Bearing Fault Diagnosis with Two-stage Multi-source Information Fusion Based on Improved Stacking Algorithm and D-S Evidence Theory

To address the limitations of fault diagnosis performance based on a single sensor and a single fault feature, a two-stage multi-source sensor information fusion method for bearing fault diagnosis is proposed. First, considering sample imbalance characteristics and multi-feature correlation, an improved Stacking algorithm is proposed to construct a one-dimensional residual feature fusion network. This network improves the training of multi-sensor features and achieves fault feature fusion. Then, to cope with the uncertainty of multi-source sensor information, an improved evidence fusion rule considering the amount of evidence and its reliability is proposed to realize fault decision fusion based on Dempster-Shafer evidence theory. A case study on rolling bearings shows that the proposed method achieve all above 0.98 on average accuracy, precision, recall and f1-score under different working conditions. Compared with other seven methods, the proposed method shows superior performance and good generalization.