With manned expeditions to planetary bodies beyond our own and the Moon currently intractable, the onus falls upon robotic systems to explore and analyse extraterrestrial environments such as Mars. These systems typically take the form of wheeled rovers, designed to navigate the difficult terrain of other worlds. Rovers have been used in this role since Lunokhod 1 landed on the Moon in 1970. While early rovers were remote controlled, communication latency with bodies beyond the Moon and the desire to improve mission effectiveness have resulted in increasing autonomy in planetary rovers. With an increase in autonomy, however, comes an increase in complexity. This can have a negative impact on the reliability of the rover system. With a fault-free system an unlikely prospect and human assistance millions of miles away, the rover must have a robust fault detection, isolation and recovery (FDIR) system. The need for comprehensive FDIR is demonstrated by the recent Chinese lunar rover, Yutu (or “Jade Rabbit”). Yutu was rendered immobile 42 days after landing and remained so for the duration of its operational life: 31 months. While its lifespan far exceeded its expected value, Yutu's inability to move severely impaired its ability to perform its mission. This clearly highlights the need for robust FDIR. A common approach to FDIR is through the generation and analysis of residuals. Output residuals may be obtained by comparing the outputs of the system with predictions of those outputs, obtained from a mathematical model of the system which is supplied with the system inputs. Output residuals allow simple detection and isolation of faults at the output of the system. Faults in earlier stages of the system, however, propagate through the system dynamics and can disperse amongst several of the outputs. This problem is exemplified by faults at the input, which can potentially excite every system state and thus manifest in every output residual. Methods exist for decoupling and analysing output residuals such that input faults may be isolated, however, these methods are complex and require comprehensive development and testing. A conceptually simpler approach is presented in this paper. Inverse simulation (InvSim) is a numerical method by which the inputs of a system are obtained for a desired output. It does so by using a Newton-Raphson algorithm to solve a non-linear model of the system for the input. When supplied with the outputs of a fault-afflicted system, InvSim produces the input required to drive a fault-free system to this output. The fault therefore manifests itself in this generated input signal. The InvSim-generated input may then be compared to the true system input to generate input residuals. Just as a fault at an output manifests itself in the residual for that output alone, a fault at an input similarly manifests itself only in the residual for that input. InvSim may also be used to generate residuals at other locations in the system, by considering distinct subsystems with their own inputs and outputs. This ability is tested comprehensively in this paper. Faults are applied to a simulated rover at a variety of locations within the system structure and residuals generated using both InvSim and conventional forward simulation. Residuals generated using InvSim are shown to facilitate detection and isolation of faults in several locations using simple analyses. By contrast, forward simulation requires the use of complex analytical methods such as structured residuals or adaptive thresholds.
展开▼
