Positive-incentive noise in artificial intelligence-enabled machine fault diagnosis
Changpu Yang, Zijian Qiao, Li Liu, Anil Kumar, Ronghua Zhu
Abstract
Noise often weakens the fault characteristics of machinery, resulting in the degradation of the accuracy of intelligent fault diagnosis and prediction models. However, noise is not completely harmful, and the emergence of the idea of positive-incentive noise (PI noise) has prompted researchers to rethink the role of noise. Most of the research on intelligent diagnosis belongs to noise aversion, while ignoring the PI noise in intelligent diagnosis. Therefore, the PI noise is explored in depth by adding noise to the intelligent diagnosis model. The positive impact of injected noise on the intelligent diagnosis model is explained through the perspective of expected loss. Meanwhile, two datasets, including bearings and hydraulic motors, were collected for experimental validation. The experimental results show that the existence of noise can improve the diagnosis accuracy of the model. This proves the existence of PI noise in the intelligent diagnosis model.