Onboard sensors-based road surface roughness identification using multi-module LSTM-DKF algorithm

To realize the preview control of the intelligent chassis suspension and improve the vehicle ride comfort based on onboard sensors, an accurate and rapid road roughness identification algorithm is proposed, which considers varying road conditions at all wheels using data-model fusion method integration. Multi-module long short-term memory network combined with a discrete Kalman filter (LSTM-DKF) is proposed in this paper. The algorithm employs parallel LSTM neural networks for each wheel, leveraging vehicle response data obtained from onboard sensors. The hyperparameters of the LSTM networks are optimized using a genetic algorithm to ensure accurate identification of road surface levels. Furthermore, the noise matrix within the discrete Kalman filter of each sub-module is iteratively updated based on the identified road surface level. Therefore, multi-module LSTM-DKF can adaptively identify the road surface roughness under four wheels simultaneously in complex road conditions. Simulation and vehicle field test results show that the proposed multi-module LSTM-DKF can quickly and accurately identify the level and profile of road roughness. Compared with the road roughness identification algorithm based on Kalman filter, the multi-module LSTM-DKF can improve the correlation coefficient r of the identification results by 3.11%, and reduce both the root mean square error (RMSE) and maximum absolute error (MAE) by more than 20%. Those outcomes validate the effectiveness of the proposed algorithm. • A data-model fusion algorithm has been proposed for identifying road roughness. • The algorithm can simultaneously detect road level and roughness under all wheels. • It shows strong adaptability to complex and varied road conditions. • It requires fewer sensors, enhancing its practicality and cost-effectiveness.