Abstract:Abstract: When kernel discriminant analysis is applied to machinery fault diagnosis, the main challenge is to determine the parameters of kernel function. Aiming at this problem, a method for machinery fault diagnosis based on bayes optimal kernel discriminant analysis is proposed. Firstly, the optimal kernel parameter is selected by optimizing the homoscedastic criterion using gradient descent; Secondly, with the optimal kernel parameter, the original examples are projected onto an optimal subspace where the examples are best separated. Finally, fault classification is performed in the optimal subspace by nearest neighbour method. the proposed method is applied to fault diagnosis of roller bearings and comparison is made with the results of related methods, experimental results demonstrate the performance of the proposed method is comparable to that of support vector machines, but avoiding the problem that kernel parameters need to be manually selected.