Professional Documents
Culture Documents
Design Engineering-A Need To Rethink The Solution Using Knowledge Based Engineering
Design Engineering-A Need To Rethink The Solution Using Knowledge Based Engineering
Design Engineering-A Need To Rethink The Solution Using Knowledge Based Engineering
Abstract
This paper discusses the current limitations of Computer Aided Design (CAD) tools and reports on the use of knowledge Based
Engineering (KBE) in the creation of a concept development tool, to organise information flow and as an architecture for the effective
implementation of rapid design solutions. The KBE tool along with supporting analytical solutions has been applied to the Body-In-White
area of automotive design. The present methods of using CAD and Finite Element Analysis (FEA) systems do not use a unified product/
process model representation and lead to the creation of separate non-relation data models that only capture the result of the engineering
process. The KBE method unifies the engineering intent into a single model that allows for existing or novel design solutions to be assessed.
These design solutions can then represent themselves in the correct form to the analysis systems. Automeshing is achieved using a rule-base
that meshes the model with respect to the analysis solution required, materials and processes. q 1999 Elsevier Science B.V. All rights
reserved.
Keywords: Knowledge based engineering; Computer aided design; Object oriented; Body-in-white
1. Introduction: the designers dilemma of the whole product cycle, are not readily made available or
captured within the design model. Competitive business
Engineering design is a multi-disciplined environment, a time-scales constrain designers to make critical decisions
highly integrated and integrating process. Designers juggle without exploring alternative strategies. The designer then
with many, and often conflicting, constraints to balance constrains the manufacturing engineers, limiting their abil-
aesthetics, manufacturability and functional objectives ity to improve the process. Decisions taken at the early
within a marketable product specification. The designers design stages may not be evident as having a negative
must transcend their perceived view of the world, to impact on the downstream process until the product model
shrug-off the narrow boundaries of specialisation, while is complete. Design decisions are made continuously during
exploiting specialised technical problem solving skills and the product development cycle. In fact it has been suggested
be open to external information at all stages of the design that designers make million dollar decisions every minute
cycle. This requires the integration and utilisation of infor- without ever knowing it [1]. Early decisions can determine
mation, supplied from many sources both internal and exter- almost 80% of the product costs (determined costs), at a
nal and in many formats. The information is constantly stage where knowledge about the product, customer and
refined by negotiations, clarifications, discussions and the processes involved is low or vague, and the actual devel-
evaluations, until a optimised or compromised solution is opment costs (incurred costs) are low [2] (Fig. 1). Making
agreed. The success of a traditional design project is deter- changes to early decisions may necessitate a great repetition
mined in two stages, by the effective communication of of the design processes steps and incur costs that increase
information relevant to a productive solution and then the strongly as we step through the product development cycle
capturing of the resultant solution or solutions. This iterative [3] (Fig. 2).
communication cycle with the results captured has inherent Concurrent Engineering (CE) allows the engineering
limitations. Designs can be produced that cannot be manu- team to utilise the varied inputs, knowledge and technology
factured. The manufacturing experiences, or the experience to speed up product development by integrating down-
stream concerns as early as possible in the design process,
* Corresponding author. Tel.: 1 44-1203-524721. by performing simultaneously many activities that used to
E-mail address: c.b.chapman@warwick.ac.uk (C.B. Chapman) be performed in sequence. For CE to work the agents in the
0950-7051/99/$ - see front matter q 1999 Elsevier Science B.V. All rights reserved.
PII: S0950-705 1(99)00013-1
258 C.B. Chapman, M. Pinfold / Knowledge-Based Systems 12 (1999) 257–267
2.1. Body-in-white
Table 1
Structural performance Table ULSAB [6]
Structural performance Current average Future reference ULSAB target ULSAB result
and safety legalisation. The Partnership for a New Genera- development of a vehicle body structure using hybrid
tion of Vehicles (PNGV) is an American government materials (Fig. 5).
initiative that aims to provide the framework to create
It is the second objective described within this paper, the
vehicles that have a target fuel consumption of 54.6 mpg.
creation of a Design Analysis Response Tool (DART) using
To achieve this the reduction of the overall vehicle weight
a Knowledge based engineering system (KBE). The first
by 40% is the key. [5]. Other initiatives such as the Ultra
objective will feed the second with rules and guidelines to
Lightweight Steel Auto Body (ULSAB) [6] project aim to
fuel the KBE system. The DART system has been applied to
improve vehicle characteristics, see Table 1, to compete
the automotive BIW area.
with lighter weight materials such as aluminium and
composites increasing in use within the automotive
industry. 3. The DART system
Design analysis tools are needed for a much wider range
of materials and constructions than hitherto. More design To understand DART, we shall look at KBE, stepping
effort and time will be needed to ensure optimum design though the development cycle and then the implementation
from this wider range of options (Fig. 3). This in turn used in the creation of the DART KBE system.
requires a faster method for concept structure design in
order to avoid the lengthening design and development 3.1. Knowledge based engineering
programmes. The conventional concept design route must
be cut short using more intelligent, automatic design KBE is an engineering method that represents a merging
development to take advantage of the established knowl- of object oriented programming (OOP), Artificial Intelli-
edge. gence (AI) techniques and computer-aided design technol-
The objectives of this research project are twofold: ogies, giving benefit to customised or variant design
automation solutions. The KBE systems aim to capture
1. To develop design analysis tools and methodologies for product and process information in such a way as to allow
the structural simulation of new materials (specifically businesses to model engineering design processes, and then
composites), manufacturing methods and assembly use the model to automate all or part of the process. The
methods for stiffness, strength and impact (Fig. 4). emphasis is on providing, informational complete product
2. To develop a fast concept design tool which will use representations, captured in a product model. The product
the existing knowledge to guide and facilitate the model represents the engineering intent behind the
product design, storing the how, why and what of a design. There have been many development methodologies
The product model is an internal computer representation of suggested for the Knowledge Based System (KBS) domain.
the product design process and can contain information on The methodologies have been aimed at assisting the devel-
both the product and processes that go to create the part. opers to define and model the problem in question. Methods
Attributes can describe geometry, functional constraints, such as Structured Analysis and Generation of Expert
material type and processes such as the methods required Systems (STAGES) and Knowledge Acquisition Documen-
to analyse, manufacture and cost a part. The KBE product tation System (KADS) (an acronym that has been redefined
model can also use information outside its product model many times, e.g. Knowledge Acquisition Documentation
environment such as databases and external company System and Knowledge-based system Analysis and Design
programs. The ultimate goal of the KBE system should be Support) [8]. These methods have normally been applied to
to capture the best design practices and engineering exper- areas other than the engineering design area, and have not
tise into a corporate knowledge base (Fig. 6). The KBE been commercially used for KBE development, where
methodology should provide an open framework for systems tend to follow a traditional design life cycle pattern.
formally capturing and defining the process of design Mulvenna [9] suggest that the KBS needs to be object-
creation [7]. oriented in nature to support an iterative prototyping area
such as design, and that KADS does not support this type of
3.2. Developing with KBE environment. Church [10] at the keynote speech for the
British Computer Societies Specialist Group on Expert
To develop a KBE system we need to first acquire, repre- Systems in 1993, recognised the fact that such methods
sent, reason and then communicate the intent of the design proved difficult for the dedicated Expert System (ES) prac-
process, Developing a KBE system is similar in nature to titioner to use and that this would hinder development by
developing a solution in the design environment. The non-dedicated personnel within companies using a multi-
problem is first understood at a conceptual level, then skilled workforce. The authors concur with this statement
decomposed into understandable working objects, devel- about KBS and would add that the KBE technology has the
oped further through an iterative process until a satisfactory ability for widespread growth within the design environ-
outcome has been reached. ment as the technology is developed along the methods
User Interface parts the style, packaging, structural members, joints and
panels. These parts then decompose further, each stage of
decomposition varying the intelligence of the objects. For
example primitive objects might only know how to draw
themselves, whereas higher level objects will use the knowl-
edge base to make specific decisions regarding their shape,
relationships, manufacturing process and cost.
Part object terminology changes depending on the appli-
Part Libraries cation domain. As most KBE systems are based on OOP
Conceptual Product Model methodology, they solve problems by defining, creating and
manipulating collections of data and procedures called
Design Output
objects. The terminology and concepts behind OOP and
Solid/Surface FEA M esh / output Decks KBE relate to how objects are defined, have properties
M odels
Drawings
assigned to them, how they interact and how they are
NC Code
combined to form more complex objects.
M anufacturing Plan Product / Process Reports
The object encapsulates the data or properties, and the
methods to manipulate that data into a single unit. The class
Fig. 8. Architecture of a typical KBE application.
is the template, and the object is specific case, (Fig. 11). An
object with a particular set of values is called an instance, of
of this design will be assessed using finite element simula- a class definition.
tion, for which the system will generate appropriate finite When values change that are used by the object to calcu-
element models and analysis input data. For the evaluation late properties or as parameters for method procedures, the
of the effects of modifications on the structural performance object is re-evaluated. This is called demand driven compu-
the system will revise the design in accordance with any tation and is a fundamental difference between OOP and
geometric changes which the designer makes, and generate conventional procedural programming. It is typical that
updated analysis input to re-evaluate its structural perfor- any change made to a procedural language or CAD/CAM
mance within a timescale needed to support the design deci- macro language, will result in re-computing everything,
sion processes. The tools will be developed for use by since the computations are performed in sequence. In a
structural analysts and body concept engineers, see Fig. 10. procedural language, a value is computed when the proce-
After stating the problem and identifying the information dure for computing it is encountered in the sequence. In the
required by the model, key objects were created, to form a KBE systems, values are computed only when required.
parts library. As each of the part objects were defined they This technique is called the dependency backtracking. For
were individually tested, by creating an instance of the example if the topology changes, the model will not be re-
object. Part objects created within the KBE system form computed until a request is made to draw the object on
the building blocks of any model, often representing real screen, or a calculation for the changed objects mass
world objects. The DART model, breaks down into six main properties are made.
The method of creating objects is very similar in most means the decomposition of information into levels of
commercial KBE systems. The Adaptive Modelling increasing detail [15]. The decisions made by the system
language (AML) from TechnoSoft [14] used within this to create a model will be based on the user inputs and the
project has the ability to create objects and write rules knowledge bases.
from templates that assist the designer in writing the Using the initial part library and conceptual product
programme. The object or define-class form is used to define model as a starting point, a subset of the overall system is
new classes. In any define-class, the following may be defined, complete with user interfaces (UI). The system is
specified: extended by increasing the part library, UI and by expanding
the object class definitions. The incremental development
1. Other classes which the new class should inherit from
approach means that the system can be continuously eval-
(superclasses).
uated and utilised, as it is always operational (Fig. 13).
2. The properties for the new class and their formulas (attri-
butes).
3. The sub-objects of the new class (children).
4. Current status
The next step was to create an initial conceptual product
model. This is not a complete representation, but acted as a The DART project has currently been through three
first draft, a blueprint of the implementation. The conceptual prototype models and is now in the final development
product model can be thought of as a schematic of the stage after the prototype models were used to promote
completed system. A particular design instance of a product discussion and as a training aid to the developer. Fig. 14
model is described by a product model tree (Fig. 12). This shows the initial prototype and training aid, Fig. 15 shows
structure describes the hierarchical relationships between the agreed UI layout containing a concept layout, sketched
the various components (parts and processes) of the around the imported surface style model and the resulting
model. The model also determines the methods of how basic structure definition. All the solid and surface geometry
the objects will obtain the correct information, in order to requirements for the structural members and panels have
carry out their specific tasks and as to the information they been coded into part objects. The next interactive require-
will provide to other objects. This hierarchical abstraction ments are the selection and application of the joint members
into the overall structural model. The solid model has the
ability to be altered through the graphical interface or
Object Name through altering the non-geometrical properties that
constrain the BIW process model. As this is a conceptual
Property 1 model any change is allowed, although if a working “Best
Property 2 Practice” rule is violated the system will inform the current
…………
Value 1 user and audit the change. Taking the structural body model
Property n
Value 2 into the analysis phase has been proven by automeshing the
Location
……… solid model within the KBE system. The meshing is done
Visualisation
Value n with respect to heuristic, material and analysis solution rule
Class Definition
Vector bases at an object level. If a structure is altered, for example
Draw Method a beam section the mesh automatically updates itself. Prior
Object to the mesh being created the rule bases are checked to see if
the structures material and manufacturing process are still
Fig. 11. Class and object relationship. maintained (Fig. 16). If the part cannot be created in reality
264 C.B. Chapman, M. Pinfold / Knowledge-Based Systems 12 (1999) 257–267
C ar
......
then there is no point analysing it, unless we are trying The next stage was to create part objects in a conceptual
something outside the current “Best Practices”, at this model that satisfied the results of stage one.
point the engineer can override the system. Sharing knowledge: KBE can utilise the stored knowl-
edge, making it available at design time to anyone at any
stage of the product cycle. We have already stated that
incompatible systems exist throughout the design process,
5. Summary KBE can store knowledge within its own structured “smart
parts” or can be used as an interface to bring together the
This paper looks at our present CAD tools with respect to incompatible data sets, transforming their representation
the design process and reports on the first phase of a KBE into one that can be used in an automatic fashion. The
tool for the creation of the BIW structures. The DART DART system uses process knowledge within its object
system aims to overcome the limitations imposed by the structure and also uses external information such as a target
computer centred approach in today’s design analysis vehicle rule database to populate the structural model. The
process by creating a concurrent unified modelling environ- ability to populate the object model with external data was
ment. The first stage in the DART development was to critical in this development as the rules resulting from other
understanding the BIW process. The mapping of this research sources were being conducted in parallel with the
process and subsequent acquisition of rules and relation- KBE system.
ships was achieved by the use of standard CE methods. Smart parts: The KBE systems allow the engineer to
By bringing all the people involved in the product life create what are sometimes termed as “smart parts”. These
cycle together and utilising their specialist knowledge to parts are created in the KBE system as objects. The parts can
come to a common solution, through negotiation and have rules and methods applied to them, rules such as cost-
compromise. Costs, functionality, materials, manufacturing ing rules, stress rules and process rules. With the advent of
and processes are all considered. This result of the group OOP the capability to build and manipulate parts within a
knowledge was then captured as a formal specification on KBE system is in tune with the real world situations.
paper. To create an integrated solution where this knowl- Objects with knowledge are manipulated in a non-proce-
edge can be automatically processed it was necessary to dural way that is particularly adept at modelling the design
capture intent behind the results in a computer system. engineering process. Object Oriented (OO) technology is
forgiving, accepting that we cannot get it correct the first
time and provides an easy mechanism for quickly changing
and modifying the conceptual product model. The object
structure has been developed with reuse in mind, this way
future systems can be built rapidly. The objects are built
using a programming language supplied with the KBE
system and require the engineer to learn programming tech-
niques and the particular language syntax. Although creat-
ing this product/process model in a KBE system was
relatively easy, this was due to the author’s background as
a designer, where decomposition and relationships are
understood. The KBE vendors should not just concentrate
on their particular KBE software when training new users,
but treat KBE as a methodology and also give an under-
standing of the philosophy of OO techniques. The creation
of a system is normally done using evolutionary program-
Fig. 13. Incremental development of application. ming techniques, but before mass acceptance by the
C.B. Chapman, M. Pinfold / Knowledge-Based Systems 12 (1999) 257–267 265
engineering community more research into the management community it should offer Computer Aided Software Engi-
and planning of the KBE systems should be encouraged. neering (CASE) type visual programming aids. The engi-
The main drawback of creating a KBE project is the exten- neers should be allowed to express themselves and not be
sive development time scales, due to the training of engi- bogged down in syntax.
neers to become users of a programming tool. The KBE Modelling and representing the results (views of the
vendors offer consultancy to assist at this stage, but this is world): The traditional way to create a model was to take
expensive. If a software tool is aimed at the engineering a functional specification and give it to a design engineer
who would draw the model in a CAD system using their key issue was geometry processing, agreeing that data
experience to interpret the specification. As happens all too exchange using an international standard was not the solu-
often within industry design models are duplicated, not due tion. The DART system overcomes these limitations by
to integration issues, but due to the fact that the different having the model transform itself automatically depending
engineering disciplines have a different yet valid view of the upon on which process is required. This was possible
world. because of the dynamic nature of the object-oriented repre-
The KBE systems allow for easily customised UIs to be sentation of the design process. Thus a material change,
created and the product model created by the rule base to be such as a change from a steel member to aluminium,
presented to the user in various ways. There are two main would entirely restructure the design and design process.
“views” of the product process required in DART, a full This method of generative modelling allows the designer
product representation and a FEA representation supporting to use the KBE model and rapidly perform a series of
both stiffness/strength and impact modelling. The design “what-if” analyses during the concept phase, allowing a
model is a complete representation. The analyst normally number of alternative design solutions to be considered.
needs to remodel the initial design. This will either be done The preparation of the model for FEA has shown that a
by editing the designer’s CAD data or by creating a comple- typical BIW structure of 250,000 elements can entail 15
tely new model within the analysis software. man weeks of effort upon receipt of the CAD model. The
Progressive CAD and analysis software companies have time taken and cost incurred in this effort has meant that
recognised the need for integration and have either inte- typically the analysis models are often used in a post design
grated with other products or have modelling and analysis phase to evaluate a final design that will only be modified if
software within one system environment. This, however, the results are found to be unacceptable [17]. The DART
does not go far enough and may well lead to rework. A system using an automatic analysis model based on an
design system must have within it the ability to represent understanding of the processes and materials will deliver a
a single product model as various alternative engineering prepared model in minutes to the analytical solution tools.
views. There is a great deal of confusion in this area. As This will allow the body engineers to try more structural
CAD systems create a geometric model and through data combinations in a greatly shorter period of time, as the
transfer or direct integration pass the model to other models are all created from a unified model description
systems, the CAD vendors suggest that the design/analysis duplication and the subsequent management costs will be
modelling route is automated. From understanding the eliminated. Downstream processes will benefit from having
design/analysis process we can see that there is a stage available to them 3D solid geometry models and not meshed
hardly discussed, that of simplification or model transforma- models as they have now. The models will also have the
tion. This step is critical for the engineering process and benefit of being known at concept time to be able to be made
often creates extra models that have no associatively. The reducing costly downstream change orders.
SANDIA Meshing Round Table 1996 [16] stated that auto- The final stages in the DART system will be completed at
mation of this design/analysis loop is hindered by the trans- the end of the first quarter of 1999, where a technical system
formation process being a judgmental operation and that the description and results will be made available.
C.B. Chapman, M. Pinfold / Knowledge-Based Systems 12 (1999) 257–267 267