Variational Residual Model for Machinery Condition Monitoring under Complex Degradation Processes

Deep-learning based approaches have been widely used for constructing health indicator from machinery monitoring signals. However, the constructed health indicators (HIs) perform unstable under complex degradation processes. To address the problem, a variational residual network (VRN) is proposed in this paper. By implanting variational mechanism in the forward propagation network, VRN could infer the distribution characteristics of the monitor signals during degradation processes. Compared with the existing methods under turbopump bearing degradation dataset, health indicators constructed by the VRN perform higher monotonicity and trendability.