Search Results

A model-based sensor fault detection algorithm is proposed in this paper to detect and isolate the faulty sensor. Wheel speeds are validated using the wheel speed deviations before being employed to check the sensor measurements of the vehicle dynamics. Kinematic models are employed to estimate yaw rate, lateral acceleration, and steering wheel angle. A Kalman filter based on a point mass model is employed to estimate longitudinal speed and acceleration. The estimated vehicle dynamics and sensor measurements are used to calculate the residuals. Adaptive threshold values are employed to identify the abnormal increments of residuals. Recursive least square method is used to design the coefficients of the expressions for adaptive threshold values, such that the false alarms caused by model uncertainties can be prevented. Different combinations of estimations are employed to obtain 18 residuals.

Time-to-collision (TTC) can be used to design the forward collision warning system (FCWS). Conventional TTC is defined as the relative distance divided by the relative velocity. It might result in estimation errors for maneuvers with non-zero relative accelerations. In order to improve the accuracy of TTC estimation, we consider the relative acceleration and extend the region of interest for FCWS from the main lane to adjacent lanes. If we assume constant relative acceleration within the sample time, the relative distance can be approximated using a second order polynomial. Recursive least square technique is employed to estimate associated coefficients. Variable forgetting factor adjusted using relative acceleration is proposed to enhance the approximation accuracies. Similar process is applied to estimate the coefficients of a second order polynomial for the relative polar angle.