Reasoning under Uncertainty for Knowledge-Based Fault Diagnosis: A Comparative Study

Abstract

This paper addresses reasoning under uncertainty for knowledge-based fault diagnosis. We illustrate how the fault diagnosis task is influenced by uncertainty. Furthermore, we compare how the diagnosis task is solved in the Bayesian and the Dempster-Shafer reasoning framework, in terms of both diagnostic performance and additional objectives, like transparency, adaptability, and computational efficiency. Since the diagnosis problem is influenced by different kinds of uncertainty, it is not straightforward to determine the optimal reasoning method. First, the different uncertain influences all have their own characteristics, asking for different reasoning approaches. So, to solve the whole problem in one reasoning framework, approximations and trade-offs need to be made. Second, which types of uncertainty are present and to what extent, is highly application-specific. Therefore, the best framework can only be assigned after the problem, the uncertainty characteristics, and the user requirements are known.