Here,aiming at problems of low efficiency,long time-consuming and complex testing process of subjective evaluation method for vehicle creeping flutter,a method for extracting temporal features and time-frequency domain features of creeping flutter signals was studied.Combining spatial processing ability of 2D-CNN with temporal processing ability of 1D-CNN,a creeping flutter evaluation method was proposed based on a dual-stream convolutional neural network(DSCNN)under fusion features.The input of one branch was time series features extracted with variational mode decomposition(VMD),while the input of the other branch was image features extracted with fast Fourier transform.1-D temporal features were fused with high-dimensional image features to train a model for scoring.This method could fully capture local waveform features and spatial texture features of creeping flutter by fusing information from different modes.The results showed that the 8-classification accuracy of the scoring model fusing the two features can reach 87.13％to verify the effectiveness of the feature fusion method in evaluating creeping flutter.
维普
万方数据
