Testing on the Curve: Nonlinear Analytical Redundancy for Fault Detection
Leuschen, Martin L.
Cavallaro, Joseph R.
Walker, Ian D.
One of the most important areas in the robotics industry is the development of robots capable of working in hazardous environments. Providing a high level of functionality in these arenas is important simply because humans cannot safely or cheaply work there. Our work focuses on a fault detection method known as analytical redundancy, or AR. AR is a model-based state-space technique that is theoretically guaranteed to derive the maximum number of independent tests of the consistency of sensor data with the system model and past control inputs. AR is only valid for linear sampled data systems. AR is a model-based technique, and is thus extremely sensitive to differences between the nominal model behavior and the actual system behavior. A system with strong nonlinear characteristics, such as a hydraulic servovalve, can be impossible to model properly in the linear domain, creating significant differences between the model and the system that will generate false error signals. In this paper we discuss the application to a hydraulic servovalve system of our novel rigorous nonlinear AR technique that maintains traditional linear AR's theoretical guarantee of the maximum possible number of independent tests in the nonlinear domain. This technique allows us to gain the benefits of AR testing for nonlinear systems with both continuous and sampled data.