Model-Based Fault Detection and Isolation for a Novel X-by-Wire Road Vehicle Architecture

Faults in actuators and sensors of vehicle dynamics control systems can result in erroneous control, leading to degraded control performance or even loss of vehicle stability. While many faults in a vehicle dynamics system can be adequately diagnosed at a component level, treating the diagnosis problem using model-based methods on a global vehicle level can offer significant improvements in detectability and sensitivity for certain faults. This thesis presents two model-based approaches for fault detection and isolation (FDI) design, each using a vehicle dynamics model which cover different operating regions of horizontal vehicle motion. The first approach is based on the linear parameter-varying (LPV) single track model (STM), which adequately models cruising operation with moderate lateral accelerations, low path curvature, and time-varying longitudinal velocity. The design method also accounts for parameter uncertainties arising from variabilities in the tyre cornering stiffness. This application motivates the main theoretical contribution of this thesis, which is concerned with developing a generic FDI design approach for uncertain LPV systems. Methods are proposed to cover the components of the residual-based FDI with structured residuals, comprising the residual generator, single residual evaluation function, and residual pattern evaluation for fault isolation. The residual generator design method takes a reference model following approach, in which the fault-to-residual responses are optimised to follow a specified reference. The synthesis procedure exploits recent developments in robust LPV estimator synthesis that account for uncertainties by means of Integral Quadratic Constraints. To reduce the potential conservatism caused by fixed residual references, this work extends the common formulation by allowing magnitude parameters from the reference model to be optimised concurrently in the semi-definite programs of the synthesis. For residual evaluation, the proposed norm-based evaluation function follows the same robust control paradigm, and a model-based procedure is proposed for the determination of evaluation and threshold parameters. Lastly, for residual structure design, an algorithm is presented to find the smallest set of feasible residuals that achieves the best possible fault isolation. These residual generation and evaluation methods are then applied to the uncertain LPV STM to design the FDI functions. The second presented FDI design approach is based on a nonlinear kinematic model (NKM), which relates wheel speeds and steering angles to the horizontal motion of the vehicle body. The absence of linearisation makes this model valid over all motion directions, but only under the constraint that the tyre slips are negligible. In order to account for the influence of deviations from the model such as small tyre slips and measurement tolerances, this work presents a method to check the feasibility of sets of measurements with respect to the NKM, under the assumption that the deviations are bounded. Functions are also provided to recognise and exclude operating states when low tyre slip is improbable. Fault isolation is achieved by means of structured residuals. After analysing the fault diagnosis requirements of the German Aerospace Center's (DLR's) ROboMObil, a highly actuated and manoeuvrable X-by-Wire test platform, the two model-based FDI approaches are implemented for this application. Their effectiveness is demonstrated using simulations and experimental data