9612190 Murphey This Grant Opportunity for Academic Liaison with Industry cooperative research project between the University of Michigan-Dearborn and Ford will provide new ways to do end of line test and fault identification in automobile manufacture. Vacuum leaks, transmission shift, idle quality and axle ratio are the variables that the diagnostic system will consider, with the constraints that diagnostic methods must use existing sensors and be executed quickly. The technical approach will use a distributed architecture with diagnostic agents for each potential fault. A central agent will evaluate the results from each diagnostic agent, taking into account interactions between faults. The decision making and classification approach uses fuzzy logic, with membership functions based on engineering heuristics, the easy-out rules applicable for the majority of vehicles, and complex fuzzy rules to refine and identify the specific faults. Implementation and testing will use real data provide by Ford to the University of Michigan-Dearborn, then move to a test cell at Ford, finally migrating to plant operation. The strength of this proposal is the significant industry commitment and participation, and the potential to bridge a technical gap between fuzzy systems and neural networks by focusing on this important diagnostic problem. The architecture of distributed agent based fuzzy systems for diagnostics can be transferred to other applications, improving quality and reliability of manufactured products.
