Steer-by-wire technology promises to deliver numerous benets, both to auto manufacturers and end customers, making cars that are safer, more ecient, easier to design and manufacture, and more fun to drive. One of the most compelling aspects of steer-by-wire is the potential to improve the safety of vehicles. While the nominal role of a steering system is to reproduce the driver’s steering command at the road wheels, a steer-by-wire system provides the opportunity for a software layer to intervene on behalf of the driver in dangerous driving situations. The simplest example of this is a car that could automatically counter-steer when starting to skid on wet or icy pavement, in order to prevent loss of control of the vehicle. In the case of higher-center of gravity vehicles, such as passenger vans and SUVs, where vehicle rollover becomes a significant safety issue, a steer-by-wire system could prevent the driver from executing a maneuver that would result in rollover.
Despite all of the benefits of steer-by-wire, there are no production vehicles with steer-by-wire on the road today. The potentially catastrophic nature of a steering system failure requires that any replacement for a conventional steering system be extremely reliable. One approach for achieving the necessary level of reliability relies upon a diagnostic system that can quickly and accurately detect and isolate a fault. This information is then used to switch over to a redundant component or a modified control law that can accommodate the fault. This strategy significantly relaxes the reliability requirements of the individual components in the system, without reducing the overall reliability of the system.
The research presented here demonstrates how a model-based diagnostic system can detect a wide range of potential steering system failures without the need for v redundant sensors. In many cases the diagnostic system can detect a steering system fault at a level well below that of driver perception. The performance of this system is demonstrated experimentally on a full-scale steer-by-wire research vehicle, developed here at Stanford.
The task of diagnostic filter design can be posed as an optimization problem, using channel capacity as measure of diagnostic performance. This eliminates the need for hand-tuning and allows design and evaluation of the diagnostic system to precede final construction of the system to be diagnosed. The usefulness of channel capacity as a diagnostic performance metric is experimentally demonstrated, as it can differentiate filter designs that provide good spectral separation of fault information and noise, from those that do not, unlike existing diagnostic performance metrics.