Low latency change detection aims to minimize the detection delay of an abrupt change in probability distributions of a random process, subject to certain performance constraints such as the probability of false alarm (PFA). In this paper, we study the low latency detection of bearing faults of direct-drive wind turbines (WT), by analyzing the statistical behaviors of stator currents generated by the WT in real-time. It is discovered that the presence of fault will affect the statistical distribution of WT stator current amplitude at certain frequencies. Since the signature of a fault can appear in one of the multiple possible frequencies, we need to monitor the signals on multiple frequencies simultaneously, and each possible frequency is denoted as a candidate. Based on the unique properties of WT bearing faults, we propose a new multi-candidate low latency detection algorithm that can combine the statistics of signals from multiple candidate frequencies. The new algorithm does not require a separate training phase, and it can be directly applied to the stator current data and perform online detection of various possible bearing faults. The theoretical performance of the proposed algorithm is analytically identified in the form of upper bounds of the PFA and average detection delay (ADD). The algorithm allows flexible parametric adjustment of the tradeoff between PFA and ADD. INDEX TERMS Bearing fault, fault detection, quickest change detection, wind turbine.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.