Quickest Fault Detection over Lossy Networks for Industrial Machinery

# 189


We study quickest fault detection for industrial machine condition monitoring where sensor data is transmitted over lossy networks. Existing techniques suffer from long detection delays despite high sensitivity. We pose this as a non-Bayesian quickest change detection problem where the post-change probability distribution is unknown but belongs to a known set. For this, we derive a computationally efficient generalized CUSUM algorithm based on the generalized likelihood ratio principle, and provide analytical bounds on the false alarm probability and expected detection delay. We apply our sequential detection techniques to detect bearing faults, by recognizing the cyclostationary nature of bearing vibration signals, constructing a cyclic spectrum statistic to detect the induced cyclostationarity, and combining this with our generalized CUSUM algorithm. We show that our algorithm is able to detect small changes even at very low SNR while ensuring practical detection delays and false alarms. For the case when measurements are transmitted over lossy wireless networks subject to packet losses and queuing delays at the sensor, we model the problem as quickest change detection for a Markov process. We derive a CUSUM-based algorithm and prove its asymptotic optimality. Furthermore, our analysis extends to cases with incomplete data due to packet drops. We also investigate the impact of different queuing disciplines employed at the sensor on the detection performance.

Dr. Krishna Chaythanya KV , Dept. Of ECE, IISc Bangalore

Krishna Chaythanya KV recently defended his PhD thesis at the Dept of ECE, IISc, where his research focused on developing quickest fault detection algorithms for cyber-physical systems like networked industrial machine monitoring. Prior to his doctorate, he spent several years in the industry applying signal processing techniques to wireless communications and designing wireless networks for IoT devices after earning his master's degree from IISc in 2011. Krishna Chaythanya's research interests lie at the intersection of statistical inference, data analytics, signal processing, and communication theory.