Fault diagnosis based on support vector machine ensemble

Accurate fault diagnosis for large machines is very important for its economic meaning. In essence, fault diagnosis is a pattern classification and recognition problem of judging whether the operation status of the system is normal or not. Support vector machines (SVMs), well motivated theoretically, have been introduced as effective methods for solving classification problems. However, the generalization performance of SVMs is sometimes far from the expected level due to their limitations in practical applications. Ensemble learning of SVMs provides a promising way for such cases. In this paper, a new ensemble learning method based on genetic algorithm is presented. Genetic algorithm is introduced into ensemble learning so as to search for accurate and diverse classifiers to construct a good ensemble. The presented method works on a higher level and is more direct than other search based ensemble learning methods. Different strategies of constructing diverse base classifiers are also studied. Experimental results on a steam turbine fault diagnosis problem show that the presented method can achieve much better generalization performance than other methods including single SVM, Bagging and Adaboost.M1 if the strategies used are appropriate for the practical problem.