Expert systems for engineering design : Michael D Rychener

The field of engineering design has long provided researchers in artificial intelligence with a rich vein of challenging research issues, and has led to many important technical developments m the field. Although the penetration of AI into commercial applications software such as Computer Aided Design (CAD) is negligible today, manufacturing industry is beginning to show an increasing interest in 'intelligent' design support systems, spurred on by unprecedented pressure from skills shortages and competitive demands. It is therefore timely that this book should be published now. The book is an edited collection of papers describing case studies drawn from the research carried out at Carnegie Mellon University, USA, into expert systems applied to design problems. The contributions are loosely grouped into sections tackling synthesis (the generation of alternative designs), expertise (the nature of expert decisions), and integration (design tools and environments), with some editorial 'glue' linking these issues together. I have two concerns with the book. My first is the selective view of other AI research in engineering design that the book presents, which makes it difficult to place the work at Carnegie Mellon in context. The research in AI-based engineering design that is being carried out in many universities in the US is sparsely reviewed, and there are no references to any of the important equivalent work being carried out elsewhere in the world. The broader but equally relevant research in design carried out under the label of human factors, psychology or social science is not mentioned, nor is there reference to any of the advanced commercial systems now available. Truth maintenance concepts, which together with blackboard architectures have become important research tools in AI design research, are not discussed. My second concern with this book is the bold claim it makes that expert systems are well suited for application to design without acknowledging the sometimes conflicting major debates in artificial intelligence applied to design. The early years of AI research were noted for the excessively optimistic claims made for the capabilities of expert systems. The resulting cynicism put back development of AI several years, particularly in the UK. It is disappointing, therefore, that this book seems to follow the same path. Within narrowly defined system domains, expert systems have been shown to work well, and good examples of this are described in the first section of the book, 'Synthesis', which covers specific expert system applications in architecture, chemical engineering and computer hardware design. The papers on expert systems for computer hardware design, catalyst selection and alloy design are particularly interesting, describing both the implementation and evidence of the effectiveness of their approach. All the contributors have had to balance a sufficient discussion of the application domain against an examination of the implementation issues. In general, though, the balance achieved is only partially successful, with the result that I found it difficult to assess the real success of expert systems in the design domains described. The implementations reported give 'intelligent' support for much simplified design problems, but one is left with little evidence to answer the key question: could such systems cope with the complexity of the complete problem? The last section of the book offers a possible answer. By integrating stand-alone expert systems, it may eventually be possible to produce design support environments that will be sufficiently powerful to handle the full complexity of industrial design problems. Blackboard architectures have been widely used to integrate inference engines with a shared knowledge base holding the design information. In a development of this, the paper by Talukdar and Cardozo describes an environment that offers powerful integrating capabilities for building organizations of distributed collaborating expert systems. The environment provides basic primitives to allow cooperative interaction of independent, distributed blackboard systems. The blackboard paradigm has also been used to provide powerful models of human team-based design processes, and perhaps this offers a route to buildlng expert systems that can interwork with human users? The middle section of the book, concerned with the nature of human expertise, considers important isues in graphical techniques for human-computer interaction. Although designers use graphical techniques extensively to express complex concepts, there is little reported research that considers this in the context of knowledge representation and user interfaces to expert systems. A fundamental issue is the control of complexity by the use of abstract models, highlevel representation languages, and graphical tools for building and iterating hierarchically structured diagrams which represent issues in the problem domain. The individual contributions that make up this work vary widely in style and quality, and mostly give an overview of both issues and implementations. Provided the selective nature of the book is recognized, it makes useful background reading for students taking undergraduate design or expert system courses, as well as for the technically-minded general reader, although it is expensive at its price.