Hierarchical Modular Fault Diagnostics for Complex Systems

Ohio State University Research Foundation -Do Not Use, Columbus OH

Investigators

Abstract

Laci Jalics, Delphi Research Labs Giorgio Rizzoni, The Ohio State University Andrea Serrani, The Ohio State University Ahmed Soliman, The Ohio State University Hierarchical, Modular Fault Diagnosis in Complex Systems The objectives of this GOALI project program are: 1) to develop a methodology that permits the design of modular, hierarchical, model-based diagnostic algorithms that can result in re-usable software modules, and that can be methodically organized into a complete, self-contained monitoring system; and 2) to demonstrate the methodology in collaboration with an industry partner. Fault tolerance is a critical element in complex automotive systems because of the introduction of "by-wire" systems in vehicles. "by-wire" systems generate electrical commands through driver input and transfer them to the computer-controlled electro-mechanical actuators. With respect to traditional systems, "by-wire" systems require complex control and diagnostic strategies to guarantee proper vehicle behavior and to prevent system malfunctions. A significant roadblock for safety-critical by-wire systems is the realization of a robust fault-diagnostic strategy that can be accomplished quickly and with limited or no additional hardware. This project addresses these issues in three stages: 1) through the use of physically based system models to understand interactions and dependencies among subsystems in a complex system; 2) with the design of residual generator units that can locally diagnose faults in components, sensors and actuators; and 3) using a hierarchical structure that links components, systems, and subsystems through the use of fault propagation digraphs. The fault propagation digraph contains a hierarchical representation of available knowledge about the characteristics of fault propagation within the system. Each level in the hierarchy contains one or more structures that together represent a view of the system under a particular granularity. The granularity of view increases with levels. Thus as one traverses down the levels of the model, the resolution of view increases. If a high resolution of diagnosis is not required, failure analysis may be confined to the upper levels of the hierarchy, thereby preventing a slowdown of the diagnosis due to excessive detail.