Int J Adv Manuf Technol (2000) 16:370–375

 2000 Springer-Verlag London Limited

A Hybrid Intelligent Systems Approach for Die Design in

Sheet Metal Forming
R. Pilani1, K. Narasimhan1, S. K. Maiti2, U. P. Singh3 and P. P. Date2
1
Department of Metallurgical Engineering and Materials Science, IIT Bombay, India; 2Department of Mechanical Engineering, IIT
Bombay, India; and 3School of Electrical and Mechanical Engineering, University of Ulster, NI, UK

Die design is heavily experience based and the die design Knowledge-based blackboard architecture is used for the
process is an iterative procedure of trial and error in order integration of various analysis models such as CAD, FEA, and
to obtain a final die design for the successful manufacture of ANN, as an intelligent framework for die design [4]. The
stampings. Most automotive industries use internal guidelines hybrid intelligent system provides an integrated decision sup-
and past experience for die design. Even though powerful port environment for simulation and analysis of the forming
computer-aided design systems are being used in automotive process, both during the initial die design phase and during
industry, the lack of adequate analysis tools at the initial die the die tryout phase. The hybrid intelligent systems approach
geometry design stage hinders the die manufacturing process, supports the capability for automatic evaluation of prospective
and also necessitates lead times of the order of 5–30 weeks die design for manufacturability, and performs automatic modi-
[1]. At the concept design stage, and during the initial die fication of design inputs. Applications of the hybrid intelligent
development process, the variations in geometry and process system for die design are described together with a comparison
conditions are so large that it is prohibitively expensive to use with shop floor data.
3D finite element analysis. The complexity of die design heuris-
tic knowledge hinders the development and application of Keywords: Die design optimisation; Hybrid intelligent system
knowledge-based systems.
Hybrid intelligent systems are computer programs in which
at least one of the constituent models simulates intelligent
behaviour [2]. These models could be knowledge-based systems, 1. Introduction
artificial neural networks, fuzzy logic systems, etc. In this
approach both artificial neural networks, knowledge-based In the automotive industry, dies are used to transform a sheet
systems and finite-element analysis (FEA) for modelling the metal blank into a sheet metal component having complex
design process are used. A simulation-based design approach geometry with well-defined shape, size, appearance and proper-
[3] for the die design process is followed. Artificial neural ties. The desired component geometry is stored in the dies and
networks (ANNs) are preliminary design tools which indicate imparted to the material with pressure through a die–material
the formability of the component geometry, for the selected interface [5]. Design of dies is a complex procedure and,
process and material conditions. The ANN module is trained typically, die design takes 20% of the lead time from concept
from FEA results for a generic set of component geometries, design to the final stamping manufacture [1]. It is estimated
process conditions, and material properties. The final die that decisions made at the product design stage determine
design validation is carried out by FEA. The intelligent frame- 70%–80% of the manufacturing productivity [6]. Hence, deter-
work incorporates rules for material selection, process para- mination of initial die geometry is an important phase of die
meter selection and their modification. design, as it influences the final product manufacturability and
Component geometry is a critical parameter which affects appearance. The shape of the die face geometry for automotive
the manufacturability of the given part. Hence, an intelligent panel surfaces is of critical importance in stamping design.
geometry handling module, which automatically modifies and Non-homogeneous die face geometry results in imperfect high-
optimises the geometry of the designed die, is implemented in light lines, dimensional inaccuracies, and failures such as tear-
the present system. ing, splitting, and wrinkling. In conventional die design prac-
tice, a process of trial and error is used, resulting in
considerable lead times in the production of the stamping dies.
Correspondence and offprint requests to: Dr U. P. Singh, The CAD/CAM systems offer die-designers productivity increase
Advanced Forming Group, University of Ulster, Newtonabbey, County through reduction in drafting, visualisation, and accurate stor-
Antrim, BT37 0QB, UK. E-mail: up.singh얀ulst.ac.uk age, and retrieval of component geometric data. However,
 Hybrid Intelligent Systems for Die Design 371

CAD/CAM outputs are based on die designers’ experience for

die face design. Hence, for new component geometries and
innovative materials, the die design process becomes a highly
complex and time consuming task. Application of finite-element
analysis (FEA) for assessing formability during the initial die
face design phase is prohibitively expensive [7]. FEA requires
considerable expertise in setting up the initial model, meshing
the geometry, applying the material and processing boundary
conditions, executing the FEA model, and analysing the out-
puts. In addition, FEA provides forming simulation for full
component geometry, which is not required at the initial phase
of die design.
Knowledge base systems (KBSs) are intelligent computer
programs which simulate human decision-making ability though
the use of a separate reasoning engine and a set of design
rules stored in the knowledge base. Since the die design is
highly experience based and depends upon designers’ past
experience and company guidelines, very few KBSs have been
developed in the sheet metal forming arena and they are
usually limited to prototype models [5,8–10]. Development and
application of KBSs for die design is highly complex and
these knowledge bases are brittle. Additionally, the knowledge
acquisition process for gathering design knowledge from human
experts is a major bottleneck in the development of knowledge-
based systems. Traditional die design practice consists of the
various stages starting from product conceptual design to initial
die face design, die manufacturing and die tryout. During
this cycle of design, manufacture, and verify, human experts
generalise from past experience, perform modifications to
obtain defect-free design, and learn. Current CAD/CAM/FEA
tools lack this important self-learning characteristic.
In this work, a hybrid intelligent system for design is
developed to aid the engineer during the die design process.
Figure 1 presents the various modules and the design method-
Fig. 1. A hybrid intelligent system for die design.
ology used by the hybrid intelligent system. The system incor-
porates FEA modules and artificial neural network (ANN)
modules in an intelligent knowledge-based system framework.
The ANN module is a preliminary design tool which indicates as process parameter values for lubrication, sheet thickness,
formability of the component geometry, for the selected process punch velocity, and blankholder pressure. Automatic selection
and material conditions. The ANN module is trained from of the design parameters is based on heuristic rules in a design
FEA results for a generic set of component geometries, process knowledge base. During the stamping process, defects such as
conditions, and material properties. ANNs are computational splitting, tearing, and wrinkling are encountered. Human experts
structures modelled on the basis of biological neural networks. apply their past experience to diagnose these defects and
The potential benefits of ANNs extend beyond the high compu- rectify them. Similarly, in the present hybrid intelligent systems
tation speeds provided by massive parallelism. ANNs are approach, heuristic rules for defect diagnosis and rectification
robust, fault tolerant, and have the ability to adapt and continue are stored. Similarly, selection of material properties is made
learning. An ANN works in parallel with the design process from the material knowledge base.
in an off-line mode (training phase). The fully trained network The geometric optimisation module assists in the detection
is then invoked when needed for the actual design process. and automatic rectification of forming defects such as splitting
This mode of off-line training of the network and using the failures. Case studies illustrating this procedure are included.
trained network for the actual design reduces the number of A comparison of the hybrid intelligent systems approach for
FEA-based design iterations required, leading to considerable die design optimisation with shop floor data is also discussed
savings in time and cost. FEA is used for final validation of the in this paper.
design. The intelligent knowledge based framework supports The hybrid intelligent systems approach demonstrates the
integrated simulation and analysis. The blackboard architecture capability of evaluation of die designs for manufacturability
is used for the development of the intelligent knowledge-based and performs automatic modification of various design para-
framework [11]. meters to obtain a defect-free design. Application of the hybrid
The hybrid intelligent systems approach for design incorpor- intelligent systems approach, reduces the total lead time for
ates rules for the selection of various design parameters such stamping die design and manufacturing.
372 R. Pilani et al.

2. Die Tryouts that supports both symbolic and numerical processing tech-
nology is required.
During the die tryouts the formability of the component is With the above requirements in mind, a hybrid intelligent
assessed by stamping trial components to investigate the sheet system for die design with the following objectives has been
material, process parameters, and die geometry combination. If developed:
any forming defects are detected, then, based on past experi- 1. Integrated design and simulation.
ence, the experts modify the initial design conditions, such as 2. Intelligent set-up and automatic modification of design para-
lubrication, drawbead shape, size, location, blankholder press- meters.
ure, etc. to enable the component to be formed successfully.
Common forming defects are splitting failures, tearing, wrink- 3. User friendly.
ling, etc. The die tryout process consists of the initial selection 4. Self-learning and generalising capability.
of various design parameters, forming the component, checking By the integration of design and simulation, the expensive and
the formed component, defect diagnosis, and then rectification, laborious task of fine-tuning the analysis model is reduced.
to eliminate any failures that might have occurred. Additionally, the effect of changing one design parameter may
Currently, for 80% of the time when forming defects are affect other parameters which requires accurate bookkeeping
detected, the experts rely on heuristic knowledge gained over of cause and effect which are taken care of in the hybrid
the years for die modification and design optimisation to avoid system. The incorporation of an ANN model provides a self-
the failures. Only in about 10% cases, do they seek the learning and generalisation capability.
assistance of FEA tools and design handbooks to rectify for- The application of intelligent tools to incorporate human
ming defects [1]. design expertise enhances the performance of analytical tools
In the automotive industry, CAD/CAM and FEA tools pro- by the automatic selection of appropriate factors for the follow-
vide separate support for geometrical design and analysis, but ing design parameters:
do not provide integrated design and simulation support, includ-
ing automatic set-up, execution and analysis. Additionally, the Analysis model: ANN or FEA.
need for heuristic knowledge during the die design process is Material model: voce or holloman hardening laws.
not considered in these systems. Lack of user friendliness and Sheet material: DQ, EDDQ, DDQ, IF, CQ, etc.
the complicated set-up and analysis of FEA systems are some Lubrication: high, low, or medium.
of the reasons for the lack of widespread applications of
Forming mode: draw, stretch or mixed mode of forming.
FEA [1].

3.2 System Architecture

3. Hybrid Intelligent System
The hybrid intelligent system developed in this work, intelligent
3.1 Objectives design environment (IDE) is based on an artificial intelligence
blackboard architecture. Blackboard systems are conceptual
Since the die design process is an iterative trial and error frameworks for integrating multiple independent programs, hav-
process, and the design knowledge and data are so complex, ing diversity in knowledge representations and opportunistic
there is a critical need for integrated systems which combine problem solving [11]. IDE incorporates groups of independent
various problem-solving technologies such as FEA, ANN, and modules consisting of ANN programs, FEA programs, and
KBS in an integrated design framework. Such a system needs KBS. The intercommunication between the programs is through
to possess an intelligent reasoning mechanism for automatic the common blackboard. An intelligent scheduler reviews the
selection of the appropriate problem-solving technology, and blackboard after each call to the knowledge source. Based on
the ability to schedule and process the results from the problem the success or the failure of the called and executed knowledge
solver, to achieve the desired design goals. source, the scheduler decides the agenda and plans the design
Hybrid intelligent systems are problem-solving systems that process dynamically. IDE incorporates various modules such
are based on the integration of several technologies where at as defect diagnosis module, geometry modification module, etc.
least one is an intelligent technology [2]. Hybrid systems are which are discussed in detail elsewhere [15,16].
available in areas such as architectural design [12], turbine The design methodology is shown in Fig. 1. Initially the
design [13], and aluminium alloy design [14]. These hybrid user develops the component geometry through IDE CAD,
knowledge-based systems integrate expert systems with numeri- after which the various process parameters such as lubrication,
cal methods and computer-aided design. These techniques pro- punch velocity, etc. are selected. Suitable material for forming
vide the accuracy needed for solving complex problems as is also selected by the hybrid intelligent system. The ANN
well as the reasoning needed for arriving at a solution quickly. module is then invoked which predicts the peak strain value
The development of such systems for sheet metal forming is for the given geometry. If the peak strain value exceeds the
difficult since the die design process is very complicated, and forming limit value for the selected material, then the rede-
the Human experts use various kinds of information at different signer module optimises various design parameters such as
times during the design process. The information could be lubrication or blankholder pressure to arrive at a design which
either numerical or symbolic. Hence, a hybrid intelligent system exhibits lower peak strain. In the next phase, the FEA module
 Hybrid Intelligent Systems for Die Design 373

is invoked which predicts strain distribution over the whole

component geometry. If forming defects such as splitting are
detected then, by optimisation of the process, and material
or/and geometry parameters, the defect is eliminated. The user
can override the modifications suggested by the system.

3.3 Knowledge Bases

The knowledge bases of the present hybrid system are similar

to the knowledge base developed in [13]. There are four main
knowledge bases:
Scheduler KB
Parameter KB
Analysis KB Fig. 2. Tool geometries: original, modified and optimised.
Optimisation KB
the initial conditions are very severe, and in which splitting
Scheduler knowledge base. This knowledge base consists of failure is detected. By performing automatic modifications in
rules for managing the selection of suitable knowledge sources, process, for material and geometrical parameters, the hybrid
the selection of suitable design plans and the execution of the system was able to obtain a defect-free die design. The second
plan. The plan specifies the sequence in which the knowledge case study illustrates the application of the hybrid design
sources are to be executed. system for the die design of an automotive panel section. To
model shop floor conditions, only process parameters and
Parameter knowledge base. The ANN and FEA programs and
geometrical modifications were permitted.
their input are determined through this module. The parameters
include material properties: n, m , r, K, etc. and process
parameters such as friction coefficient , punch velocity, blank- 5. Intelligent Die Design Optimisation
holder pressure. IDE attaches current values, minimum
The objective of this hypothetical case study is to show how
increment, default values, etc. to the various parameters. The
design optimisation as performed by human experts is simulated
material parameters are selected from the stored material data-
by IDE. The optimisation is related to the decrease in the peak
base.
strain values for a stretch formed component. The optimisation
parameters are material, processing conditions, and component
Analysis knowledge base. This knowledge base contains rules
geometry. Traditionally, such design iterations and optimisation
for the extraction of useful information from the output of the
are performed at the die tryout stage, where the die design is
various ANN/FEA analysis modules. This includes checking
optimised to prevent any forming defects such as tearing or
for various forming defects from the data. At present, only
splitting, occurring during manufacturing.
splitting failures are analysed.
The initial geometry of the component is shown in Fig. 2.
The material selected is commercial quality steel having poor
Optimisation knowledge base. Heuristic rules for modification
formability characteristics with a limit strain of about 0.20%.
and optimisation of the various design parameters to prevent Process parameters such as lubrication coefficients are also
failures are stored in this knowledge base. chosen in such a way that severe forming conditions are
induced. Figure 3 shows the strain distribution for this compo-
Geometry modification. During die tryouts, experts frequently
modify the die face geometry by polishing, spotting, and
smoothing the sharp corners. In the hybrid system, the
component geometry is translated into a B-spline represen-
tation form. B-spline representation is the most compact and
commonly used geometrical representation format in current
CAD/CAM/CAE systems.

4. Hybrid Intelligent System: Applications

Two case studies describing the application of the hybrid

design approach for an industrial automotive component are
given in the following sections. The first case study illustrates
the full range of automatic modifications performed by the
hybrid design system for a hypothetical component in which Fig. 3. Strain distribution: original, modified and optimised.
374 R. Pilani et al.

nent. The peak strain values exceeds 0.30, which is above the Table 1. Design variables and the predicted peak strain values.
forming limits for this material.
As the peak strain exceeds the forming limit for this material, Trial Design variables modified Peak strain
the scheduler KS executes the KS for design modification
asserting tearing failure at specific locations on the component. Original geometry 0.30
Change process conditions Increase lubrication 0.25
The design modification KS calls design rules for design Geometry optimise 1 Geometry at critical section 0.18
parameter modification relating to material and process para- Optimised geometry Geometry at critical section 0.14
meters.
After the change in lubrication condition and material proper-
ties, the peak strain value falls to 0.25 which is still higher
than the forming limit for this material. Note these design
modification rules are user modifiable and can be customised Table 2. Design variables and the predicted peak strain values.
for individual sites. In the next iteration the tool geometry is
modified within user specified limits, which is comparable to Trial Design variables modified Peak strain
the polishing of the sharp edges and corners of the die. This
rule is activated since there are no further modifications poss- Original geometry 0.38
ible for process or material conditions. Optimised geometry Geometry at critical 0.13
The modified geometry and resulting strain distribution is section
Shop floor data Polishing and smoothing 0.14
shown in Figs 2 and 3, respectively. Now, the peak strain is
below 20%, which is still in the marginal zone. The user can
go through one more round of design iteration for still further
reduction in peak strain values. After this design modification,
the peak strain value falls to 18% which is in the safe zone automotive inner panel is demonstrated. This component was
(Table 1). An added advantage of the last design iteration, is selected to illustrate the IDE capabilities in modelling the die
that during production, even if there is deterioration in process design process as carried out on the shop floor. Table 2
and material conditions, the safety margin makes the forming illustrates the reduction in peak strain achieved using IDE and
process more robust. the corresponding peak strain values from the shop floor
measurement.
The original and optimised component geometry is shown
6. Design Optimisation of an Automotive in Fig. 4. The material used is a deep drawing quality (DDQ)
steel of 0.64 mm thickness. The peak strain obtained was well
Component above the limit strain of 23% for this material (Fig. 5). By
using current design variable settings, splitting failure would
IDE was tested using shop floor data. For this case study, an have taken place. Hence, the scheduler opted for modifications
existing automotive inner panel section was selected, which in the geometry. Figure 5 shows the geometry after complete
satisfies plane–strain conditions approximately [17]. The orig- optimisation, at which stage the strain level attained was within
inal tooling was as designed, the tool shape is shown in Fig. the forming limits for the material.
4. This component, however, was found to fail by splitting at This case study shows the ability of the hybrid systems to
the critical section. By trial and error, die design experts suggest and make modifications automatically to critical design
modified the geometry to prevent failure in this component. In variables such as process, material, and geometrical parameters.
the section below, the use of IDE for optimising the above

Fig. 4. Die geometries: original, optimised tooling and shop floor

final data. Fig. 5. Strain distribution: original, optimised and shop floor final data.
 Hybrid Intelligent Systems for Die Design 375

7. IDE Implementation References

1. P. Fallbohmer, T. Altan, H.-K. Tonshoff and T. Nakagawa.

The current version of the hybrid intelligent system, known as “Survey of the die and mold manufacturing industry”, Journal of
Intelligent Design Environment is v1.5.1. IDE was developed Material Processing Technology 59, pp. 158–168, 1996.
using Prolog, C/C⫹⫹ and FORTRAN and runs on a DOS 2. G. R. Madey, J. Weinroth and V. Shah, “Hybrid Intelligent
platform. The blackboard system is implemented in Prolog. Systems: ‘Tools for decision making in intelligent manufacturing’”,
in Cihan H. Dagli (ed.) Artificial Neural Networks for Intelligent
IDE consists of around 5500 lines of Prolog code, 2500 lines Manufacturing, Chapman and Hall, London, p. 72, 1994.
of C/C⫹⫹ code and the FEA code is in FORTRAN. The 3. K. F. Reinschmidt and G. A. Finn, “Smarter computer aided
knowledge base consists of about 150 rules. design”, IEEE Expert, 23, p. 50, 1994.
4. D. Stacey, “Intelligent systems architecture: design techniques”,
in Artificial Neural Networks for Intelligent Manufacturing, Chap-
man and Hall, London, p. 229, 1995.
8. Summary 5. T. Altan, “Design and manufacturing of dies and molds”, Annals
CIRP, 36(2), p. 455, 1987.
6. A. Makinouchi, “Sheet metal forming simulation in industry”,
A hybrid intelligent systems approach for die design for sheet Journal of Materials Processing Technology 60, p. 19, 1996.
metal manufacturing has been developed. Using powerful arti- 7. Mustafa A. Ahmetoglu, Gary Kinzel, and Taylan Altan, “Computer
ficial intelligence techniques, the hybrid system integrates KBS, simulation for tool and process design in sheet forming”, Journal
FEA and ANN. The hybrid intelligent system consists of of Materials Processing Technology 49, p. 432, 1994.
8. A. Makinouchi and M. Kaska, “Process simulation in sheet metal
various modules as knowledge sources, and a scheduler linked forming”, Journal of Materials Processing Technology, 46,
through blackboard architecture. The knowledge sources are p. 303, 1994.
the individual expert systems, the FEA programs, ANN and 9. M. P. Sklad, “Modelling of part shape and deformation evolution
the graphics handlers. The hybrid approach has the following in conventional forming of large sheet metal components”, Pro-
ceedings of Numisheet 93, 3rd International Conference of 3D
capabilities: Sheet Metal Forming Process, Isehara, Japan, p. 295, 31 August–
1. Hybrid intelligent system integrating finite element analysis, 2 September 1993.
10. T. Nakagawa, “Recent manufacturing technologies for auto-body
artificial neural networks, and knowledge based systems. panel forming tools”, Journal of Materials Processing Technology,
2. Supports conceptual design, and rapid prototyping 46, pp. 277–290, 1994.
11. P. H. Nii. “Blackboard systems: The blackboard model of problem
3. Automatic evaluation of new designs quickly and accurately. solving and the evolution of blackboard architectures”, AI Maga-
4. 1. Auto modification and optimisation of geometry. zine, pp. 38–53, Summer 1996.
5. Auto process optimisation for the given component. 12. D. Sriram, R. Logcher, N. Groleau and J. Cherneff, “DICE: An
object oriented programming environment for cooperative engin-
6. Self learning capability. eering design”, Artificial Intelligence in Engineering Design, 3,
pp. 230–252, 1992.
The hybrid intelligent system improves the design quality 13. S. Tong, D. Powell and D. Cornett, “Engineous: A unified method
and reduces the set-up time. The hybrid system is shown to for design automation, optimization and integration”, Artificial
perform automatic setting up and modification of the input Intelligence in Engineering Design, 3, pp. 210–228, 1992.
process, material and geometry, for evaluating manufactur- 14. I. A. E. Hulthage, M. S. Fox, M. D. Rychener and M. L. Farinacci.
“Alladin: A knowledge based system for aluminium alloy design”.
ability. It is shown by the case studies that the hybrid systems IEEE Expert, pp. 56–73, August 1990.
approach optimises design by modifying automatically the 15. R. Palani, R. H. Wagoner and K. Narasimhan. “Intelligent design
punch velocity, drawbead force and geometry, blankholder environment: A knowledge based simulations approach for sheet
pressure and position, blank size, thickness, material properties, metal forming”, Journal of Materials Processing Technology 45,
pp. 703–708, 1994.
and friction, to eliminate the forming defect. 16. R. Palani and K. Narasimhan, “An integrated approach to design
in sheet metal forming”, in T. S. Mruthyunjaya (ed.), Proceedings
Acknowledgement of Conferences of Advances in Mechanical Engineering (ICAME),
Narosa Publishing House, New Delhi, India, pp. 637–650, 1995.
17. Y. T. Keum, C. Wang and R. H. Wagoner. Sheet-S v 2.0:
The research project is supported by UK-India Scientific Tech- Industrial Trials and User’s Guide, Ohio State University, Ohio,
nical Research Funding. USA, 1990.