Study on Fusion Estimation of Adhesion Coefficient Based on Multimodal Fusion Recognition Strategy

Estimation of the coefficient of adhesion for pavement conditions with different material properties is a difficult engineering problem. The study proposes a tripartite fusion estimation technology framework based on vision, pavement unevenness information, and vehicle dynamic response. First, an extended Kalman filter estimator based on seven-degree-of-freedom dynamics and the Dugoff tire model is established to estimate the adhesion coefficient value according to the dynamic response of the vehicle; second, a pavement type fusion recognition strategy based on pavement image information and pavement unevenness information is proposed to realize the pavement unevenness prediction through the nonlinear autoregressive networks with exogenous inputs, correct the pavement type recognition results of GoogLeNet convolutional neural network, and look up the table to get the range of adhesion coefficient prediction. Finally, the spatiotemporal synchronization method of pavement recognition and dynamics estimation and the fusion estimation mechanism of the adhesion coefficient are established to achieve an accurate estimation of the adhesion coefficient. The experimental results show that the proposed method can make up for the dependence of the dynamic estimation on the reliability of observations and sufficient excitation, and is more recognizable to the pavement, which improves the estimation accuracy, convergence, and tracking of the adhesion coefficient.