Bearing Fault Diagnosis Based on Probability Boxes Theory and SVM Method

To solve the information loss on the feature extraction process in the traditional fault diagnosis, this paper proposes a new method which based on probability boxes and Dempster Shafer Structure (DSS). The DSS was extracted from the raw data and then transformed into a probability box. The bearing fault diagnosis was done by the probability boxes images recognition. To solve the excessive computing cost caused by large sample frequency and the overlaps among p-boxes, the Support Vector Machine (SVM) was involved. The SVM features database was established by some cumulative uncertainty measures methods of p-boxes. The test result shows that this method is fast, not sensitive to noise and has high recognition rate with high accuracy.