Start date: 24th September 2001 Finish date: 24th September 2004 (subject to thesis completion)
The relation between humans and machines is a strong one. Over the years we come to depend on the correct functioning of the tools we deploy to produce work for us. When these machine break down we are faced with two prime options: replace the entire machine or repair it. For financial reasons we may not be able to replace every machine that breaks down and in many cases we are left with the option of repairing it.
Repairing damaged machinery could be a costly process involving the use of system experts for both diagnosis and repair. To reduce costs, the diagnosis part of the process could be automated. A proper implementation of an automated diagnostic engine should yield a 'short-list' of possible failure reasons for a given system. This project, which is closely related to the GenMech project, will attempt to provide such 'short-lists' for a range of electro-mechanical systems.
The aim of this PhD research is to develop a generic as possible engine, capable of diagnosing multiple failures with electro-mechanical systems utilising the systems' devices own failure modes. Combining multiple models to account for how such devices might fail should enable the engine to diagnose different types of devices, operating in both two-dimensional and three-dimensional planes. As not much work has been done in model-based diagnosis of mechanical failures we also aim to determine to what extent could the model-based diagnostic approach could be useful in the diagnosis of these kinds of systems. For example, we aim to determine the ability of diagnosing failures in an industrial robot arm, an electro-mechanical system operating in a three-dimensional configuration within the physical world.
The project will:
- Develop a series of electro-mechanical failure modes.
- Develop an engine able to utilise the above modes to produce a causal explanation or in the very least a causal path between the failure initiator and the given observations.
- Create a simulation of relevant systems and test the comprehensiveness of the models used.
- Deploy the engine to diagnose actual physical systems, from a simple two-dimensional configuration to the more elaborate industrial robot arm.
- Evaluate the feasibility of proper diagnosis under uncertain/erroneous input.
- Evaluate the results of the project.
The first year of this project has been dedicated to exploring the field of diagnosis concentrating on model-based techniques. An overview of what has been done in the field from early stages until now can be found in the first year's report.
Into the second year of the project, we are currently investigating how might we build a stable model of cohesion failures, having already established a Kinematic failure model (due to Prof. Mark Lee's previous work).
- Supervisor: Prof. Mark Lee
- Supervisor: Prof Chris Price
- Student researcher: Coby Karp