Fault Diagnosis and Prognosis Based on Lebesgue Sampling

Traditional fault diagnosis and prognosis (FDP) approaches are based on periodic sampling, i.e. samples are taken and algorithms are executed both in a periodic manner. As the volume of sensor data and complexity of algorithms keep increasing, the bottleneck of FDP is mainly the limited computational resources, which is especially true for distributed applications where FDP functions are deployed on microcontrollers and embedded systems with limited computation resources. This paper introduces the concept of Lebesgue sampling in FDP and proposes a Lebesgue sampling based fault diagnosis and prognosis (LS-FDP) framework. In the proposed LS-FDP, a novel diagnostic philosophy of “execution only when necessary” is developed in computation cost reduction and uncertainty management. For prognosis, different from traditional approaches in which the prognostic horizon is on the time axis, the proposed approach defines prognostic horizon on the state axis. With a reduced prognostic horizon, the LS-FDP naturally benefits the uncertainty management. The goal is to create the fundamental knowledge for LS-FDP solutions that are cost-efficient, capable for the deployment on systems with limited computation sources, and supportive to the trend of distributed FDP schemes in complex systems. The design and implementation of LS-FDP based on particle filtering algorithms are presented with experimental results to verify the effectiveness of the proposed approaches.