Abstract: Aiming at the difficulty to identify the faults of combustion chamber components in diesel engines under multiple working conditions,a multi-level deep stack network（DSN） fault diagnosis algorithm was proposed and verified on-line on Z6170ZICZ-1 diesel engine based on the vibration signal of the cylinder head. Firstly,the construction process of the DSN was analyzed. Combined with the idea of multi-level diagnosis,a multi-level DSN fault diagnosis model was constructed to identify different faults and the degree of faults under various working conditions. Then,a comparison between DSN algorithm and classical extreme learning machine（ELM） and support vector machine（SVM） algorithm was completed based on the accuracy. The experimental results show that the multi-level DSN diagnostic model has better classification effect. Finally,an on-line fault diagnosis system for the marine diesel combustion chamber components was designed based on dSPACE platform and the feasibility and effectiveness of the fault diagnosis algorithm were verified.