For the 1D non-negative matrix factorization (1DNMF) technique, when the dimension of the 2D matrix is reduced, the problem of huge data volume, low computational efficiency and loss of original data structure information is generated. 2D non-negative matrix factorization (2DNMF) technique is introduced. The time-frequency image of the vibration signal is obtained by S-transformation, and the time-frequency image is compressed by 1DNMF and 2DNMF respectively, and the compressed image information is classified, and the vibration signals of the diesel engine in 8 states are collected, and the nearest neighbor classifier is adopted. The naive Bayes classifier and the support vector machine classifier are used for experimental comparison. The results show that the 2DNMF is more efficient and accurate in fault diagnosis than the original 1DNMF.