. For this reason, the proposed control system consists of two
different parts. The feedforward part, which links the measurable influences to the knowledge base and a feedback
loop using the draw-in to compensate non-measurable influences. The required knowledge base is derived from
numerical experiments (FEA simulations). As the knowledge base can only be used for the compensation of
measurable influences, the later on shown feedback loop was designed and tested in production. The control system
described in this work uses the draw-in acquired after the first deep drawing operation and the blankholder forces to
adapt the process. In this contribution the necessary equipment as well as the approach of designing a feedforward
control based on simulation data is shown.
Nomenclature
S Virtual draw-in sensor position
TM Measured blank thickness
F Force until 80mm
2. Virtual process tryout
The first step in building up a knowledge based control, is a thorough investigation of the process. As the
demonstrator part is already in series production, two different methods could be used. The first method would be
experiments in the press shop. As most of the influences on the part quality (e.g. material properties) cannot be changed
according to a design of experiment, virtual process modelling is chosen for building up the knowledge. In detail, the
kitchen sink that is chosen as demonstrator is modelled with the finite element method in AutoForm. As knowledge
based control works only when the model has a good agreement with the reality, a nominal simulation is build up
which corresponds to the series production part. Therefore the material is characterized with tensile tests at different
temperatures as well as the bulge test. As the material is the stainless steel 1.4301, the martensitic content during
tensile tests is measured as well. The Hänsel model is chosen as baseline model for the material. The baseline model
is afterwards collapsed to a single yield curve at 30 degree Celsius for the calculation in AutoForm. As yield surface
the BBC model is chosen [9]. The forces in the nominal model are chosen according to the press settings in series
production, which means that the blank holder forces are changing after a drawing depth of 80mm while the complete
2 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
Peer-review under responsibility of the scientific committee of the International Conference on the Technology
of Plasticity.
Keywords: Deep drawing; control
1. Introduction
In production of deep drawing parts, scattering material properties as well as process related changes like increased
temperatures in the tools through the forming process, have a significant impact on the scrap rate and therefore the
costs. In an automotive press shop up to 89% of the part costs are material cost [1]. To realize a zero defect production,
a combination of measurement systems and knowledge based control is required. For a knowledge based control it is
necessary to measure all relevant process data. In the past different approaches based on in-line measurement systems
and data mining have been tested. While feedback solutions are covered well in literature, the implementation of
feedforward control in deep-drawing based on material properties is only proposed by Mork, Neumann and
Heingärtner [2][3][4]. In all systems for deep-drawing the state variable (draw-in) is, either measured in the forming
tool as done by Bräunlich [5][2][3][6][7][8] or it is acquired after the drawing operation [9]. Also the location of the
intervention to adapt the process can change. The actuators can be integrated in the forming tool [2][7][8] or the
blankholder forces of the press line can be adjusted [9][3]. For strip bending a possible solution for a model based
control is shown by van den Boogaard [10]. In deep-drawing, Mork [2] gathered data and trained a neural network
using these data. He proposed to directly control the process through neural networks. The approach of Neumann [3]
focused on data collection and evaluation, as well as proposing a possible system for process control. The success of
all these approaches heavily depends on the evaluation of the acquired data. As Neumann showed in her work, not all
influences on the process can be measured directly. For this reason, the proposed control system consists of two
different parts. The feedforward part, which links the measurable influences to the knowledge base and a feedback
loop using the draw-in to compensate non-measurable influences. The required knowledge base is derived from
numerical experiments (FEA simulations). As the knowledge base can only be used for the compensation of
measurable influences, the later on shown feedback loop was designed and tested in production. The control system
described in this work uses the draw-in acquired after the first deep drawing operation and the blankholder forces to
adapt the process. In this contribution the necessary equipment as well as the approach of designing a feedforward
control based on simulation data is shown.
Nomenclature
S Virtual draw-in sensor position
TM Measured blank thickness
F Force until 80mm
2. Virtual process tryout
The first step in building up a knowledge based control, is a thorough investigation of the process. As the
demonstrator part is already in series production, two different methods could be used. The first method would be
experiments in the press shop. As most of the influences on the part quality (e.g. material properties) cannot be changed
according to a design of experiment, virtual process modelling is chosen for building up the knowledge. In detail, the
kitchen sink that is chosen as demonstrator is modelled with the finite element method in AutoForm. As knowledge
based control works only when the model has a good agreement with the reality, a nominal simulation is build up
which corresponds to the series production part. Therefore the material is characterized with tensile tests at different
temperatures as well as the bulge test. As the material is the stainless steel 1.4301, the martensitic content during
tensile tests is measured as well. The Hänsel model is chosen as baseline model for the material. The baseline model
is afterwards collapsed to a single yield curve at 30 degree Celsius for the calculation in AutoForm. As yield surface
the BBC model is chosen [9]. The forces in the nominal model are chosen according to the press settings in series
production, which means that the blank holder forces are changing after a drawing depth of 80mm while the complete
© 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientic committee of the International Conference on the Technology of Plasticity.
P. Fischer et al. / Procedia Engineering 207 (2017) 42–47 43
Available online at www.sciencedirect.com
ScienceDirect
Procedia Engineering 00 (2017) 000–000
www.elsevier.com/locate/procedia
1877-7058 © 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientific committee of the International Conference on the Technology of
Plasticity.
International Conference on the Technology of Plasticity, ICTP 2017, 17-22 September 2017,
Cambridge, United Kingdom
A knowledge-based control system for the robust manufacturing of
deep drawn parts
P. Fischera*, D. Harschb, J. Heingärtnerb, Y. Renkcic, P. Horaa
aETH Zurich, Institut of Virtual Manufacturing, Technoparkstrasse 1, 8005 Zurich, Switzerland
bInspire-ivp, Technoparkstrasse 1, 8005 Zurich, Switzerland
cFranke Technology and Trademark Ltd, Franke-Strasse 2, 4663 Aarburg, Switzerland
Abstract
Throughout a single batch of deep drawing parts the settings of the press have to be adjusted to account for several
influences. These can be divided in influences originating through the process, like heating of the tools or
aggregation of the lubricant in the tool, and influences originating in the manufacturing of the blanks, like scattering
material properties within a coil or between different coils. In the present paper, a method is shown to minimize the
effects of both types of influences. The first step in building up a knowledge based control is the quantification of
the influences. This is done by running a virtual process tryout based on FEM simulations in order to predict the
influence of the scattering material and process properties on the process outcome. For an effective feed forward
control based on the variant system, the blank properties are measured during the cutting stage and every part is
labeled with a unique identification. The yield strength and ultimate tensile strength are measured by an eddy-
current system, while the blank thickness is measured via laser triangulation. As the knowledge of the blank
properties alone is not sufficient, a feedback loop is introduced to compensate for the non-blank related influences.
For the feedback control, an optical measurement system is proposed, which is able to calculate the draw-in at pre-
defined points. The relevant measuring points are defined by evaluation of the correla tion between draw-in and
changing properties in the virtual process tryout. Both control mechanisms are solely using the usual available and
adjustable press settings. In the presented case, the position of the blank as well as the different blankholder forces
were chosen. Finally the applicability of the proposed method is evaluated virtually.
© 2017 The Authors. Published by Elsevier Ltd.
* Corresponding author. Tel.: +41 79 902 09 79; fax: +41 44 633 15 96.
E-mail address: fischer@ivp.mavt.ethz.ch
Available online at www.sciencedirect.com
ScienceDirect
Procedia Engineering 00 (2017) 000–000
www.elsevier.com/locate/procedia
1877-7058 © 2017 The Authors. Published by Elsevier Ltd.
Peer-review under responsibility of the scientific committee of the International Conference on the Technology of
Plasticity.
International Conference on the Technology of Plasticity, ICTP 2017, 17-22 September 2017,
Cambridge, United Kingdom
A knowledge-based control system for the robust manufacturing of
deep drawn parts
P. Fischera*, D. Harschb, J. Heingärtnerb, Y. Renkcic, P. Horaa
aETH Zurich, Institut of Virtual Manufacturing, Technoparkstrasse 1, 8005 Zurich, Switzerland
bInspire-ivp, Technoparkstrasse 1, 8005 Zurich, Switzerland
cFranke Technology and Trademark Ltd, Franke-Strasse 2, 4663 Aarburg, Switzerland
Abstract
Throughout a single batch of deep drawing parts the settings of the press have to be adjusted to account for several
influences. These can be divided in influences originating through the process, like heating of the tools or
aggregation of the lubricant in the tool, and influences originating in the manufacturing of the blanks, like scattering
material properties within a coil or between different coils. In the present paper, a method is shown to minimize the
effects of both types of influences. The first step in building up a knowledge based control is the quantification of
the influences. This is done by running a virtual process tryout based on FEM simulations in order to predict the
influence of the scattering material and process properties on the process outcome. For an effective feed forward
control based on the variant system, the blank properties are measured during the cutting stage and every part is
labeled with a unique identification. The yield strength and ultimate tensile strength are measured by an eddy-
current system, while the blank thickness is measured via laser triangulation. As the knowledge of the blank
properties alone is not sufficient, a feedback loop is introduced to compensate for the non-blank related influences.
For the feedback control, an optical measurement system is proposed, which is able to calculate the draw-in at pre-
defined points. The relevant measuring points are defined by evaluation of the correla tion between draw-in and
changing properties in the virtual process tryout. Both control mechanisms are solely using the usual available and
adjustable press settings. In the presented case, the position of the blank as well as the different blankholder forces
were chosen. Finally the applicability of the proposed method is evaluated virtually.
© 2017 The Authors. Published by Elsevier Ltd.
* Corresponding author. Tel.: +41 79 902 09 79; fax: +41 44 633 15 96.
E-mail address: fischer@ivp.mavt.ethz.ch
2 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
Peer-review under responsibility of the scientific committee of the International Conference on the Technology
of Plasticity.
Keywords: Deep drawing; control
1. Introduction
In production of deep drawing parts, scattering material properties as well as process related changes like increased
temperatures in the tools through the forming process, have a significant impact on the scrap rate and therefore the
costs. In an automotive press shop up to 89% of the part costs are material cost [1]. To realize a zero defect production,
a combination of measurement systems and knowledge based control is required. For a knowledge based control it is
necessary to measure all relevant process data. In the past different approaches based on in-line measurement systems
and data mining have been tested. While feedback solutions are covered well in literature, the implementation of
feedforward control in deep-drawing based on material properties is only proposed by Mork, Neumann and
Heingärtner [2][3][4]. In all systems for deep-drawing the state variable (draw-in) is, either measured in the forming
tool as done by Bräunlich [5][2][3][6][7][8] or it is acquired after the drawing operation [9]. Also the location of the
intervention to adapt the process can change. The actuators can be integrated in the forming tool [2][7][8] or the
blankholder forces of the press line can be adjusted [9][3]. For strip bending a possible solution for a model based
control is shown by van den Boogaard [10]. In deep-drawing, Mork [2] gathered data and trained a neural network
using these data. He proposed to directly control the process through neural networks. The approach of Neumann [3]
focused on data collection and evaluation, as well as proposing a possible system for process control. The success of
all these approaches heavily depends on the evaluation of the acquired data. As Neumann showed in her work, not all
influences on the process can be measured directly. For this reason, the proposed control system consists of two
different parts. The feedforward part, which links the measurable influences to the knowledge base and a feedback
loop using the draw-in to compensate non-measurable influences. The required knowledge base is derived from
numerical experiments (FEA simulations). As the knowledge base can only be used for the compensation of
measurable influences, the later on shown feedback loop was designed and tested in production. The control system
described in this work uses the draw-in acquired after the first deep drawing operation and the blankholder forces to
adapt the process. In this contribution the necessary equipment as well as the approach of designing a feedforward
control based on simulation data is shown.
Nomenclature
S Virtual draw-in sensor position
TM Measured blank thickness
F Force until 80mm
2. Virtual process tryout
The first step in building up a knowledge based control, is a thorough investigation of the process. As the
demonstrator part is already in series production, two different methods could be used. The first method would be
experiments in the press shop. As most of the influences on the part quality (e.g. material properties) cannot be changed
according to a design of experiment, virtual process modelling is chosen for building up the knowledge. In detail, the
kitchen sink that is chosen as demonstrator is modelled with the finite element method in AutoForm. As knowledge
based control works only when the model has a good agreement with the reality, a nominal simulation is build up
which corresponds to the series production part. Therefore the material is characterized with tensile tests at different
temperatures as well as the bulge test. As the material is the stainless steel 1.4301, the martensitic content during
tensile tests is measured as well. The Hänsel model is chosen as baseline model for the material. The baseline model
is afterwards collapsed to a single yield curve at 30 degree Celsius for the calculation in AutoForm. As yield surface
the BBC model is chosen [9]. The forces in the nominal model are chosen according to the press settings in series
production, which means that the blank holder forces are changing after a drawing depth of 80mm while the complete
2 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
Peer-review under responsibility of the scientific committee of the International Conference on the Technology
of Plasticity.
Keywords: Deep drawing; control
1. Introduction
In production of deep drawing parts, scattering material properties as well as process related changes like increased
temperatures in the tools through the forming process, have a significant impact on the scrap rate and therefore the
costs. In an automotive press shop up to 89% of the part costs are material cost [1]. To realize a zero defect production,
a combination of measurement systems and knowledge based control is required. For a knowledge based control it is
necessary to measure all relevant process data. In the past different approaches based on in-line measurement systems
and data mining have been tested. While feedback solutions are covered well in literature, the implementation of
feedforward control in deep-drawing based on material properties is only proposed by Mork, Neumann and
Heingärtner [2][3][4]. In all systems for deep-drawing the state variable (draw-in) is, either measured in the forming
tool as done by Bräunlich [5][2][3][6][7][8] or it is acquired after the drawing operation [9]. Also the location of the
intervention to adapt the process can change. The actuators can be integrated in the forming tool [2][7][8] or the
blankholder forces of the press line can be adjusted [9][3]. For strip bending a possible solution for a model based
control is shown by van den Boogaard [10]. In deep-drawing, Mork [2] gathered data and trained a neural network
using these data. He proposed to directly control the process through neural networks. The approach of Neumann [3]
focused on data collection and evaluation, as well as proposing a possible system for process control. The success of
all these approaches heavily depends on the evaluation of the acquired data. As Neumann showed in her work, not all
influences on the process can be measured directly. For this reason, the proposed control system consists of two
different parts. The feedforward part, which links the measurable influences to the knowledge base and a feedback
loop using the draw-in to compensate non-measurable influences. The required knowledge base is derived from
numerical experiments (FEA simulations). As the knowledge base can only be used for the compensation of
measurable influences, the later on shown feedback loop was designed and tested in production. The control system
described in this work uses the draw-in acquired after the first deep drawing operation and the blankholder forces to
adapt the process. In this contribution the necessary equipment as well as the approach of designing a feedforward
control based on simulation data is shown.
Nomenclature
S Virtual draw-in sensor position
TM Measured blank thickness
F Force until 80mm
2. Virtual process tryout
The first step in building up a knowledge based control, is a thorough investigation of the process. As the
demonstrator part is already in series production, two different methods could be used. The first method would be
experiments in the press shop. As most of the influences on the part quality (e.g. material properties) cannot be changed
according to a design of experiment, virtual process modelling is chosen for building up the knowledge. In detail, the
kitchen sink that is chosen as demonstrator is modelled with the finite element method in AutoForm. As knowledge
based control works only when the model has a good agreement with the reality, a nominal simulation is build up
which corresponds to the series production part. Therefore the material is characterized with tensile tests at different
temperatures as well as the bulge test. As the material is the stainless steel 1.4301, the martensitic content during
tensile tests is measured as well. The Hänsel model is chosen as baseline model for the material. The baseline model
is afterwards collapsed to a single yield curve at 30 degree Celsius for the calculation in AutoForm. As yield surface
the BBC model is chosen [9]. The forces in the nominal model are chosen according to the press settings in series
production, which means that the blank holder forces are changing after a drawing depth of 80mm while the complete
44 P. Fischer et al. / Procedia Engineering 207 (2017) 42–47
Pascal Fischer/ Procedia Engineering 00 (2017) 000–000 3
drawing depth is 180mm. The friction coefficient is used to match the draw-in of a 3D scanned real part with the
nominal simulation.
2.1. Process model
The varied parameters and their limits can be seen in table 1. In the input variations, it is necessary to distinguish
between observable and unobservable parameters. In the present case all parameters except friction are observable as
they are either measurable by the systems in section three or they are adjustable on the press.
As the draw-in of the blank can be used as indicator for the strain distribution, different virtual draw-in sensors
(figure 1) are evaluated and modelled by quadratic approximation functions.
Fig. 1. Position of virtual draw-in sensor
Table 1. Input variations for process model
Parameter
Lower limit
Nominal
Upper limit
Blank position in material flow direction (arrow fig. 1)
-5
5
15
Blank thickness
0.775mm
0.8mm
0.825mm
Friction
0.05
0.07
0.12
Force until 80 mm
1600kN
2000kN
2400kN
Force after 100 mm
450kN
750kN
1050kN
Force distribution
-40mm
Center
40mm
Yield stress/Yield strength delta
-25/-50 MPa
0
25/50MPa
2.2. Analysis of process model
To be able to control the parts, the influence of the changing parameters has to be detectable in the chosen sensor
position, therefore a global sensitivity analysis after Cannavó [11] is done.
Table 2. Global sensitivity analysis for the sensors
Parameter
S01
S02
S03
S04
Blank position in material flow direction
51%
52%
1%
0%
Blank thickness
1%
1%
0%
1%
Friction
35%
33%
66%
68%
Force until 80 mm
4%
4%
6%
5%
Force after 100 mm
5%
4%
9%
11%
Force distribution
6%
5%
14%
11%
Yield stress/Yield strength delta
2%
1%
1%
1%
The results in table 2 show that friction has an important influence on all sensors and therefore a change in friction
should be easily detectable in a change of the sensor value. For sensors S01 and S02, the main influence is the blank
P. Fischer et al. / Procedia Engineering 207 (2017) 42–47 45
Pascal Fischer/ Procedia Engineering 00 (2017) 000–000 3
drawing depth is 180mm. The friction coefficient is used to match the draw-in of a 3D scanned real part with the
nominal simulation.
2.1. Process model
The varied parameters and their limits can be seen in table 1. In the input variations, it is necessary to distinguish
between observable and unobservable parameters. In the present case all parameters except friction are observable as
they are either measurable by the systems in section three or they are adjustable on the press.
As the draw-in of the blank can be used as indicator for the strain distribution, different virtual draw-in sensors
(figure 1) are evaluated and modelled by quadratic approximation functions.
Fig. 1. Position of virtual draw-in sensor
Table 1. Input variations for process model
Parameter
Lower limit
Nominal
Upper limit
Blank position in material flow direction (arrow fig. 1)
-5
5
15
Blank thickness
0.775mm
0.8mm
0.825mm
Friction
0.05
0.07
0.12
Force until 80 mm
1600kN
2000kN
2400kN
Force after 100 mm
450kN
750kN
1050kN
Force distribution
-40mm
Center
40mm
Yield stress/Yield strength delta
-25/-50 MPa
0
25/50MPa
2.2. Analysis of process model
To be able to control the parts, the influence of the changing parameters has to be detectable in the chosen sensor
position, therefore a global sensitivity analysis after Cannavó [11] is done.
Table 2. Global sensitivity analysis for the sensors
Parameter
S01
S02
S03
S04
Blank position in material flow direction
51%
52%
1%
0%
Blank thickness
1%
1%
0%
1%
Friction
35%
33%
66%
68%
Force until 80 mm
4%
4%
6%
5%
Force after 100 mm
5%
4%
9%
11%
Force distribution
6%
5%
14%
11%
Yield stress/Yield strength delta
2%
1%
1%
1%
The results in table 2 show that friction has an important influence on all sensors and therefore a change in friction
should be easily detectable in a change of the sensor value. For sensors S01 and S02, the main influence is the blank
4 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
position as it is shifted in this direction. As all variations have at least some sensitivity, regarding the input variations,
they can be chosen for the control algorithm. In addition to corresponding to the input influences, an analysis of the
correlations shows, that the sensors also correspond to post processing criteria like thinning and maximum failure in
the corners of the sink (e.g. the correlation between S01 and the thinning in the edge between S01 and S03 is 0.82.
3. Measurement systems
The later on proposed control algorithm consists of two independent parts, which need different inputs to react. For
the feedforward control, it is necessary to gain as much knowledge about the blank as possible. Therefore the
mechanical material properties and the blank thickness are measured. For the usage in a fast reacting feedback control
based on the instantaneous draw-in, the draw-in of each part has to be measured in-line. The details of the currently
at Franke installed system can be found in the following sections.
3.1. Inline measurement of material properties
In order to achieve a zero defect manufacturing, already the first part has to be produced using the correct process
settings Batch related influences have to be accounted for before producing the first part. Therefore the blank thickness
of each blank is measured as well as the mechanical properties yield strength and tensile strength. These two properties
are measured non-destructive via eddy current. More details can be found in the work of M. Ruzovic [12] and J.
Heingärtner [13]. For traceability reasons, each blank is marked with a data matrix code and scanned before processing
in the press line.
3.2. Inline measurement of draw-in
A feedback loop has to be designed to compensate the non-measurable influences. As input for the feedback
control, the draw-in is measured by an optical measurement system between the first and second drawing stage. To
be able to cover different parts, a camera with a large enough image section to view the whole draw-in of different
parts is chosen. Further details can be found in the paper of J. Heingärtner [4] , while an overview of other possible
sensors can be found in the paper of Allwood[6] and in the thesis of Neumann[3].
4. Design of control algorithm
Fig. 2. Control scheme
To account for the known fluctuations as well as unknown effects, the control algorithm (fig. 4) consists of two
independent parts. The previously generated knowledge from numerical simulations of the process is used in the feed
forward approach to improve the part quality even before the feedback loop would be able to react. Based on the meta
models the feedback loop is designed to compensate the influences of the non-measurable effects.
4.1. Feedforward control
For the implementation of feedforward control different approaches are feasible, e.g. linearization around the
nominal value for linear correction or the usage of the local gradient and last but not least an optimal control approach
46 P. Fischer et al. / Procedia Engineering 207 (2017) 42–47
Pascal Fischer/ Procedia Engineering 00 (2017) 000–000 5
which corrects the influence based on the occurring change in the meta model. This paper deals with the approach of
linearization around the nominal working point and the theoretical performance of this approach. The steps from
modelling the process by a quadratic model with interactions to reaching the final correction value can be seen in
equation one to three. In the beginning the interaction terms are neglected for simplification, as well as the forces until
80mm and after 80mm are coupled by a factor. Therefore a change in a sensor can be reduced to a change in the yield
stress and a change in the force until 80mm. As the strain distribution between parts should not change, a change in
the sensor reading should be prevented and leads to equation three, which gives the force change to suppress an output
change by a changing yield stress.
2
4
2
3210 ...
yy FFS
(1)
yynomnom
FFS
)2()2(
4231
(2)
)2(
)2(
31
42
nom
yynom
F
F
(3)
4.2. Feedback control
A pure feed forward control would need a perfect model for the correction, as well as the feed forward control is
not able to react on non-measurable influences, therefore a feedback loop is developed depending on the sensor values
S03 and S04 as they show a high sensitivity regarding the friction which is not measurable. For a better understanding
of the behavior, the feedback control is chosen to be a P-controller. The calibration of the controller is done by a
parameter variation in a Simulink environment, representing the process.
5. Virtual performance check
For the purpose of testing the proposed control algorithms, a testing environment is build up in Simulink. In the
testing environment, different scenarios can be evaluated. For testing the feedback control, a kind of coil change after
9 and after 10 parts is assumed (left hand figure 3) while all other parameters stay constant and the reaction on the
different sensors is checked. On the right side of figure 3, the reaction in sensor S03 is shown, which clearly shows
that the effect of the changing yield stress can be suppressed quite well. Due to the linearization, a small error remains,
which could be removed with a feedback loop. In the real process, the error would add up, as already before the
linearization, the model might diverge from the real process.
Fig. 3. Step function in σy and response of the draw-in sensor S03
The second scenario for the evaluation of the control algorithm is sudden change in friction which might occur due
to a break or a change in the lubrication. As no estimator is used for the friction, only the feedback control can react
on this change. The step function in the lubrication is displayed on the left side of figure 4, while the reaction on the
6 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
change can be seen on the right. The feedback control needs the measurement of one part to detect the change and is
able to bring the part back to the reference draw- in in a few parts.
Fig. 4. Step function in friction and response of the draw-in sensors
6. Conclusion
The proposed method for the design of a feed forward control in combination with a feedback control looks
extremely promising. In the defined testing environment, the controller is able to compensate for both influences very
well. For the usage in the industrial environment, further tests regarding the influence of noisy measurements regarding
the stability have to be performed. As the first tests with only feedback control worked out well [9], the robustness
should be granted with the feed forward system as well. In a next step, the control system as a whole system will be
tested in the production line by using blanks of different coils for the feedforward approach as well as changing
lubrication for a second verification of the feedback control.
Acknowledgements
The authors are grateful for the support of the CTI (Commission for Technology and Innovation) within the project
17366.1 PFIW-IW and also for support of the additionally participating companies (Franke, AutoForm and GOM).
References
[1] D. Hortig, “Experiences with the robustness of sheet metal forming processes,” in 4th Forming Technology Forum, 2011.
[2] R. Mork, “Qualitätsbewertung und Regelung für die Fertigung von Karosserieteilen in Presswerken auf Basis Neuronaler Netze,”
Technischen Universität München, 2011.
[3] A. Neumann, “Konzept zur Beherrschung der Prozessschwankungen im Presswerk,” Meisenbach, Bamberg, 2014.
[4] J. Heingärtner, P. Fischer, P. Horat, D. Harsch, Y. Renkci, and P. Hora, “Optical draw-in measurement in kitchen sink production,” in
Proceedings of IDDRG 2016, 2016.
[5] H. Bräunlich, Blecheinzugsregelung beim Tiefziehen mit Niederhalter - ein Beitrag zur Erhöhung der Prozeßstabilität. Zwickau:
Verlag Wissenschaftliche Scripten, 2002.
[6] J. M. Allwood et al., “Closed-loop control of product properties in metal forming,” CIRP Ann. Technol., vol. 65, no. 2, pp. 573–596,
2016.
[7] B. Endelt, S. S. Tommerup, and J. Danckert, “A novel feedback control system - Controlling the material flow in deep drawing using
distributed blank-holder force,” J. Mater. Process. Technol., vol. 213, no. 1, pp. 36–50, Jan. 2013.
[8] T. Bäume, W. Zorn, W.-G. Drossel, and G. Rupp, “Step by step control of a deep drawing process with piezo-electric actuators in serial
operation,” MATEC Web Conf., vol. 21, p. 4008, Aug. 2015.
[9] P. Fischer, D. Harsch, J. Heingärtner, Y. Renkci, and P. Hora, “Inline feedback control for deep drawing applications,” IOP Conf. Ser.
Mater. Sci. Eng., vol. 159, p. 12006, Nov. 2016.
[10] T. van den Boogaard, J. Havinga, and R. van Tijum, “Model-based control of strip bending in mass production,” CIRP Ann. - Manuf.
Technol., vol. 64, no. 1, pp. 297–300, 2015.
[11] F. Cannavó, “Sensitivity analysis for volcanic source modeling quality assessment and model selection,” Comput. Geosci., vol. 44, pp.
52–59, 2012.
[12] M. Ružovič, “Die zerstörungsfreie Ermittlung von genauen Zugversuchsdaten mit dem Wirbelstromverfahren,” Zürich, Zürich, 2004.
[13] J. Heingärtner, Y. Renkci, and P. Hora, “Non-destructive testing of material properties,” in 6th Forming Technology Forum, 2013.
P. Fischer et al. / Procedia Engineering 207 (2017) 42–47 47
Pascal Fischer/ Procedia Engineering 00 (2017) 000–000 5
which corrects the influence based on the occurring change in the meta model. This paper deals with the approach of
linearization around the nominal working point and the theoretical performance of this approach. The steps from
modelling the process by a quadratic model with interactions to reaching the final correction value can be seen in
equation one to three. In the beginning the interaction terms are neglected for simplification, as well as the forces until
80mm and after 80mm are coupled by a factor. Therefore a change in a sensor can be reduced to a change in the yield
stress and a change in the force until 80mm. As the strain distribution between parts should not change, a change in
the sensor reading should be prevented and leads to equation three, which gives the force change to suppress an output
change by a changing yield stress.
2
4
2
3210 ...
yy FFS
(1)
yynomnom
FFS
)2()2(
4231
(2)
)2(
)2(
31
42
nom
yynom
F
F
(3)
4.2. Feedback control
A pure feed forward control would need a perfect model for the correction, as well as the feed forward control is
not able to react on non-measurable influences, therefore a feedback loop is developed depending on the sensor values
S03 and S04 as they show a high sensitivity regarding the friction which is not measurable. For a better understanding
of the behavior, the feedback control is chosen to be a P-controller. The calibration of the controller is done by a
parameter variation in a Simulink environment, representing the process.
5. Virtual performance check
For the purpose of testing the proposed control algorithms, a testing environment is build up in Simulink. In the
testing environment, different scenarios can be evaluated. For testing the feedback control, a kind of coil change after
9 and after 10 parts is assumed (left hand figure 3) while all other parameters stay constant and the reaction on the
different sensors is checked. On the right side of figure 3, the reaction in sensor S03 is shown, which clearly shows
that the effect of the changing yield stress can be suppressed quite well. Due to the linearization, a small error remains,
which could be removed with a feedback loop. In the real process, the error would add up, as already before the
linearization, the model might diverge from the real process.
Fig. 3. Step function in σy and response of the draw-in sensor S03
The second scenario for the evaluation of the control algorithm is sudden change in friction which might occur due
to a break or a change in the lubrication. As no estimator is used for the friction, only the feedback control can react
on this change. The step function in the lubrication is displayed on the left side of figure 4, while the reaction on the
6 Pascal Fischer/ Procedia Engineering 00 (2017) 000–000
change can be seen on the right. The feedback control needs the measurement of one part to detect the change and is
able to bring the part back to the reference draw- in in a few parts.
Fig. 4. Step function in friction and response of the draw-in sensors
6. Conclusion
The proposed method for the design of a feed forward control in combination with a feedback control looks
extremely promising. In the defined testing environment, the controller is able to compensate for both influences very
well. For the usage in the industrial environment, further tests regarding the influence of noisy measurements regarding
the stability have to be performed. As the first tests with only feedback control worked out well [9], the robustness
should be granted with the feed forward system as well. In a next step, the control system as a whole system will be
tested in the production line by using blanks of different coils for the feedforward approach as well as changing
lubrication for a second verification of the feedback control.
Acknowledgements
The authors are grateful for the support of the CTI (Commission for Technology and Innovation) within the project
17366.1 PFIW-IW and also for support of the additionally participating companies (Franke, AutoForm and GOM).
References
[1] D. Hortig, “Experiences with the robustness of sheet metal forming processes,” in 4th Forming Technology Forum, 2011.
[2] R. Mork, “Qualitätsbewertung und Regelung für die Fertigung von Karosserieteilen in Presswerken auf Basis Neuronaler Netze,”
Technischen Universität München, 2011.
[3] A. Neumann, “Konzept zur Beherrschung der Prozessschwankungen im Presswerk,” Meisenbach, Bamberg, 2014.
[4] J. Heingärtner, P. Fischer, P. Horat, D. Harsch, Y. Renkci, and P. Hora, “Optical draw-in measurement in kitchen sink production,” in
Proceedings of IDDRG 2016, 2016.
[5] H. Bräunlich, Blecheinzugsregelung beim Tiefziehen mit Niederhalter - ein Beitrag zur Erhöhung der Prozeßstabilität. Zwickau:
Verlag Wissenschaftliche Scripten, 2002.
[6] J. M. Allwood et al., “Closed-loop control of product properties in metal forming,” CIRP Ann. Technol., vol. 65, no. 2, pp. 573–596,
2016.
[7] B. Endelt, S. S. Tommerup, and J. Danckert, “A novel feedback control system - Controlling the material flow in deep drawing using
distributed blank-holder force,” J. Mater. Process. Technol., vol. 213, no. 1, pp. 36–50, Jan. 2013.
[8] T. Bäume, W. Zorn, W.-G. Drossel, and G. Rupp, “Step by step control of a deep drawing process with piezo-electric actuators in serial
operation,” MATEC Web Conf., vol. 21, p. 4008, Aug. 2015.
[9] P. Fischer, D. Harsch, J. Heingärtner, Y. Renkci, and P. Hora, “Inline feedback control for deep drawing applications,” IOP Conf. Ser.
Mater. Sci. Eng., vol. 159, p. 12006, Nov. 2016.
[10] T. van den Boogaard, J. Havinga, and R. van Tijum, “Model-based control of strip bending in mass production,” CIRP Ann. - Manuf.
Technol., vol. 64, no. 1, pp. 297–300, 2015.
[11] F. Cannavó, “Sensitivity analysis for volcanic source modeling quality assessment and model selection,” Comput. Geosci., vol. 44, pp.
52–59, 2012.
[12] M. Ružovič, “Die zerstörungsfreie Ermittlung von genauen Zugversuchsdaten mit dem Wirbelstromverfahren,” Zürich, Zürich, 2004.
[13] J. Heingärtner, Y. Renkci, and P. Hora, “Non-destructive testing of material properties,” in 6th Forming Technology Forum, 2013.
