Phase Space Feature Based on Independent Component Analysis for Machine Health Diagnosis

Abstract

This paper proposes a new feature extraction method based on Independent Component Analysis (ICA) and reconstructed phase space. The ICA-based phase space feature unifies the system dynamics embedded in vibration signal and higher-order statistics expressed in phase spectrum and hence, is effective for machine health diagnosis. The new feature extraction is done in three steps: first, the Phase Space Reconstruction (PSR) is performed to reconstruct a phase space with the dimension covering dynamic structure information; second, the ICA bases are trained by a number of constructed phase points; and finally, the new feature is quantitatively calculated by evaluating the correlation property of transformed coefficients based on ICA bases. The presented feature contains plentiful phase information with the training pattern, which is often under evaluated when using existing methods. It has excellent pattern representation property and can be applied for signal classification and assessment. Experiments in an automobile transmission gearbox validate the effectiveness of the new method.