A bearing fault multiple-weight fusion identification method under uncertain speed conditions

As an important component in mechanical equipment, the health status analysis and fault identification of bearing have signification meaning for the operation and maintenance of equipment. Since the speed conditions have a serious influence on the periodic characteristic and statistical feature distribution of vibration signal, The conventional fault diagnosis methods and fault identification theories based on vibration signal have high misjudgement performance for rolling bearing fault classification once the speed information was uncertain. In this manner, based on architecture artificial neural network theory, this paper constructed a new weighted neural network architecture consisted of several different units and proposed a multiple-weight fusion identification method for bearing under uncertain speed conditions. Compared with the traditional networks, a speed-insensitive fault identification model is built by fusing different units to extract and integrating different information of the input samples, and hence the fault recognition accuracy of bearing faults under uncertain speed condition is greatly improved. The feasibility and effectiveness of the proposed method are further validated by experimental data analysis.