Neural Network-Based Virtual Measurement of Road Vehicle Wheel Displacements

In the automotive industry, sensors play a crucial role in enabling the advancement of autonomous vehicles. However, physical sensors often present challenges due to their cost or practical limitations. Consequently, virtual sensors offer a viable and cost-effective alternative, where relevant metrics can be estimated using available data and mathematical models. This paper introduces a virtual sensor designed to measure the vertical displacement of a road vehicle’s wheels, specifically aimed at enhancing the control of the vehicle’s suspension systems. The virtual sensor utilizes a multi-output neural network to predict the displacements of all four wheels simultaneously. Inputs to the network include longitudinal, lateral, and vertical accelerations from an Inertial Measurement Unit (IMU), along with anti-roll torques from both front and rear anti-roll bars, and currents in the actuators of active suspensions. Experimental validation was conducted using data collected from a Range Rover vehicle, demonstrating the efficacy of the virtual sensor in measuring vertical displacements of unsprung masses. Results were assessed through comparison with measurements obtained from a physical sensor, with Root Mean Square Error (RMSE) values confirming the accuracy of the virtual sensor’s estimations.