Use this URL to cite or link to this record in EThOS:
Title: Fault diagnosis for a new generation of intelligent train door systems
Author: Dassanayake, Hemendra Parakrama Bandara
Awarding Body: University of Birmingham
Current Institution: University of Birmingham
Date of Award: 2001
Availability of Full Text:
Access from EThOS:
This thesis presents the fault diagnosis of an electric train door system. To operate an efficient rail system it imperative that equipment such as train doors are maintained thoroughly, preferably using a predictive maintenance scheme. Initially, a life size testrig was constructed and information regarding frequent door faults collected from both the maintainers and manufacturers. Thirteen faults were induced on the test-rig and a comprehensive dataset collected. It was found that the main incipient faults are related to changes in the friction. By examining the steady state operating conditions it was found that the friction could be adequately described by first order characteristics. The dynamics of the electro-mechanical system can be described by a basic second order differential equation. The most logical approach to diagnose physical faults is to estimate the physical parameters of the system. For this purpose, various continuous-time model identification techniques are discussed. State variable-pre-filtering and linear integral pre-filtering can be used to solve the problem of noise-accentuated derivatives; these filters are also convenient for on-line implementation. However, linear integral filtering is much more straightforward in design and computationally faster for lower order systems. The prefilters are carefully designed to accommodate fast system responses under fault conditions; however, adaptive type pre-filtering provides an alternative solution. A neuro-fuzzy svstem that processes the parameter estimates is proposed to isolate faults', such a system combines the strengths of neural networks and fuzzy inference systems. Initially, a clustering algorithm is used to identify a fuzzy model, which is implemented in a neural architecture for training. Prior to training, redundant membership functions are removed, increasing the readability of the inference system. The neuro-fuzzy system has high classification success and a transparent rule-base that can be used by maintainers. Fault identification is achieved through direct implication of the parameter estimates. To make practical implementation feasible, it is proposed that a three-layer distributed hierarchical architecture be used. This incorporates the manufacturer's control alarms, dedicated fault diagnosis algorithms and the train management system for data communication; an interface to the maintenance system is also available. The architecture is generic and could be used for fault diagnosis of multiple low-cost assets.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available