Knowledge Based System Approach for Concurrent Engineering of PM Components

The importance of the technological, and especially ecological advantages offered by PM are well known to those working in the PM field and related industries. Despite these advantages, PM continues to be under-utilised. Application of Knowledge Based Techniques offers potential for enabling increases in the utilisation of PM, through implementation of Concurrent Engineering methods, such as design for manufacture. A Knowledge Based System (KBS), can be employed to process the empirical and heuristic knowledge held by manufacturers, and to combine it with information from theoretical and numerical process models. Through enhancements in knowledge quality and system sophistication, it should be possible to develop systems that are of practical use for advising on design optimisation, material selection, and process control. This paper describes preliminary work which has the goal of establishing methodologies suitable for development of industrially applicable KBS’s for PM. Additionally, directions for further research and system development, are identified. Requirements for Concurrent Engineering of PM Parts A thorough implementation of Concurrent Engineering (CE) requires information relating to all the stages of the product life cycle to be made available to Engineers during all the critical stages of product design and manufacture. However, by focusing on the product design stage (where most of the manufacturing costs are determined), it is possible to demonstrate how computing techniques can be employed to provide the process related knowledge needed for utilisation of CE techniques. One may ask how this knowledge is made up, and the answer is that it invariably comprises diverse, extensive, and often non-standard knowledge representations; certainly not the kind of information which is suited to storage and processing within conventional computer programs. These programs have traditionally been suited to high speed calculation and solving of formulae, in other words the processing of crisp data rather than the ‘fuzzy’ information which makes up descriptions of the real world. When designing a Knowledge Based System (KBS) for PM we cannot afford to adopt the assumptions and abstractions of the pure mathematician. In the real world of PM manufacturing, dealing with incomplete or error ridden data is the norm rather than the exception. Consequently it is necessary to employ techniques which can identify patterns with the data, and this is generally a task which is difficult for a computer but easy for an engineer. When solving production problems the engineer applies deductive reasoning, in other words he asks what production parameters need to be set to achieve the desired outcome. For example, the required outcome often involves production of a material which exhibits a specified tensile strength, and the question is what chemical composition, compaction density, and sintering conditions are needed to achieve this result. There may be a number of ways of achieving the desired outcome, and it is necessary to choose between these using a suitable selection criterion, such as cost. The question is, how can the skills required for evaluating different options, be encapsulated into a procedure or a system for use by a non specialist? The solution suggested here is to employ an hybrid Knowledge Based System approach. There are many reasons for drawing this conclusion, but the following factors provide a summary of the rationale: Hybrid KBS can be used to store, model, and make accessible, the vast amounts of process related knowledge needed for effective implementation of CE for PM  Hybrid KBS incorporate diverse types of knowledge representation techniques (E.g. databases, theoretical numerical and statistical models, fuzzy logic, and rule based inferencing). Each of these techniques has its own advantages which can be incorporated into the KBS. For example, Neural Network techniques could be used to determine optimal processing scenarios for arriving at specified process outcomes, and then high level rules could be used to determine which scenario is acceptable. Suppose, for example, that the desired mechanical properties were input to a Neural Network for determining processing parameters, and either a carbon steel or copper steel were recommended. High level rules could then be employed to choose the carbon steel, since it is likely to be of lower cost and exhibit less dimensional distortion than the copper steel.  Hybrid KBS can incorporate methods for machine learning such as Neural Networks. This enables the KBS to be constantly updated thereby preventing the knowledge base from becoming out-dated and redundant.  The KBS approach provides an excellent method for providing easy access to accurate and up-to-date process related knowledge. The Role of Computer Based Methods in the PM Industry In recent years there has been a steady but slow increase in the application of computer based methods in PM. It is true that manufacturers use Computer Aided Design systems such as AutoCAD, but these systems are mainly employed for computerised drafting. Computers are also used for statistical quality control, but here their application is often limited to providing an interface to an inspection device (such as a comparator), with data storage and processing, and possibly also a facility for printing control charts. Computer based PM materials selectors have been developed [1], however these systems function essentially as computerised look up tables. Some application of numerical methods (such as Finite Element Analysis), has also been employed for analysis of powder compaction and tool deflection. Despite these activities, and other similar developments in related areas, the application of computerised methods, (particularly Artificial Intelligence techniques) to PM has been relatively limited. Possible reasons for this include the following factors: In the early 1980’s unrealistic claims were made for the capabilities of expert systems and related techniques. A number of companies invested in the technology without realising the anticipated returns. Many industrial sectors are therefore weary of getting their fingers burned a second time.  Every company has its trade secrets which it does not wish to be disclosed to the competition. In reality it is likely that the competition is in possession of very similar information, which they also consider to be their trade secret. From the PM manufacturer’s viewpoint, a KBS can be perceived as a means of storing and making available process knowledge which was difficult and expensive to acquire. They would rather keep this know-how close to their chest and use it to help maintain a competitive advantage. Although understandable, this attitude is potentially the most detrimental for the long term success of the industry. Although it inhibits full implementation of Concurrent Engineering, it has traditionally been adopted by individuals as well as corporations.  The perceived technological challenges, complexity, and cost associated with industrial computer system developments (especially those incorporating AI methods), has contributed to the slow take-up of this technology. KBS for PM Research, Methodologies and Results of Implementations Research into methods for developing a KBS for PM have concentrated on the three key areas of Design Analysis, Materials Selection, and Process Optimisation [2] (see Figure 1). These areas were addressed through an investigation of the suitability of a range of AI methods for generation of relevant advice relating to design and production. System elements were developed using rule based inferencing methods [3], and models were incorporated following simulations of the behaviour of PM materials. (This included theoretical, statistical, and numerical simulations [2].) INTELLIGENT INTERACTIVE DESIGN MATERIAL SELECTION PM ADVISORY KNOWLEDGE BASED SYSTEM PROCESS OPTIMIZATION RULE BASED INFERENCING WITH EMPIRICAL MODELLING THEORETICAL, NUMERICAL AND EMPIRICAL MODELLING EXPERT SYSTEM TECHNIQUES Figure 1. Elements of a knowledge based advisory system for PM. Generation of Design Advice The design advisor KBS element provides the user with a library of PM design features which can be combined as desired to form the proposed design. The function of the KBS is to evaluate the emergent design for its suitability for production by means of PM. If unacceptable features are detected (e.g. a too thin a wall section), then the user is advised, and if possible the system can attempt an automatic design modification. Development of the KBS required identification of a suitable AI technique for performing the design analysis. After considering a number of options, it was decided that a rule based approach would be especially useful for development of systems for initial PM design analysis[4]. Fine tuning, or design optimisation, could utilise the results of simulations of powder compaction and/or tool deflection. The first system developed, PM design advisor [5], successfully demonstrated application of knowledge based techniques to design analysis of such commonly occurring components as washers and bearings. Application of curve fitting to empirical data to develop relevant rules, followed by use of an expert system to apply these rules to a given design, was demonstrated. Further research and development resulted in the second design advisor system (IDA), which incorporated an interactive approach [3]. This enabled development of a system that could attain a level of sophistication sufficient for application to the relatively complex design problems that are encountered when manufacturing structural parts. The Interactive Design Advisor successfully fulfilled this objective, and demonstrated how the approaches identified can be used to analyse and modify relatively complex design geometries, such as that of an hubbed gear. Generation of Materials Advice The materials advisor element of the KBS draws conclusions by inferencing, using rules that have been developed through statistical analysis of empirical data. This approach, which has also been reported to be commonly employed by powder manufacturers [6], offers a number of advantages, including easy access via proprietary statistical packages, and the generation of models that are relatively easy to understand and apply. Future developments of the materials advisor will incorporate the complex effects of variations in processing, which are be suited to the application of techniques such as Neural Networks [7]. Utilisation of these types of techniques offers potential for a KBS materials advisor to be developed for a wide range of PM applications, as well as enabling incorporation of a sufficient depth of knowledge for the system to be useful in practice. A KBS can also provide a convenient and powerful facility for future incorporation of models able to predict the effects on mechanical properties of variation in other relevant process parameters. Many such parameters are associated with the sintering process where, for example, tensile strength may depend on factors such as sintering temperature, time, and atmosphere. Generation of Production Advice Modelling of the density variations within a metal powder compact, resulting from application of a given pressure, was considered to be particularly important. The power and flexibility of Finite Element methods lead to their being chosen as a suitable method for modelling compaction behaviour. It was decided to use the ABAQUS FE system, which is generally accepted to be a powerful, well established, and flexible package. ABAQUS is shipped with ‘Porous Plasticity’ constitutive equations, for modelling the behaviour of porous metals subjected to stress. Since this incorporates a yield function that is dependent upon both applied stress and void fraction, it was considered suitable for the demonstrations of powder compaction modelling that formed part of the current work. The intention is to incorporate the results of the FE modelling into a case based reasoning KBS module for advising on powder compaction [8]. The other major area for production advice relates to powder packing. A considerable amount of fundamental research has been performed in this area by the authors. Coverage of this area of research is beyond the scope of this paper, however further details can be found in a number of technical publications [9][10][11]. Proposed Future Research Future Knowledge Elicitation Perhaps the greatest challenge associated with KBS knowledge elicitation, is the requirement for managing and efficiently utilising the large amounts of PM related data that is available. This requires the data to be processed to generate information that can be inter-related, and employed in an integrated fashion, as is required for generation of process related advice that has real production significance. To achieve this, it is useful for the knowledge engineer to study the methods employed by PM experts, so that the knowledge base contains ‘high-level’ know-how of a logistical and managerial nature as well as the ‘low-level’ PM knowledge, (an example of such low-level knowledge is the materials properties corresponding to given production conditions). Although this is essentially a problem in ‘meta-knowledge’, it is significant that application of a Knowledge Based System approach offers potential for storage and utilisation of high-level knowledge describing how the low-level PM knowledge can be employed, as well as providing an efficient means of storing the low-level knowledge itself. Future System Development System development will require transfer of geometrical information from a CAD solid model to the expert system for rule based design interrogation. The various end-users who would wish to utilise the system are likely to already employ proprietary CAD systems such as AutoCAD, SDRC I-DEAS, Pro/ENGINEER, and CATIA. The intention is to store geometrical information on designs developed with these systems, in a neutral format, prior to transfer to the expert system for geometrical evaluation. The proposed method for achieving this is to employ the ISO STEP standard (International Standards Organisation STandard for Exchange of Product Model Data) [12], for information storage and transfer. Once the relevant data is present within the expert system, the powerful and flexible information processing capabilities offered by modern KBS applications will be used to draw inferences regarding the manufacturability of the part, and subsequently, design optimisation. System development will necessitate an increase in the range and complexity of algorithms employed for inferring the production consequences of the feature combinations. However, design and compaction analysis will be simplified through application of a modular approach. To allow for a manageable size of system modules, it is suggested that each module should apply to a specific PM production method and material type (e.g. axial compaction of ferrous parts), and class of component (e.g. axisymmetric). This method of specifying applications for the modules will constrain the complexity and number of features required. In this way the feature library size, and complexity of algorithms for the analysis of feature combinations, should be kept within manageable limits. The following are the classes of axially-compacted components (in Figure 2 example components are shown for each class), that are expected to comprise the modules of the final system:1. Axisymmetric components, 2. Cyclically symmetric components, 3. Components with planar symmetry, 4. Components with repetitive symmetry, and 5. Asymmetric components. 1. Axisymmetric components 2. Cyclically symmetric components 3. Components with planar symmetry 4. Components with repetitive symmetry 5. Asymmetric components Figure 2. Classes of component to be analysed by the sub-modules of an advisory system for design of PM components (example components are shown). Each of these modules can be further categorised into the following five sub-modules:(a) Single level parts that are pressed from one direction. To avoid unacceptable density variations, the practical limitation in the thickness of such parts is 7.5 mm. (b) Single level parts pressed from both the top and bottom. (c) Two level parts pressed from both the top and bottom. (d) Multi-level parts pressed from both the top and bottom. Example components and corresponding categories include the following (category in brackets):Gear with hub (2c), Cylindrical bearing (1b), Camshaft timing pulley (2d), Washer (1a), Oil pump gears (2b), Flanged bearing (1c), Plate with central hole (3a), Plate with array of holes (4a), Crankshaft timing pulley (2d).