Sensor Fault Detection and Isolation using a Non-linear Planar Kinematic Model of Vehicle Dynamics

Abstract

An approach for fault detection and isolation (FDI) of sensors in chassis control systems is presented. This approach makes use of the analytical redundancy in the non-linear kinematic relationships between wheel speeds, steering angles and the vehicle body motion, which is valid when negligible tyre slip can be assumed. Using a bounded-in-bounded-out approach, bounds for sensor values are computed from a subset of the remaining sensor values and their tolerances. A sensor value lying outside these bounds indicates a fault candidate. The tolerance propagation is achieved by solving a linear programming problem. By estimating a variety of output signals using different combinations of input signals, each residual generator is sensitive to a different subset of sensor faults, thus permitting fault isolation through structured residuals. The effectiveness of this FDI approach is demonstrated using simulation results and experimental test drive measurements from the DLR’s highly manoeuvrable ROboMObil research platform with four wheel individual large-range steering. Because of the latter, state-of-the-art methods cannot be applied.