Vehicle Side-Slip Angle (SSA) Model Based Observer Enhancement Via NN-based SSA Virtual Sensors

To guarantee vehicle stability and maintain control over it, closed-loop control systems such as Direct Yaw Control (DYC) need to have full knowledge about its states Yaw rate $(\dot \Psi )$ and Side-Slip Angle (SSAβ). $\dot \Psi$ can be measured from onboard automotive Inertial Measurement Unit (IMU) sensor. On the other hand, β measurement is a question of research in the literature. In this paper, a hybrid approach is presented to enhance the estimation performance of model based Sliding Mode Observer (SMO) using an NN-based virtual sensor. Regarding the NN-based virtual sensor, deep Neural Networks models (Feed Forward Neural Network (FFNN), Simple Recurrent Neural Network (S-RNN), Gated Recurrent Unit (GRU-RNN), and Long Short Term Memory (LSTM-RNN)) are proposed as virtual sensors for vehicle SSA feedback. The proposed NN models are trained on a vehicle dynamics dataset acquired from IPG-Carmaker from a Mercedes-Benz C-Class 350e. The collected dataset covered a wide range of vehicle dynamics states performing either combined cornering maneuvers (such as ISO driving test scenarios: Double Lane Change (DLC) and a Slalom) at high speeds up to 150 KM/H or normal driving conditions at low and medium range of speeds. Proposed NN virtual sensors will be fed as SSA measured to a non-linear 3 DOF vehicle model with a Dugoff tire model. SSA estimation results of proposed SMO observer is compared to the results of the proposed SMO with SSA NN-based virtual sensor in terms of estimation error criteria. Results showed that proposed hybrid-SSA models enhance the estimation performance of conventional model based observers as SMO.