Active Learning of Support Vector Machine for Fault Diagnosis of Bearings

摘要

Based on traditional Active Support Vector Machine (ASVM), the learning method of Probabilistic Active SVM (ProASVM) is introduced to detect fault of bearings. Compared with the general SVM, the active learning methods can effectively reduce the number of samples on the condition of keeping the classification accuracy. ASVM actively selects data points closest to the current separation hyperplane, while ProASVM selects the points according to the probability of the sample point as a support vector. The two methods are applied to classify the practical vibration signal of bearings and the results show that ProASVM is a better algorithm of classification than ASVM.