Optimization of the Design of Diagnostic Systems through Requirement-Based Feature Selection

Abstract

The classification of faults based on available diagnostic information in large complex mechatronic systems is a well-studied research subject. One challenge is the classification of faults with missing or ambiguous diagnostic information present. In real-world applications, this has the effect of misclassified faults, resulting in costly replacements of functional components. As even the acquisition of diagnostic information is afflicted with costs and part of the diagnostic functions are mandatory, the design of diagnostic content at an early system development stage is essential. The paper presents a procedure to automatically allocate a set of diagnostic functions for a complex mechatronic system that is informative and low in costs. It proposes a granular graph structure called the Diagnostic Cover Graph (DCG) to represent the system functionalities and the respective diagnostic functions. The DCG can represent restrictions on the availability of functionalities, restrictions on the executability of diagnostic functions and potential interdependence between them. On its basis, a decision table is generated, and a minimal test cost feature selection method such as the Test-Cost-Sensitive Quick Reduct (TCSQR) is conducted. The approach guarantees the proposal of a close-to-optimal subset of diagnostic functions. The capability of the presented procedure is demonstrated for the vehicle brake system for several diagnostic scenarios.