No abstract

A practical, robust method of fault detection and diagnosis of a class of pneumatic train door commonly found in rapid transit systems is presented. The methodology followed is intended to be applied within a practical system where computation is distrib uted across a local data network for economic reasons. The health of the system is ascertained by extracting features from the trajectory pro®les of the train door. This is incorporated into a low-level fault detection scheme, which relies upon using simple parity equations. D etailed diagnostics are carried out once a fault has been detected; for this purpose neural network models are utilized. This method of detection and diagnosis is implemented in a distrib uted architecture resulting in a practical, low-cost industria l solution. It is feasible to integrate the results of the diagnosis process directly into an operator's maintenance information system (M IS), thus producing a proactive maintenance regime.

The results of tests on a new class of closed-loop electric train door were used to construct a mathematical model that could form the basis of a fast-response condition monitoring system. Faults are detected by identifying variations in selected mechanical and electrical hardware characteristics, these being evaluated through parameter estimation. The main obstacle is signal noise, but this can be overcome with linear integral pre-filtering.

A novel approach to distributed fault detection and isolation ( FDI ) is presented. Consideration is given to an appropriate architecture and communication method for a truly distributed FDI system. As part of the discussion on system architecture, regard is given to the correct level at which fault detection, diagnosis and isolation should be carried out.The case study of an automatic door as part of a building automation system is presented. Recognized faults are induced on the asset and detection methods formulated. It is demonstrated that, by implementing the fault detection algorithms at a lower level, it is possible to carry out FDI over a system consisting of many assets. This allows maintenance staV more precise information regarding the health of the whole system and allows system-wide isolation and in-depth diagnostics to be applied in the event of a fault being detected. The architecture discussed allows the implementation of system-wide FDI at a fraction of the cost of stand-alone systems. NOTATIONY output vector over L observations ð parameter vector m i weight vector of the ith winning neuron Á regression vector over L observations n useful data (bytes) ae regression vector containing observations N i * (d ) neighbourhood function de ning indices of of inputs at sample time k the neurons around a geometric distance d of the winning neuron N x,y the neuron at position (x, y) in the Abbreviations rectangular neuron grid structure EP embedded processor QU question frame FDI fault detection and isolation RE response frame FMEA failure mode and eVects analysis S CE controller eVort index FN eldbus node S CP controller performance index LAN local area network S D damping ratio of second-order model MIS management information system S G d.c. gain of second-order model PWM pulse width modulation S N nominal controller index values SOFM self-organizing feature map S P peak step response of second-order model STME single-throw mechanical equipment S T time constant of second-order model