Abstract:

This paper addresses the problem of system design for diagnosability. Specifically, it focuses on design of built-in self-tests (BISTs) for subsystems based on electronic control units (ECUs). The BISTs play a major role in diagnosis of the systems and in particular in determining if the failure is in the ECU or externally in the sensors, detectors, or actuators. The design of BISTs involves a tradeoff between the diagnostic benefit gained by the presence of a BIST versus cost of providing it in the system. We describe a systematic methodology and software tools for quantitative tradeoff analysis of BISTs. The methodology utilizes graphical probabilistic models (Bayesian networks) to represent the diagnostic properties of the system and structured equation models to perform costbenefit analysis. The models are developed from the knowledge of the systems (i.e. documentation and/or subject matter experts) and from data. The methodology is suitable for design of BIST for a broad range of systems. We illustrate the use of it on an example of a ECUbased subsystem for control of agricultural machinery.