To solve the diesel engine fault diagnosis problem, a fault diagnosis method based on the combination of variational mode decomposition (VMD) and kernel fuzzy C-means clustering (KFCM) is proposed. This paper optimizes the selection of decomposition level K in VMD algorithm, and proposes an adaptive choosing method for K. Then, three key components are selected from the decomposition results of the optimized VMD algorithm to calculate the maximum singular values, which are input into the KFCM algorithm as three-dimensional eigenvectors for classification and recognition. The optimized VMD method，VMD method and empirical mode decomposition (EMD) method are used to decompose and recognize the simulated signal and the experimental data of a diesel engine. The results show that the proposed method obviously improves the accuracy of pattern recognition. The joint algorithm proposed in this paper has better application prospects.