Three-stage hybrid fault diagnosis for rolling bearings with compressively sampled data and subspace learning techniques

To avoid the burden of much storage requirements and processing time, this paper proposes a three-stage hybrid method, compressive sampling with correlated principal and discriminant components (CS-CPDC), for bearing faults diagnosis based on compressed measurements. In the first stage, CS is utilized to obtain compressively sampled signals from raw vibration data. In the second stage, an effective multistep feature learning algorithm obtains fewer features from correlated principal and discriminant attributes from the compressively sampled signals, which are then concatenated to increase the performance. In the third stage, with these concatenated features, multiclass support vector machine is used to train, validate, and classify bearing faults. Results show that the proposed method, CS-CPDC, offers high classification accuracies, reduced computation time, and storage requirement, with fewer measurements.