Transfer fault prognosis of bearing across different machines with the application of negative transfer mitigation

Accurately predicting the remaining useful life (RUL) of rolling element bearings is essential for maintaining the reliable operation of mechanical assets. Many existing research endeavours focus on utilizing domain adaptation (DA) techniques to predict RUL across variable operating conditions within the identical machine fleets. However, this oversimplified approach may limit the applicability and prediction accuracy in real industrial settings. While DA can address domain shifts between different machines or conditions, it also has the risk of the negative transfer, which can significantly compromise the effectiveness of transfer learning (TL)-based prognostic methodology. A dual network comprising a deep convolutional neural network (DCNN) and a multi-layer perceptron (MLP) is proposed for a more challenging and realistic scenario, i.e., transfer RUL prediction between the heterogeneous rolling bearings. Additionally, a negative transfer mitigation (NTM) scheme is integrated within the DA module for mitigating negative transfer. Finally, experiments conducted on two publicly available bearing datasets highlight the effectiveness and superiority of the proposed model.