Fusion of multi-class support vector machines for fault diagnosis

Data fusion strategies based on multi-class support vector machines are proposed, in the centralized scheme, the information from several sources is combined to construct an input space. In the distributed schemes, the input space is constructed corresponding to each information source and the multi-class support vector machine is used for modeling each source. The distributed data fusion strategies are applied to combine these multi-class support vector machine models, it is taken into account that a SVM classifier realizes classification by finding the optimal classification hyperplane with maximal margin. The proposed methods are demonstrated with the fault diagnosis of a diesel engine. The experimental results show that most of the proposed approaches can largely improve the diagnostic accuracy. The robustness of diagnosis is also improved.