Prediction of hunting instability index of high-speed railway vehicles based on a coupled 2DCNN-GRU model

A coupled Two-Dimension Convolutional Neural Network-Gated Recurrent Unit (2DCNN-GRU) model is proposed to evaluate and predict the hunting instability of high-speed railway vehicles in this paper. First, vibration accelerations four measuring points on surface bogie frame a vehicle good working condition with are obtained through line test simulation. The acceleration data under different conditions cut into many pieces at equal intervals. Low-frequency band-pass filtering applied each piece obtain filtered data, which then analyzed separately get sample set spectrum images, including short-time Fourier spectrum, Hilbert time-frequency-amplitude marginal spectrum. Then, 2DCNN extract features by deeply studying images data. root-mean-square (RMS) responses mean value response envelope calculated recorded for piece. Hunting Instability Index (HII) considering weighted RMS quantitatively extent instability. Finally, GRU method predicting dynamic change HII indicators, effectiveness accuracy verified typical examples. One contribution work proposing motion image identification signals, another definition based statistics.