After explaining how inference engines work and which knowledge is used for the inference in two diagnosis task-domains of troubleshooting and enterprise diagnosis, the inference engines are described by IPST.
The troubleshooting expert system called Knowledge Complier II (T.Yamaguchi and R.Mizoguchi et al. ,92) is a kind of hybrid model-based diagnosis system that uses both the model-based process and the heuristics(surface causal association)-based process. Furthermore, the model-based process has been divided into the following two sub-processes: faulty components identification(FCI) and fault analysis from faulty components(FAFC). FCI is performed based on propagation of qualitative values, using the following two kinds of models: a device model described by media-flow (such as heat-flow or force-flow) and a physical model described by a set of physical equations. FAFC is performed by a chain of application of generic(not specific or heuristic) causal models focusing on failure mechanism. FCI and FAFC perform model-based diagnosis at different levels of granularity on which deep models are formulated. On the other hand, some heuristics have been applied to correspond between one concept in FCI and another concept in FAFC. Thus the model-based process and the heuristic-based process make up the whole diagnosis process in Knowledge Compiler II. This has been described by IPST, as shown in Figure 3.
The enterprise diagnosis expert system called FIMCOES (Pedro Garcia del Valle y Duran and T.Yamaguchi ,94) performs the task by constraint satisfaction over a set of equations of financial indices combined with a best-first search solver. After identifying bad financial indices, FIMCOES tries to choose an action available to the enterprise using knowledge to describe the relationships between financial indices and actions for enterprises. Afterwards, FIMCOES tries to simulate side effects in other financial indices brought about by taking the action. If bad side-effects arise in other financial indices, FIMCOES repeats the processes until no more financial indices coming up with problem. Thus constraint satisfaction and simulation make up the whole diagnosis process in FIMCOES. This has been described by IPST, as shown in Figure 4.
Figure 3: Inference Structure Described by IPST in a Troubleshooting Expert System
Figure 4: Inference Structure Described by IPST in an Enterprise Daignosis Expert System