Road condition estimation based on multiple vehicle dynamics information and its experimental validation

Abstract

There is a lack of experimental verification and a single system input signal in the current reverse recognition of road using vehicle dynamics response, which leads to low accuracy and reliability of road recognition, and cannot achieve accurate recognition of random road grade information. A novel road condition estimation method based on the Convolutional Neural Network and Long Short-Term Memory (CNN-LSTM) hybrid neural network is proposed in this paper, and the vehicle body vertical acceleration, wheel vertical acceleration, and pitch angular acceleration are used as road surface grade recognition inputs. Moreover, the full vehicle seven-degree-of-freedom passive suspension mathematical model is established. Through simulation and real vehicle experiments, the signals of the vehicle system dynamics response are processed, the recognition of random road is realized, and accurate road grade information is obtained. Real vehicle experimental results demonstrate that when the vehicle speed is 10 and 20 km/h, there is a 100% probability that the asphalt pavement is recognized as a B-class road. Overall, compared with the fixed speed model, the recognition accuracy of the variable speed multi-input fusion model is increased by 25.1%. The effectiveness and accuracy of the road grade estimation based on the CNN-LSTM hybrid neural network is verified in the real vehicle, which provides a basis for the intelligent switching of suspension working modes.