Robust nonlinear analytic redundancy for fault detection and isolation of robotic systems

Abstract

The demand for automation in modern society has significantly increased during the last few decades. Robotic systems play an important role in automation industries that include manufacturing, assembly, and biotechnology among others. In addition, there is a growing need for unmanned operation in different services and research sectors such as search and rescue operation, nuclear waste clean-up, and planetary exploration. Robots can perform repetitive tasks efficiently and can function in a harsh and unsafe environment. However, robots are susceptible to system faults. Faults may result in mission failures that are costly in mission critical enterprises. Therefore fault detection and isolation (FDI) is important for reliable and safe robot operations. In this dissertation, we present a new approach, called the robust nonlinear analytic redundancy (RNLAR) technique, to sensor and actuator FDI for input-affine nonlinear multivariable dynamic systems in the presence of model-plant-mismatch and process disturbances. Robust FDI is important because of the universal existence of model uncertainties and process disturbances in most systems. The new approach is based on analytic redundancy relation, which has primarily been used in the linear domain. The proposed RNLAR technique extends the current state-of-the-art in analytic redundancy relation-based FDI into the nonlinear domain. The RNLAR technique is used to design primary residual vectors (PRV) to detect actuator and sensor faults. Primary residual vectors are designed in such a manner that they are highly sensitive to the faults and less sensitive to model-plant-mismatch and process disturbances. The proposed methodology is applied to the actuator and sensor fault detection of a wheeled mobile robot as well as a robotic manipulator. The order of redundancy relation is used to characterize the robustness of the RNLAR technique. It is proved that an increase in the order of redundancy relation increases the robustness of the RNLAR technique. This result extends the existing relationship between the order of redundancy relation and robustness from the linear domain to the nonlinear domain. Finally, a robust fault isolation technique is presented in this work. The PRVs are transformed into a set of structured residual vectors (SRV) for fault isolation. Experimental results on a Pioneer 3-DX mobile robot and a PUMA 560 robotic manipulator are presented to justify the effectiveness of the RNLAR technique.