Dynamic analysis of railroad track faults based on neural network algorithms

Rolling bearing is the core component of rotating machinery, and its state is closely related to the operation of machinery. This paper takes the transmission spindle of the railroad track as the research object, and establishes a railroad track fault dynamics model from the fault dynamics analysis of the railroad track, which is the principle of fault characterization, and proposes a railroad track fault diagnosis model based on the neural network algorithm. Firstly, the model construction idea is analyzed by combining kinematics and dynamics, and then the time-varying displacement function is used to simulate the topographic characteristics of the fault area of the railroad track, and the fault information can be introduced into the fault dynamics of the model for solving simulation. The next step combines wavelet transform, CNN-BiLSTM, soft attention mechanism, and Dropout technique. The one-dimensional vibration signal is converted into a two-dimensional image that contains fault features. The attention mechanism is introduced into BiLSTM as a way to extract more accurate fault features, and the Dropout technique is applied to suppress the overfitting of the algorithm. Finally, the Softmax classifier is utilized to complete the diagnosis of railroad track faults. The experiments are designed on the relevant data set, and the experimental results show that the fault feature frequency obtained through the kinetic modeling of railroad track faults is not far from the theoretical results, and the maximum error of the outer and inner ring fault bearings and rolling element fault bearings is less than 1%, which is within a reasonable range.The combination of the BiLSTM network and the attention mechanism makes the distribution of the 10 kinds of railroad track fault features gather very tightly. It shows that the model in this paper based on a neural network algorithm can achieve efficient and accurate diagnosis of railroad track faults.