An Investigation of Reliability of Hydraulic Robots for Hazardous Environments Using Analytic Redundancy

Robotics for hazardous environments is one of the most important and fast growing areas in the robotics industry [4, 5, 6, 12]. This area focuses on providing a high level functionality in arenas where humans cannot safely or cheaply work. However, the usefulness of the robot in hazardous situations is highly dependent on its reliability. Hazardous environments can damage robotic components, are often sensitive to disturbances that can be caused by malfunctioning robots, and humans usually cannot enter these areas to repair or remove a failed robot. For these reasons, our team has investigated reliability issues for robots extensively [7, 8, 14, 9]. The Rosie mobile worksystem [1, 3] is an important and interesting robot that is on the cutting edge of hazardous environment robotics. (DOE is currently using it in the decomissioning of Argonne National Laboratory's CP-5 reactor facility [1].) Rosie, under development by RedZone Robotics Inc. and Carnegie Mellon University's Field Robotics Center, is a heavy-duty hydraulic robot designed for nuclear reactor decontamination and dismantlement. The robot consists of a wheeled platform containing a central hydraulic power supply powered by an electric tether, four independently steerable wheels, and a heavy-duty crane/manipulator. The hydraulic wheel actuator subsystem has been determined to be a vital component of the mobile platform through reliability analysis. A failure of a wheel mechanism may prevent the removal of Rosie from the reactor work site, which may very well be deadly to potential repairmen. The combination of the hazardous nuclear environment and debris from the decontamination and dismantlement tasks will put considerable stress on the robot system and requires a careful monitoring and diagnostics strategy to reduce the possibility of major failures. Our research into analyzing this robot's reliability through the technique of analytical redundancy (AR) will help provide the Department of Energy (DOE) with a more complete and e ective set of tests for monitoring and diagnostics of the Rosie system. In this paper, we discuss the derivation through AR of a suite of model based tests for the default sensor package for one of Rosie's wheel actuators. AR allows us to exploit the sensor information of the sensors values and the system model to derive tests of the consistency of the sensor data. Some of these tests are comparison of the actual system response to control inputs to predicted response indicated by the model, the other tests uncovered by the AR analysis re ect higher order state interdependencies. These tests and their use in monitoring and diagnostics for Rosie are detailed and examined in depth. This work is also an interesting example of the application of model based techniques for an important class of practical systems.