Inverse Simulation as a Tool for Fault Detection & Isolation in Planetary Rovers

Fault detection, isolation and recovery is crucial in semiand fully-autonomous vehicles, particularly such vehicles as planetary rovers, which are far-removed from assistance, human or otherwise, in the event of a fault. Residual generation is a popular and conceptually simple method of model-based fault detection, comparing estimates of system properties with other estimates or measurements of those properties. Output residual generation is a simple example of this, employing a mathematical model of the nominal system behaviour, which is driven by the measured system inputs. It then produces an estimate of the system output, which is compared to the output of the real, fault-afflicted system in the form of an output residual error. In this paper, the generation of residuals for variables beyond the output is considered. Where output residual generation employs a standard model of the system, other residuals require inverse models of the system or its subsystems. This inversion is achieved using Inverse Simulation (InvSim). The use of InvSim is shown to enable generation of residuals at the input and between subsystems. These residuals demonstrate greater clarity in certain faults than output residuals. Additionally, InvSim can produce different results when driven by different subsets of the output. This functionality permits discussion of an architecture which can employ forward simulation and multiple InvSim modules to generate a large suite of residuals for fault detection and isolation.