Feature Extraction Technique for Fault Prognosis Based on Fault Trend Analysis

Fault prognosis is one of the key techniques for prognosis and health management, and an effective fault feature can improve prediction accuracy and performance. A novel approach of feature extraction for fault prognosis based on fault trend analysis was proposed in this paper. In order to describe the ability of tracking fault growth process, definitions and calculations of fault trackability was developed, and the feature which had the maximum fault trackability was selected for fault prognosis. The vibration data in bearing life tests were used to verify the effectiveness of the method was proposed. The results showed that the trackability of energy entropy for bearing fault growth was the maximum, and it was the best fault feature among selected features root mean square （ RMS）, kurtosis, new moment and energy entropy. The proposed approach can provide a better strategy for fault feature extraction of beatings in order to improve prediction accuracy.