Prediction of Urban Rail Transit Train Door Faults Based on FOA-BP Neural Network Model

[Objective] In order to carry out preventive maintenance before urban rail transit (URT) train doors fail, avoid personnel and property damage, and ensure the operation safety of URT trains, it is necessary to study the pre-diagnosis of URT train door failures. [Method] Taking the abnormal current signal of URT train door before failure as the research object, a FIR (finite impulse response) filter is designed to filter and dimensionally normalize the collected URT train door current signal data; the FOA (fruit fly optimization algorithm)-BP (back propagation) neural network model is used to train the closed state learning sample data of different doors after dimensional normalization, and output test results; FOA-BP results and output results after BP neural network models training are compared and analyzed. [Result & Conclusion] The FIR filter+Hanning Window function method can effectively remove the interference of high-frequency noise and retain the low-frequency signal part that can correctly reflect the current change trend. Compared with the traditional BP neural network model, the FOA-BP model has the advantages of simple training method, short training time and greatly improved diagnostic accuracy. The error between the actual output value and the expected FOA-BP model output value is less than 1%, satisfying accurate diagnosis needs of URT train door faults.