Maneuver-based deep learning parameter identification of vehicle suspensions subjected to performance degradation

Abstract

A novel parameter identification method was proposed for vehicle suspensions subjected to performance degradation. The proposed method does not require the measurement of the stiffness and damping coefficients of suspensions. Instead, it uses vehicle states to calculate the stiffness and damping coefficients based on an efficient multibody model and inverse dynamics. First, a full-vehicle system was modelled using a semirecursive multibody formulation, and the dynamic properties of suspensions, chassis frame, and tires were considered. Second, dynamic simulations on a bumpy road were performed, and vehicle state data were collected. A deep neural network (DNN) model, whose inputs and outputs were vehicle states and suspension parameters, was developed. The DNN model can estimate the stiffness and damping coefficients based on vehicle states measured by sensor networks. The parameter identification was achieved by deep learning of the relationship between vehicle states and suspension parameters in a given maneuver. Finally, the model accuracy was investigated in terms of different DNN inputs, data samples, and hidden layers. The results showed that the DNN model predicts accurate stiffness and damping coefficients in real time. This maneuver-based parameter identification method can be used for the condition-based monitoring or fault diagnosis of vehicle suspensions subjected to performance degradation.