Exploiting multiple knowledge models in the development of real-time diagnostic expert systems for highly automated manufacturing plants

Abstract

This paper describes the results of the DIREK Project (knowledge-based, real-time diagnosis and repair for a robotized handling and storage system), aimed at developing a real-time diagnosis system for a highly automated SNIA fibre spinning plant. The project effectively implemented a multi-model approach to diagnosis in manufacturing environments, exploiting structural, functional, behavioural and heuristic models. Among other things, particular emphasis has been placed on the plant behavioural model, which can be derived from the software code running on the PLCs which control an automated manufacturing facility. The existing diagnostic system is now operational at an SNIA plant in Italy, fully integrated with the factory environment and able to support different levels of users through distributed man-machine interfaces. The paper provides an insight into the theoretical background of the project and describes the adopted methodology, with special attention given to the knowledge acquisition problems arising in the development of the various knowledge models included in the diagnostic system. Furthermore, the architecture and functionalities of the existing system are described, along with the achieved benefits and further exploitation potential. Both IT and user perspectives are considered in the paper.