The overall goal of this research project is to apply artificial intelligence (AI) methodology to computer aided design (CAD). In the area of computer-aided engineering design, AI techniques hold the promise of achieving two significant technical goals: 1. The production of CAD software which allows semi-skilled non-specialists to produce good candidate designs so as to dramatically hasten the dissemination and assimilation of new technology. 2. The possibility of "discovery" by CAD software of new designs and design methods not conceived by human experts. Over the past decade, the PI and his coworkers have developed two CAD packages CONSYD and POLYRED presently in use by industry as well as by universities as teaching tools. CONSYD allows the experienced user to interactively build a process model from data, determine the principal control problems through process simulation, and use a wide variety of sophisticated design procedures to synthesize several candidate designs of control systems which can be evaluated in plant testing. The current limitation of CONSYD (and other CAD packages for control system design) is that the design procedures require the user to have a relatively sophisticated knowledge of the control theory behind each design method. Similarly POLYRED can be used to design polymerization reactors and requires expertise in design, polymerization and reaction engineering on the part of the user. To overcome some of these limitations on the population that can use these CAD packages, the PI proposes to devise AI strategies which will allow the average control engineer in industry to use CONSYD and the typical industrial scientist and engineer involved in polymerization reactor modelling and design to use POLYRED. The first step will be to create an expert system for control system design which will sit above CONSYD and call the appropriate FORTRAN programs in the CONSYD and POLYRED packages for detailed technical calculations. The second part of the project will involve the study of "discovery" and "learning" features which could be added to the AI design package. For the control systems design problem there are two types of "learning" which will be studied: 1. The discovery of new, previously unknown, types of control system designs which may arise from the application of expert systems with fixed rules. 2. The formulation of new design rules based on the accumulated experience of the AI package. Similar learning features will be incorporated into the polymerization reactor system. The results will not only be new software for these two specific areas, but will also increase the knowledge base on the use of AI techniques and expert systems to nontrivial, real problems.
