Fault Feature Extraction of Bearings for the Petrochemical Industry and Diagnosis Based on High-Value Dimensionless Features

Abstract

The time and frequency domain features of a petrochemical unit have a variety of effects on the fault type of bearings, and the signal exhibits nonlinearity, unpredictability, and ergodicity. The detection system's important data are disrupted by noise, resulting in a huge number of invalid and partial records. To reduce the influence of these factors on feature extraction, this work presents a method for the fault feature extraction of bearings for the petrochemical industry and for diagnosis based on high-value dimensionless features. Effective data are extracted from the obtained data using a complex data preprocessing approach, and the dimensionless index is expressed. Then, based on the distribution rule of the dimensionless index, the high-value dimensionless features are retrieved. Finally, to ensure sample completeness, a high-value dimensionless feature augmented model is developed. This approach is applied to the bearing fault experiment platform of a petrochemical unit to effectively classify the bearing fault features, which benefits theoretical guidance for the feature extraction of bearings for a petrochemical unit.