Application of multiclass support vector machines for fault diagnosis of field air defense gun

This paper introduces multiclass support vector machines (SVM) and a back-propagation neural network (BPNN) for fault diagnosis of a field air defense gun. These intelligent methods preclude human error in fault diagnosis, and they make it possible to diagnose a new failure precisely and rapidly. Our experimental results show that both SVM and BPNN provide excellent fault diagnosis accuracy when sufficient training samples are examined, and multiclass SVM models have better fault diagnosis accuracy than BPNN models when numbers of training sets are small. Our multiclass SVM approach also offers advantages of solution stability and requires fewer control parameters; it is easier to apply it to fault diagnosis problems than BPNN.