Data Mining for Building Rule-based Fault Diagnosis Systems

This paper aims at developing rule-based fault diagnosis (RBFD) systems using data mining techniques, where we address a problem of generating rules for faults with low probability of occurrence but considerable conceptual importance. Main technical contributions include a multilayer structure of rule generation and use, and a regularization model embedding some information on recognition rate, coverage rate and generalization capability for rule optimization. A case study is carried out by an engine diagnostics to illustrate effectiveness of our methodology.