DC circuit breaker is one of the most important switch devices for controlling and protecting power system, which is reliable operation related to the safety and stability of the power system. The vibration signal can directly reflect the mechanical state of the circuit breaker during the opening and closing. Therefore, the mechanical vibration signal of the DC circuit breaker is selected as the research object. Firstly, this paper studies the influence of vibration sensor selection on vibration signal acquisition and determines the vibration sensor. Then, it studies the vibration signal preprocessing and the method of extracting the characteristic quantity of vibration signal. Finally, several common faults are simulated and simulated faults are diagnosed by Elman neural network. The results show that the DC circuit breaker mechanical state monitoring and diagnosis is realized by wavelet packet decomposition and Elman neural network. This work is supported by National Key Research & Development Program of China (No. 2017YFB1201200).