Multidomain Feature Fusion for Varying Speed Bearing Diagnosis Using Broad Learning System

Abstract

Bearing is one of the most critical mechanical components in rotating machinery. To identify the running status of bearing effectively, a variety of possible fault vibration signals are recorded under multiple speeds. However, the acquired vibration signals have different characteristics under different speeds and environment interference, which may lead to different diagnosis results. In order to improve the fault diagnosis reliability, a multidomain feature fusion for varying speed bearing diagnosis using broad learning system is proposed. First, a multidomain feature fusion is adopted to realize the unified form of vibration characteristics at different speeds. Time-domain and frequency-domain features are extracted from the different speeds vibration signals. Then, the broad learning system is employed with the fused features for classification. Our experimental results suggest that, compared with other machine learning models, the proposed broad learning system model, which makes full use of the fused feature, can improve the diagnosis performance significantly in terms of both accuracy and robustness analysis.