Predicting Torsional Resonance while driving on rough roads based on Time Series Analysis

Abstract

Electric vehicles (EVs) are prone to torsional resonance when driving on rough roads, which can cause excessive loads on the vehicle's drive shaft due to the resonance between the vibrations of the motor and tires. However, conventional damping technologies are limited by their ability to detect resonance only after the vibration amplitude has reached a certain threshold, leading to significant loads on the shaft at the time of detection. To address this limitation, this study proposes a method for predicting drive shaft vibration that will occur one second or more in the future using a network that calculates the attention between recursive features and features obtained by temporal convolution from the measured motor rpm. To evaluate the predictive performance of the proposed method, 976 pieces of driving data were generated using a vehicle motion simulator. The correlation coefficient between the true vibration waveform and the predicted waveform obtained using the proposed model was found to be 63% higher than that of the baseline model. These results suggest that the proposed method has the potential to improve the detection of torsional resonance in EVs and thus reduce the risk of excessive loads on the drive shaft.