Application of the multi-kernel non-negative matrix factorization on the mechanical fault diagnosis

Although the computation amount involved in the image processing is very large, image information which is very intuitive and easy to be understood has attracted great many attentions in the fault diagnosis of machines. In order to extract useful features from the images accurately and perfectly, a novel mechanical fault diagnosis method was proposed by the combination of the multi-kernel non-negative matrix factorization and multi-kernel support vector machine. The genetic algorithm was used to optimize the parameters of both multi-kernel non-negative matrix factorization and multi-kernel support vector machine. Experiments were used to validate the efficacy of the proposed method. It is shown that the multi-kernel function combined with the polynomial kernel function and radial-based kernel function can describe the fault feature more perfectly in the kernel space than a single kernel function. Sound accuracy can be obtained in the application of the bearing fault diagnosis. Compared with the fault diagnosis method based on the sparse non-negative matrix factorization, the proposed method is more accurate in the condition identification of rotor.