No abstract

No abstract

tSystems and Software Engineering Lab., Toshiba Corp. 70 Yanagi-cho, S aiwai-ku, Kawasaki-ci ty, 21 0 Japan iwamasa%ssel.toshiba.co.jp@uunet .uu.net $-Fuchu Works, Toshiba Corp. Abstract I n this paper we propose a model-based diagnostic system f o r continuous physical devices such as a thermal power plant. The aim of model-based diagnosis is to find faulty components i n the model f r o m observations (symptoms). The model-based diagnosis is well formalized as the diagnosis from first principle by Reiter [6]. Reiter proposed a method to calculate diagnoses from first principle using a theomm prover. In the continuous physical domain it is daficult t o obtain proper theorem prover. Set-covering is another approach t o get diagnosis when the causal relations between symptoms and disorders is clearly defined. Our new method combines the model-based approach and set-covering approach. W e introduce the Qualitative Causal Model(QCM) and define symptoms and qualitative disorders i n QCM. The qualitative propagation on QCM and the B I P A R T I T E algorithm based on set-covering calculates all the diagnoses. Finally the proposed method is proved to realize the diagnosis from first principle i n the continuous physical domain.

Model-Based Diagnosis {MBD) uses the design and physical principle of the target as knowledge base in stead of wing enyirical diagnostic knowledge. Although MDB is promising to resolve the knowledge acquisition bottleneck, it involves the problem of combinatorial explosion in making asswnptwns. This problem is remarkable especially when target is large and complex without enough observable information.In this paper, we propose a new framework of MDB that can control and avoid this dificulty. In our fromavork, diagnostic system comprises three parts, evaluation, control and reason maintenance. By introducing propagation oriented control into control part, we can control the number of assumptwns. We analyze basic tasks for MDB based on proposed framework and propose "adjustable diagnosis" by using these tasks.We also propose a shell for MDB that has "adjustable diagnosis" feature. Using this shell engineer 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.

No abstract