Machine Fault Diagnosis Based on Probabilistic Bayesian Active Learning

In the field of machine fault diagnosis, data-driven approaches have been widely applied. However, we often encounter situations where there is a lot of healthy data but a scarcity of fault data. And for some devices, sample labeling is a time-consuming and laborious process. Under these cases, developing fault diagnosis models using data-driven supervised learning approaches becomes exceedingly challenging. Therefore, this paper presents a machine fault diagnosis method based on probabilistic Bayesian active learning. The approach employs probabilistic Bayesian neural network to assess the uncertainty of samples, selects those with the highest uncertainty for labeling, and incorporates them into the training set, thereby realizing an active learning process. This method is expected to reduce the cost of sample labeling by actively selecting samples for labeling.