Probabilistic Inference of Fault Condition of Cyber-Physical Systems Under Uncertainty

Cyber-physical systems (CPS) are paving new ground with increasing levels of automation and usage in applications with complex environments, posing greater challenges in terms of safety and reliability. The increasing complexity of CPS environments, tasks, and systems leads to more uncertainties. Unless properly managed, these uncertainties may lead to false detection of real fault condition of a system, which in turn may affect decision making and potentially cause fatal consequences. In order to implement safety-critical missions, such as autonomous driving, it is essential to develop a reliable monitoring and assessment service dealing with the complexity and uncertainty issues. In this article, we propose a fault detection function based on Bayesian inference, which combines empirical knowledge with information of the specific system. By considering uncertainties as possible causes for false detection, various uncertainties during the detection process are analyzed, and the ways to quantify and propagate them are explored. As a result, probabilistic inference is achieved for distinguishing system faults from uncertainties, which contributes to more reliable detection results regarding system faults under dynamically changing environments. A case study on an microelectro mechanical system (MEMS) accelerometer is conducted and the result shows that the fault detection function effectively distinguishes system faults and uncertainties arising from the environment.