In order to identify the running state of rolling bearings during the operation of CNC machine tools and improve the fault state diagnosis rate of rolling bearings, an improved genetic algorithm based on wavelet packet decomposition was proposed to optimize the BP neural network fault identification method. Taking four fault states of rolling bearing as the research object, the vibration signal was decomposed by using wavelet packet, and the sensitive feature vector was obtained. Aimed at the shortcomings of BP neural network, the improved genetic algorithm was used to optimize the threshold and weight of BP neural network. The optimal training was achieved, and a more accurate IGA-BP state prediction model of rolling bearings was established. The results show that the IGA-BP prediction model has a faster convergence rate and higher prediction accuracy the effectiveness of the proposed method is proved