In the conventional oscillation fault diagnosis system, the noise of the collected oscillation signal is large, which leads to the low confidence of the fault category. In order to solve this problem, a fault diagnosis system for oil film oscillation of steam turbine based on cogeneration is proposed. Coupled with cogeneration production mode, the redundant exhaust steam of circulating steam turbine unit is used to increase the power generation of the system to complete the hardware design; the wavelet packet de-noising algorithm is used to eliminate the collected oscillation signal noise, decompose each time and frequency band of the signal, sort the time domain sub signal energy value in the frequency band, form the fault signal eigenvector, input the neural network, and output the oscillation fault after training The diagnosis mode completes the software design. The oil film oscillation signals under the conditions of full load, minimum load and noise are collected, and the comparative experiments are set. The results show that the designed system improves the reliability of fault diagnosis, and the fault diagnosis results are more accurate and reliable.