Engineering with Computers (1994) 10: 245-257
©1994 Springer-Verlag
London Limited
Engineering
with
Computers
Correspondence and offprint
requests to: Ian
R. Grosse, Department of Mechanical Engineering, University
of Massachusetts, Amherst, MA 01003, USA.
Concurrent Iterative Design
and the Integration of Finite Element Analysis Results
Kaushik Sahu and Ian R. Grosse
Department of Mechanical Engineering,
University of Massachusetts, Massachusetts, USA
Abstract. To
help foster concurrency in engineering design, methodologies are
needed to support the interpretation and abstraction of numerical
simulation results for design modification purposes. A methodology
is presented in this paper for the integration of continuum-based
numerical simulations into a computational system for concurrent
design of mechanical components. The technique involves conversion
of low-level numerical data, such as that obtained by finite element
analysis, into high-level symbolic representations called form
templates. The form templates are based on rudimentary
features which carry meaningful qualitative descriptions abstracted
from either manufacturing and/or functional numerical simulations.
The form templates are superimposed on the design's primary representation
in the solid modeler to facilitate the cognitive mapping of concurrent
destructive or constructive solid modeling features for design
modifications. Examples are presented to demonstrate the methodology.
Keywords. Design
methodology; Iterative design; Concurrent design; Features; Form
templates; Finite element analysis
1. Introduction
While feature-based design systems are providing more and more intelligent support to CAD systems, we still need effective tools for integrating into the design process deep physical knowledge of the design's manufacturing process and/or the design's functional behavior. For real-world problems, this is only possible through the abstraction of knowledge from computer simulation results and using this knowledge for design modifications. For example, manufacturing process simulations can be exploited not only to optimize the manufacturing process but also to make major preliminary design changes [I 4].
However, modeling tools for either functional or manufacturing process simulation normally demand extensive domain knowledge and modeling specific expertise on the part of the user. The task of analysis is therefore shifted from the designer to the experienced analyst. This de-couples the design stage from the analysis stage and defeats the entire purpose of simultaneous or concurrent engineering. A truly friendly concurrent engineering computational environment should have the ability to interpret the results from numerical analyses, and then use these inferences for automating, suggesting, or facilitating design modifications. Reasoning based on analysis results can thereby augment the intelligent thinking capacity of feature-based design systems.
The methods used in this research
exploit deep science-based principles to assist design modification
activities. This extends the design automation task from dealing
with simple rule-based information modules to deeper knowledge-based
modules in the form of mathematical models and other numerical
algorithms. What is needed, therefore, is a mechanism to extract
meaningful qualitative design information from simulation results
and to couple this information to a design modification system
at a higher level of representation. This paper discusses suitable
representations and the transformation process for integrating
the cognitive, symbolic and numerical design knowledge in a computational
system for the preliminary design of mechanical components.
2. Related Work
Chalfan [5] discusses the
need for an integrated preliminary design tool and employs
expert system technology for achieving this goal. Recent research
trends have also shown features as having an important role in
an integrated CAE environment. However, the major thrust behind
the feature-based design tool has been to provide a complete product
definition to offer a better means for transferring information
246
K. Sahu and I. R. Grosse
between design and manufacturing systems [6-10]. Patel and McLeod [8] have further indicated the lack of feature information or usage in free-form manufacturing processes such as molding and casting. The design-by-manufacturability (DBM) approach offered by Grosse and Sahu [2-4] makes it possible to bring features into this domain. Since the method is largely dependent on manufacturing simulation results, design tools are needed that are more closely coupled to the domain-specific analysis modules.
Few attempts have been made at abstracting or filtering deep physical knowledge from numerical data, and then using this knowledge for making significant preliminary design changes. The use of deep knowledge is often restricted to the evaluation stage and hence is always identified with detailed design rather than with preliminary design. However, Hart and Rodriguez [I 1] have developed a system that proposes design modifications (i.e., finite element deletions) to the user, based on a structural analysis of the initial design.
Examples of other existing structural analysis assistants are SACON [12], FEASA [13], PLASHTRAN [14] and FACS [15]. FACS has been used to a certain extent in abstracting higher level design details of structural components for FE modeling purposes. These systems exploit knowledge only in the functional domain. Also, these systems lack compact heuristic databases and do not have the capability to interpret numerical simulation results. Because of representational inadequacies, the coupling between design and analysis modules in these systems is either non-existent (such as in SACON) or very weak.
It is argued that the up-front use of analysis results in general, and manufacturing simulation results in particular, can provide expert guidance to the design engineer during the early stages of design [1]. This has been a primary motivation in building coupled systems for design and analysis tools. Kitzmiller and Kowalik [16] have discussed the need for coupling symbolic and numerical computing in knowledge-based systems. Their paper is a discussion of issues related to numerical and symbolic computing techniques that were presented in a workshop. The authors emphasize the fact that neither purely knowledge-based systems nor numerical algorithms by themselves can successfully handle complex problems in design and analysis. They suggest a proper mix of both symbolic and numerical techniques as the key to a robust automated design environment. According to their definition, coupled systems link symbolic and numerical computing in a manner not found in conventional expert or knowledge based systems. Coupled systems should have some knowledge of the embedded numerical processes and reason about the application or results of those numerical processes. They have identified interesting issues related to coupled system architecture and other key factors necessary for designing a coupled system.
Several coupled systems in different application specific areas have been discussed by Jacobstein et al. [17]. One area is in the domain of design and analysis where the coupled system concept is applied to bridge the gap between design and analysis modules. Notable research in this particular area has been carried out earlier by General Electric researchers in their Engineous project [18]. Engineous was initially developed to optimize the design of aircraft engine compressors. Expert system techniques have been used in representing domain-specific knowledge which is then used for design problem solving. Engineous optimizes using classical numerical techniques and/or heuristic search methods. The knowledge represented supports program interdependencies that effect design parameters from a relational point of view. However, Engineous allows more of a parametric design optimization rather than preliminary design of components.
Shephard and Grice [19] in their report have cited the requirements for coupling geometric modelers with FE modelers. Kissil and Kamel [20] propose using expert system technology in finite element modeling applications. Shah [9] interprets this work as being dependent partially on feature information. He has discussed features as information sets that can well be used in integrating solid modelers and finite element modelers. The expert system in Kissil and Kamel's work depends on element biasing rules which make use of feature information sets. The feature information is therefore in the form of parametric ratios defining different topological patterns. Shah has indicated the use of form features only at the idealization stage. In this paper we will discuss the use of different representative forms which will follow the work of the above researchers in providing a better form of coupling between design and analysis modules. Features will be used primarily to perform design modifications.
Finally, Rosen et al. [21]
have addressed the problem of integration of CIM (computer integrated
manufacturing) functions through feature conversion. Product
data in this environment is shared through viewpoint-specific
feature-based representations. Thus, for example, features in
the designer's viewpoint need to be converted for relevancy from
an analyst's viewpoint. We share these ideas in our design metamorphosis
cycle where design topologies need to have a primary representation
in the solid modeler
Design and Integration of
Finite Element Results
247
Feature
Based
Modeler
(infommtion)
ometty
--------- - ----------
---------- - ----------
eep owiedge
Analysis
Fig. 1. Sharing
product information between design and analysis.
which subsequently helps in obtaining a secondary representation for manufacturing or functional simulations in the finite element modeler. The deep knowledge, abstracted from the numerical simulation results, is used for making design modifications in the feature-based modeler.
Feature-based systems, as
seen earlier in Nielsen [22], have the capability of associating
both form and intent for representing designs. Intent data is
mostly manifested in the form of either the functional or the
manufacturing requirements that are provided to constrain the
design evolution process. Thus, the coupling between design and
analysis can be through information comprising form, intent and
deep knowledge (see Fig. 1). The details of our approach are
discussed in the following sections.
3. The Problem Domain
The problem domain discussed in this paper pertains to the preliminary design domain of mechanical load-carrying components manufactured by the injection molding process. This domain presents a formidable research challenge because of the complexity of the manufacturing process and the potential geometric complexity of the design. Furthermore, plastic components may serve a variety of functional purposes. For simplicity, we will restrict the functional objective to be structural, although secondary functional objectives will be permitted.
Furthermore, the geometric complexity of the components in this domain presents some difficult problems for purely rule-driven computational systems for configuration design. It is therefore necessary to exploit deeper knowledge of the design and manufacturing processes along with some shallow heuristic rules. Since analytical solutions do not exist for these complex geometries, deep physical knowledge must be abstracted from the numerical data obtained from either functional and/or manufacturing simulations.
Accordingly, the broad objective
of this research has been to abstract deep knowledge from numerical
simulation results, and then, through the aid of an integrated
system, modify designs using a featurebased modeler [2-4]. The
initial designs could be
preparation
gate location
elem size
runner dia
inputcondn
Fig. 2. Finite element modeling complexity in the injection molding
demain.
either 'pre-existing' designs or to-be-determined'nonexisting' designs which are required to satisfy certain functional needs. The latter task of generating nonexisting designs is mainly carried out in the preliminary or configuration design stage, whereas treatment of pre-existing designs is similar to shape optimization methods. The details have been discussed in refs [2, 4]. In contrast to a conventional concurrent design approach, in our computational model new design configurations are generated primarily by the manufacturing simulation results with functionality playing more of a secondary role. It is argued that using deep physical knowledge of the manufacturing processs, abstracted from process simulations, to help generate alternative design configurations will in general enhance the design's inherent manufacturability. The authors have focused on the injection molding process where qualitative understanding of the mold-filling analysis results are primarily based on expert interpretations of the quantitative data.
Because of the geometric complexity
of the parts involved, the domain presents a formidable challenge
which is further complicated by the task of analysis. Figure
2 shows the complexities involved in the domain of in . ection
mold-filling simulation analysis. The complicated task of preparing
the model for finite element analysis for abstracting manufacturing
knowledge requires the designer to be proficient with various
injection molding issues. This provides an excellent example
of a currently decoupled design
248
and analysis environment, and hence is the target of our research.
The secondary objective is
to transform low-level numerical information into suitable higher
level representations necessary for performing intelligent design
modifications within an integrated computational environment.
The technique is discussed using examples from both the manufacturing
and the functional domain. The acronyms DBM and DBF are used
to represent two fundamentally different design methods, namely,
design-by-manufacturability and design-by-functionality. Both
DBM and DBF approaches are largely dependent on finite element
analysis results.
4. The Design Metamorphosis
The design procedure by Grosse and Sahu, which is based on either the manufacturing or functional strategies, is primarily iterative in nature [1, 2]. Thus, at the end of each iterative cycle, a configuration exists which is the result of feature-based design modifications on the parent design configuration. Within each iterative cycle there is a noticeable transformation in the design's representation as it progresses from the design module to the analysis module. This cyclic change in the design's representation characteristic is called the design metamorphosis.
The various design and analysis activities performed by the system are possible only through suitable representations. At this point we have identified three different representations needed to link the design and analysis modules successfully. A primary representation is needed to link the human designer's concept with a computer-based model in the form of a solid model representation with associated high-level design, analysis and manufacturing information. Both the geometric and augmented models in ref. [9] can be considered as a subset of this primary representation.
The secondary idealization,
obtained from the primary representation, links the solid modeler
to the analysis module. Like the primary representation, the
secondary representation can be considered as a super-set of both
the idealized and the discretized models shown in Shah's paper
[9]. The secondary representation is then subjected to either
a manufacturing and/or functional simulation to provide deep knowledge
which is normally in the form of raw numerical data. Since this
is the lowest level of information available to the designer in
this framework, it is designated as the tertiary representation.
Finally, a qualitative description is necessary to
K. Sahu and 1. R. Grosse
Intent
I
THIN SHELL
DESIGN BLANK
PF detive
EPRE b
(soli
1)
modiftl
QUALITATIVE abstmc
DESCRIPTION
Fig. 3. The
design metamorphosis.
link the deep design knowledge
contained in the tertiary representation back into the solid modeler
for design modifications. The design metamorphosis cycle is shown
in Fig. 3. Each representative design form will now be examined
in more detail.
4.1. Primary Representation
First, the design intent is
captured either automatically or interactively within an initial
form known as the design blank which is also the parent configuration.
As discussed in refs [1-4], the design blank is a geometric entity
heuristically constructed to include all geometric problem specifications.
This enveloping model has both form (or topology) and intent-specific
attributes. While the intent for a particular design problem
cannot be altered, topological modifications are normally unavoidable
from a preliminary optimization point of view. In this research,
topological modifications are based on deep knowledge interpretations
of the design's behavior under the influence of actual processing
conditions. All relevant distinct design configurations are then
obtained by appropriate geometric operations, usually subtractive
operations, on the design blank. Figure 4 shows an example where
toplogical modifications were performed for a certain set of intent-specific
data. Thus, while the intent remains the same in successive iterations,
the design form evolves during the iterative design process.
The dynamic nature of the design's form could, in general, impose
downstream problems related to finite element modeling. However,
by using a primary representation in the first place, such problems
can be easily averted, as discussed in the next section.
Design and Integration of
Finite Element Results
249
Fig. 4. Primary
representations of initial design blank and DBM and DBF child
designs.
CHI@DBM
The automated design blank generation capability provides a geometric shape given the intent-specific data. This method of blank formation restricts the initial design formative steps to an algorithmic procedure coded into the generator. Flexibility is also needed to permit the user to interactively generate designs in the solid modeler, and later to capture his design intent through a dialogue session. This information, i.e., geometric shape and intent, is necessary to provide full primary definition to the parent configuration. The intent-specific data, such as load and support point details in the functional domain, and injection molding gate location and other details in the manufacturing domain, serve as the critical design points within the design space during the design modification task.
In summary, a primary representation
of a design consists of a high level representation of its geometry,
such as that contained by a feature-based solid model, and associated
intent-specific information concerning both the design's function
and its manufacturing process. It should be noted that this definition
is sufficient for the scope of the design task considered in this
paper. If the design task is more comprehensive in nature, then
a richer primary design representation is needed. For example,
if the design task is to obtain a completely specified final design
fully optimized with respect to all aspects of the design, such
as cost, reliability, maintainability, serviceability, environmental
impact, aesthetics, etc., then the primary representation must
be extended to include explicit representation of design knowledge
associated with these design issues. This can be easily handled
in an object-oriented environment with the definition of new attribute
slots.
PARENT
LOAD VECTOR
)- DISPL. CONSTRLM
H HOT
R@
CHI@DBF
4.2. Secondary Representation
The next step in the design process is to exploit physical information related to the design's functional behavior (elastic deformation) or to the design's manufacturing process (injection molding) for improving the design blank. Since closed form solutions to these phenomena do not exist, numerical solutions must be sought. Sophisticated mold-filling simulations and optimization techniques may then be applied to optimize each candidate design configuration. Similarly, in a design-by-function approach we need to perform stress-based finite element analysis. However, because of software limitations, the solid models have to be pre-processed to obtain a finite element model suitable for analysis purposes. Since automatic meshing of solid models has not reached the level of sophistication required to represent part geometries for analysis in the application-specific domain, customization of the primary geometric representation is needed to facilitate successful integration.
A secondary representation
is therefore necessary to provide the knowledge source (i.e.,
the numerical solver) with a front-end link to the solid modeler.
In its simplest form, the secondary representation consists of
a complete and valid finite element model of the design's primary
representation for either functional or manufacturing simulations.
Thus, for the functional simulation (i.e., structural finite
element analysis), the static boundary conditions and loads are
imposed with the help of the intent data stored in the primary
representation to obtain the DBF secondary representation (see
Fig. 5(a)). For the manufacturing simulation (i.e., injection
molding simulation), the model is represented by the mid-plane
surface of
250
K. Sahu and I. R. Grosse
(a) DBF REPRESENTATION
(a) DEF PZPRESENTATION
DBM REPRESENTATIOW
(injection node at the functional
controid)
LOAD VECTOR
.L DISPL.
CONSTRAINT
I H HOT
RUNNER
(b) DBM REPRESENTATION
(injection node at the functional
controid)
parts and attributed with thin shell characteristics. The representative surfaces from the solid model are converted to mesh areas in the finite element module. Similarly, the DBM secondary representation is created by imposing the appropriate boundary condition. The boundary conditions in the DBM approach are created by locating the injection node and properly defining the hot runner dimensions. The placement of the injection node is based on heuristics [4]. Figure 5(b) shows the DBM secondary representation for an example problem. For simplicity, the rules embedded in the computational system are such that the same finite element mesh is constructed for the first DBF and DBM secondary representations, although this is by no means necessary or even desirable.
The information stored in
the primary representation helps in determining and imposing the
boundary conditions during the secondary characterization stage.
Since the system is dependent on the free meshing capabilities
of a commercial mesh generator, the global element size varies
from model to model. As a result, nodes cannot be anchored to
the critical design points. Therefore, a program has been written
Fig. 5. Secondary
representations for the design
blank.
Fig. 6. Secondary
representations for the
DBM-child.
to identify the nodes in the vicinity of the critical design points and distribute the loads and/or supports over the region. Figures 5, 6 and 7 show how the intent data is exploited in imposing the boundary conditions even as the topology changes from model to model.
Selection of the injection
node is also dependent on the intent data stored in the primary
representation. Even though the intent data are significant entirely
from a functional point of view, they are exploited by the DBM
approach in ascertaining the injection node location. Specifically,
the DBM representations shown in Figs 5, 6, and 7 have the injection
node located at the functional centroid. The functional centroid
is defined as the centroid of all relevant functional entities.
The coordinates of the centroid for a system of functional entities
fi with weights wi and occurring at points (xi, yi, zi) for
(i = 1, 2, . . . , n), is given by
i=n
Y-i -- I wj xi
i-n
Y-i=1 wj
assuming equal weights,
i.e, WI = W2
=... = W.,
Design and Integration of
Finite Element Results
251
Fig. 7. Secondary
representations for the
DBF-child.
we have
similarly
(a) DBF REPRESENTATION
(k)) DBM REPRESENTATION
(injection node at the functional
c*ntroid)
I
i=n
i=n
Y. = Y- yi
ni=l
I i=n
Z@ = Y zi
ni=l
The use of a functional centroid eliminates the need for process optimization at the end of each iterative cycle. This is a common drawback with the selection of gate locations at the geometric centroids which are sensitive to topological variations. Results indicate how the selection of gate location at the functional centroid also helps in generating good designs [4]. The design modification strategies employed in the research also show how the topology adapts itself to the choice of the initial gate location.
Model preparation for plastic flow simulation requires considerable experience and therefore deters the designer from using such analysis tools. A rule-driven approach is used here to guide the model preparation stage. The meshing entities such as element size and transitioning factors like dimensions of neighboring elements or the dimension between part and runner are controlled heuristically to generate the mesh [4]. The manufacturing requirements (obtained through manufacturing process plans) stored in the primary description impose restrictions on the gate location.
The design is meshed using
an automatic mesh generator. The finite element mesh, material
properties and appropriate boundary conditions are passed to a
finite element solver which computes a numerical solution for
the flow model of the injection molding process.
4.3. Tertiary Representation
The numerical solution consists of one or more sets of values of functional or manufacturing variables at specific points in the domain. In the case of functional simulations used for the DBF strategy, the solution consists of a set of stress components at the node points of the finite element model. In the DBM approach, the numerical simulation yields a variety of numerical information such as the shear stress and shear stress rate values at nodes at various times during the mold-filling process and the time to fill each node point.
While the numerical solution
is very important for abstracting deep functional or manufacturing
knowledge concerning the design, the numerical solution itself
represents the lowest level representation of the design. It
is merely an array of raw data which has little meaning until
it is associated with the secondary representation and transformed
into a qualitative description useful for design modification.
Thus, we refer to the numerical finite element solution as the
design's tertiary representation.
4.4. Qualitative Definiti6n
This numerical solution serves as a basis for modifying
the design blank, thereby obtaining new design
configurations. The raw data
obtained from the analyzer are transformed into packets of geometric
information with a qualitative description. This is accomplished
by the use of rudimentary features and form templates. Rudimentary
features are crude geometric shapes which enclose or approximately
enclose regions of the domain whose numerical solution is less
than a certain threshold value. For example, a rudimentary feature
may be a region of the domain formed by a group of nodes with
a stress value below a specific fraction of the mean stress.
The term ,rudimentary' is used here to indicate the lack of any
252
K. Sahu and 1. R. Grosse
@x
PARENT
FORM TEWLATE SUPER
ON THE SOLID MODEL
Library of Features
DB d
pre-defined shape or form for this feature. However, a rudimentary feature does contain qualitative information about the physical state of this region. In the case of the stress example discussed here, the rudimentary feature contains the qualitative information that this geometric portion of the domain exhibits an lunderstressed' physical state. Note that the quantitative numerical solution is not carried with the rudimentary feature. The rudimentary feature has a single qualitative description, such as being understressed.
The set of rudimentary features based on a design's finite element simulation is called a form template (see Fig. 8) and represents a 'blueprint' for modifying the design based on the qualitative abstraction of the numerical results. The form template is therefore a high-level design modification representation which consists of a set of points defining regions that are considered to have a homogeneous qualitative state (e.g., the state of being understressed). Such high-level definitions, when used to modify designs, render intelligence to the design tool. An example of such an operation is shown in Fig. 8.
The form template is used
to transfer this design modification information from the secondary
representation of the design to the primary representation where
sophisticated solid modeling tools are available for design modifications.
As the name implies, the form template provides the designer
with a template for
Fig. 8. Use
of form templates and cognitive feature mapping for design modifications.
performing cognitive design modifications. Because of the lack of automatic pattern matching capabilities, the user must perform cognitive mapping of features from a library onto the form templates displayed on the current configuration. Figures 8 and 9 show how the user can perform cognitive mapping of form features onto the highlighted form templates. Note that the user could have selected a different form feature from the library and performed the destructive geometric operation; or the user could even have used a number of hole features thereby, unknowingly, degrading the manufacturing quality of the parent design. Good design for manufacturability (DFM) rules are therefore necessary to prevent the designer from making such mistakes.
Qualitative association of engineering data with geometric entities is possible through abstraction of numerical data at discrete points within a continuum. The strategies used for data abstraction act as filters that provide meaningful qualitative information within quantized intervals. Logical operators are employed to define each strategy. However, not all strategies used here are based on deep knowledge. This means that the system has to rely partially on human experience as well.
Thus, qualitative understanding
of the numerical behavior is initially provided by human experts
and subsequently stored as rule-based strategies to be used by
the system. This allows the system to grow by the
Design and Integration of
Finite Element Results
253
Fig. 9. Cognitive mapping onto
the rudimentary features.
Rudimentary fe&ures hemng
homogeneous qualitative state definftions
addition of more and more
strategies. The following section describes the environment which
implements these ideas.
5. The CSN-Designer
A system, called Cognitive Symbolic and Numeric Designer (CSN-Designer), has been developed for assisting the designer in making intelligent manufacturing and functional based design changes. CSNDesigner contains encoded heuristic rules, which are based on domain-specific human experience, for near seamless integration of finite element analysis into a design environment.
CSN-Designer consists of a
CAD environment supported by a knowledge-based environment running
on a VAS station 3100 under VASJVMS v5.4-1. Figure 10 shows a
screen capture of the user interface offered by CSN-Designer.
The environment is built around the commercial package I-DEAS.*
The system is essentially menu driven (see Fig. I 1). One menu
set is within I-DEAS and is used to activate the various design
and analysis modules during the design modification task. This
auxiliary controlling menu, developed using the I-DEAS programming
language
* I-DEAS is an integrated
design and analysis software package developed and distributed
by Structural Dynamics Research Corporations.
capability, extends the CSN-Designer functions into I-DEAS. The other menu set is outside the commercial CAD environment and provides an interface, called UMDA (the UMASS Design Advisory), which allows the user to perform additional functions necessary for knowledge based design modifications. The user selects the appropriate menu items when prompted by the system. UMDA is developed using Common Lisp X.
The auxiliary controlling menu within I-DEAS, together with the main UMDA-menu outside I-DEAS, provide access to the various modules and processes (see Fig. I 1). Inter-process communication and coordination between the CAD environment and the external knowledge based environment are done through suitable files. The KB environment generates program files that are necessary for driving the modules within the CAD environment. In this particular case prg files are used for this purpose. The I-DEAS Command Language (with programmability capabilities) is used to control the various functions within the CAD package (whose menu interface is extended to provide additional relevant functions to facilitate the activities pertaining to analysis-based design methods).
The Lisp listener is activated
each time the user selects a menu item from the UMDA menu and
the controller takes over in controlling the corresponding module.
The different modules controlled through the UMDA menu are:
254
K. Sahu and 1. R. Grosse
I-DEAS
THE UMASS DESIGN AD% ISORY
(@UIT INMDDUCTION
BLANK OBJ-DATA
CONVERTER FE13CS
FORM-TEMF
f Y-pl.-. 2
0
3
T TAL 14WIBER OF PLANES 3
IIM ER OF LOAD$ TR X-DIR =@ 2
)FM[DER OF LOADS IR Y-DIR 2
IMMER OF LOADS IN Z-DIR 0
, nnMER OF SUPPORT-POINTB
4 HIII-QUITGC-QUITKIN-IN
@ACTIO"
(INTO S 1) MODELER FOR lt4TERACTIVE
EXECUTE)
. . . . . . . . .F...................................
I
(RUNS DESIG@BLANK.PRG)
tll QUIT => Exits program
t2l INTRODUCTION =:> Provides help information
t3l BLANK =>
Generates the design
f 31 BLANK => Generates the design
blank automatically
141 OBJ-DATA => Provides the primary-
representation
f 51 CONVERTER => Converts
the primary
rep to a secondary rep
141 OBJ-DATA
UMDA
{61 FEBCS
f7l SET-TEMP
{81 FORM-TEMP
{91 MOLDFILL
Fig. 10. A screen capture of the
CSN-Designer interface.
Fig. 11. The menu interface for
the CSN-Dcsigner.
=> Provides the boundary conditions to the FEM => Initializes Params for the transfer of form temp => Creates the form
templates
=> Activates the moldfilling
analysis solver
Design and Integration of
Finite Element Results
255
Fig. 12. Analysis for design
(DBF example).
PAREN . T
DBF-Child
. PRIMARY REPRESENTATION
OUD OR FEATURE -BASED MODELE
I
I
FINITE ELE
I
I SOLVE
I FEATURE BASED DESTRUCTIVE -
i_x 1 SOUD GEOMETRY OPERATlOt4
I [SOUD OR FEATURE-BASED
MODELER]
I
COGNITIVE MAPPING OF A DESIGN
MODIFICATION FEATURE ONTO A RUDIMENTARY FEATURE (SOLID OR FEATURE-
BASED MODELER)
I
I
I
TEMPLATE QUAUTAWE
FORM
I
EXTRACTOR TEMPLATES
The moldfill window waits for the mold-filling simulation to start when requested by the user. The user is prompted to perform the various functions from time to time. The control structure is either built into the Lisp code that performs the various UMDA tasks, or it is written using the I-DEAS command language that controls the various operations within the CAD/ CAE package and prompts the user to perform the external operations when necessary.
The automatic design blank,
generator consists of the UMDA-design-blank module and generates
a designblank.prg file that is executed from the automatic option
of the auxiliary menu within I-DEAS. The pre-existing design
customizer has the primary rep stored in the obj-data modules
and uses the interactive option in the CAD environment. The alternative
design generator consists of the UMDA-set-temp and UMDA-form-temp
modules and uses the knowledge extractor within the CAD environment
to perform design modifications.
; @ihi f ehipy@ iG 9c;
ON THE SOUD MODEL
---------------------
Some of the major system capabilities
are:
· Automatic generation of the initial design blank
· Primary representation of the initial design
· Automatic creation of the secondary representation both for manufacturing and functional simulation (little user interaction)
· Automatic generation of the initial gate location
· Automatic imposition
of boundary conditions for both manufacturing and functional simulation.
6. Example
As discussed in the previous
sections, the design modification task is dependent on various
modules, such as the feature-based and/or solid modeler, finite
element modeler, finite element solver and the finite element
post-processor. The complete iterative design loop is illustrated
in Fig. 12 for a specific design
256
K. Sahu and I. R. Grosse
example where the finite element analysis is based on the design's functional behavior. The figure shows how the design metamorphosis takes place as the design progresses through the different design and analysis modules. CSN-Designer integrates the various modules through suitable design representations, consequently making it feasible to use analysis for design.
The iterative design loop begins with the primary representation. In Fig. 12, the initial primary representation, i.e., the parent design, was automatically created by CSN-Designer based on the functional specifications for the design problem. For example, CSN-Designer invokes a set of rules to automatically create the design's secondary representation (i.e., finite element model) from the design's primary representation. This involves not only creation of the finite element mesh but also the specification of material properties, loading conditions and boundary conditions based on the intent data stored in the design's primary representation. The numerical results of the finite element analysis are then used to form rudimentary features and the form template which is displayed on the design in the solid modeler. The user then performs design modification tasks in the solid modeler through cognitive mapping of features from the feature library onto the superimposed form templates. CSN-Designer therefore functions as an intelligent interactive modeling system which is directly dependent on human interaction for such modeling tasks which are inherently cognitive and not amenable to automation, while automating much of the tedious modeling and analysis activities within the design metamorphosis cycle.
For the design-by-manufacturing
approach, manufacturing process simulations serve as a basis for
modifying the design in the iterative design loop to improve the
design's manufacturability. For the injection molding domain
studied here, the secondary representation created by CSN-Designer
must include various injection molding data needed by the moldfilling
simulation software, such as runner size, injection pressure or
fill time, etc. Rules, which represent compiled knowledge, are
used to construct this secondary representation automatically
from the primary representation. The DBM approach at the end
of the first design iteration loop yields the DBM child shown
in Fig. 4.
7. Conclusion
A methodology has been discussed
for transforming finite element results to support progressive
design modifications. Suitable representations have been identified
for ensuring successful coupling between the design and analysis
modules. These representations make it possible to augment the
feature-based design tool with deep knowledge-based modules.
The proposed methodology has been realized in CSNDesigner, a
computational system which integrates numerical (finite element),
as well as compiled and cognitive sources of knowledge, for preliminary
design of injection molded components.
References
1. Grosse, I.R.; Sahu, K. (1989) A design by manufacturability approach for computer-aided configuration design, Preprints of NSF Engineering Design Research Conference, University of Massachusetts, Amherst, June 11-14
2. Grosse, I.R.; Sahu, K. (1990) Extending the manufacturability approach to 3D configuration designs, Proceedings of the 1990 ASME International Computers in Engineering Conference, Boston, MA, Vol. 1, August, 25-32
3. Sahu, K.; Grosse, I.R. (1991) A manufacturing process driven approach for preliminary design, Proceeding 17th NSF Design and Manufacturing Systems Grantees Conference, Janaury
4. Sahu, K. (1991) A manufacturing based approach to preliminary design using coupled symbolic and numerical computing techniques, PhD Thesis, Dept of Mechanical Engineering, University of Massachusetts, Amherst, MA
5. Chatfan, K.M. (1986) An expert system for design analysis, Coupling Symbolic and Numerical Computing in Expert Systems (Kowalik, J.S., Editor), North-Holland, Amsterdam, 179-190
6. Dixon, J.R.; Cunningham, J.I.; Simmons, M.K. (1987) Research in designing with features, IFIP WG 5.2 Workshop on Intelligent CAD, Cambridge, MA, October 6-7
7. Dixon, J.R. (1988) Designing with features: building manufacturing knowledge into more intelligent CAD systems, ASME Manufacturing International'88, Atlanta, Georgia, April 17-20
8. Patel, R.M.; McLeod, A.J. (1988) Engineering feature description in mechanical engineering design, Computer-Aided Engineering Journal, October, 180-183
9. Shah, J.J. (1988) Feature transformations between applicationspecific feature spaces, Computer-Aided Engineering Journal, December, 247 255
10. Requicha, A.A.G.; Vandenbrande, J.H. (1989) Form features for mechanical design and manufacturing, Proceedings of the 1989 ASME International Computers in Engineering Conference, Anaheim, California, 47-52
11. Hart, P.K.; Rodriguez, J. (1989) A hybrid knowledge-based expert system for design of structural elements, Proceedings of the 1989 ASME International Computers in Engineering Conference, Anaheim, California, August, 287-292
12. Bennett J,, et al. (1978) SACON: a knowledge-based consultant for structural analysis, Technical Report STAN-CS-78-699, Stanford Heuristic Programming Project
13. Taig, I.C. (1986) Expert aids to finite element system applications, Applications of Artificial Intelligence to Engineering Problems (Sriram, D. and Adey, R., Editor), SpringerVerlag, New York
14. Cagan, J.; Genberg, V.
(1987) PLASHTRAN: an expert consultant on two-dimensional finite
element modeling techniques, Engineering with Computers, Vol.
2, No. 4, 199-209
Design and Integration of
Finite Element Results
257
15. Gregory, B.L.; Shephard, M.S. (1986) Design of a knowledge based system to convert airframe geometric models to structural models, Expert Systems in Civil Engineering (Kostem, C.N. and Maher, M.L., Editors), ASCE
16. Kitzmiller, C.T.; Kowalik, J.S. (1986) Symbolic and numerical computing in knowledge-based systems, Coupling Symbolic and Numerical Computing in Expert Systems (Kowalik, J.S., Editor), North-Holland, Amsterdam, 3-17
17. Jacobstein, N.; Kitzmiller, C.T.; Kowalik, J.S. (1988) Integrating symbolic and numeric methods in knowledge-based systems: current status, future prospects, driving events, Coupling Symbolic and Numerical Computing in Expert Systems 11 (Kowalik, J.S. and Kitzmiller, C.T., Editors), North-Holland, Amsterdam, 3-11
18. Nicklaus, D.J.; Overton,
K.S.; Tong, S.S.; Russo, C.J. (1988) Knowledge representation
and technique for engineering design
automation, Coupling Symbolic and Numerical Computing in Expert Systems 11 (Kowalik, J.S. and Kitzmiller, C.T., Editors), North-Holland, Amsterdam, 67-76
19. Shephard, M.; Grice, K. (1988) The coupling of geometric modeling systems with finite element modeling, Report R-88-GM-01.1, CAM-1, Arlington, TX, April
20. Kissil, A.; Kamel, H.A. (1986) An expert system finite element modeler, Proceedings of the ASME Winter Annual Meeting
21. Rosen, D.W.; Dixon, J.R.; Dong, X. (1991) A methodology for conversions of feature-based representations, Proceedings of the Design Theory and Methodology - DTM '91 - DE Vol. 31, ASME, 45-51
22. Nielsen, E.H. (1990) Designing
mechanical components with features, PhD Thesis, University of
Massachusetts, Amherst, MA