Model-based diagnosis shell and its application to steel plant

Model-based diagnosis (MBD) uses the design and physical principle of the target as a knowledge base instead of using empirical diagnostic knowledge. Although MDB is promising to resolve the knowledge acquisition bottleneck, it involves the problem of combinatorial explosion in making assumptions. This problem is remarkable especially when the target is large and complex without enough observable information. We propose a new framework of MDB that can control and avoid this difficulty. In our framework, the diagnostic system comprises three parts: evaluation, control and reason maintenance. By introducing propagation oriented control into the control part, we can control the number of assumptions. We analyze basic tasks for MDB based on the proposed framework and propose “adjustable diagnosis” by using these tasks. We also propose a shell for MDB that has an “adjustable diagnosis” feature. Using this shell engineers can build an application that can focus diagnosis. We show an example of a system for steel plant diagnosis which is built using the shell