The research aims at the serious economic loss and life-threatening problems caused by mechanical and electrical system accidents and bearing failures. The fault detection and identification of bearings in mechanical and electrical systems are discussed. The structure, fault cause, vibration mechanism, and characteristic frequency of rolling bearing are analyzed, and the noise of vibration signals is removed and eliminated in light of the characteristics of the initial fault signal of rolling bearing. Because of the shortcomings of the wavelet, wavelet packet transform is proposed to characterize the normal state of rolling bearing, rolling element fault, inner ring fault, and outer ring fault signal. Based on the characteristics of global optimization of GA (genetic algorithm), the algorithm falls into local optimal value following the defects of BP (backpropagation) neural network and uses GA to optimize the BP neural network algorithm for fault diagnosis of rolling bearings. According to the experimental results, when the evolution algebra of the fault diagnosis model GA-BP is 8 at the drive end, the optimal classification accuracy of the population reaches 98.83%. In this case, a rolling element fault in the test data is misclassified. When the evolution algebra of the GA-BP fault diagnosis model is 2 at the fan side, the overall optimal classification accuracy reaches 97.62% in total. Under this condition, a rolling element fault and an outer ring fault are misclassified in the test data. Through the comparison experiment with the traditional optimized BP neural network, it is found that the GA-BP neural network algorithm model is suitable for the fault classification of rolling bearings.