Machinery Fault Diagnosis Based on Weighted 2D Fault Feature Extraction and Multi-level Information Fusion

Machinery fault diagnosis is critical to the reliability and safety of modern industrial systems. In order to better diagnose the faults of rotating machinery, this paper presents a fault diagnosis model based on a weighted twodimensional (2D) time-frequency spectrum and multi-level information fusion. Firstly, multiple time-frequency analysis methods are performed on the collected vibration signals to get 2D time-frequency spectrum diagram. Secondly, a feature weighting strategy is proposed to weight the 2D time-frequency features to further enhance the fault information and reduce the disturbing of noise. Then, a 2D dimensionality reduction method is applied to reduce the dimensionality of the weighted 2D timefrequency features and extract the primary fault information contained in the 2D fault features. Finally, the extracted fault features are inputted to a support vector machine (SVM) for fault diagnosis model training. The proposed multi-level information fusion strategy based on Dempster-Shafer (D-S) evidence theory is also applied to resolve the contradiction of fault diagnosis results between different features and different sensors.