Model Prediction Based Fault Detection Algorithm of Sensor for Longitudinal Autonomous Driving Using Multi-Sliding Mode Observer

This paper describes the model prediction-based fault detection algorithm of a sensor for longitudinal autonomous driving using a multi-sliding mode observer. In order to detect the faults in radar and acceleration sensors used for longitudinal control of autonomous vehicles, a sliding mode observer and model predictive algorithm was used. In an actual driving situation where the subject vehicle drives with the preceding vehicle, the sliding model observer was used to reconstruct the relative acceleration while the model predictive algorithm was used to predict relative values such as relative displacement and velocity. The predicted states were saved in finite time, and relative accelerations were reconstructed based on the multi-sliding mode observer using the predicted states that represent the current state. Based on the predicted states and reconstructed accelerations, the faults in the sensors can be detected. The performance evaluation of the proposed model predictive algorithm was conducted using actual driving data and a 3D vehicle dynamics model