Multiple waveforms are received 119, each associated with an operating behaviour of a railway point switch. Characteristic features are extracted 121 which represent a shape of respective waveforms. They are used for constructing multiple feature sets, each corresponding to an operating behaviour of the switch. Logic rules may be generated for each feature set for classifying a real-time waveform. The features may be extracted using wavelet approximation after normalisation of the waveforms. The features sets may form part of a supervised machine learning algorithm, e.g. as the trees of a Random Forest 123. Also disclosed is diagnosing point switch faults by receiving 125 a waveform associated with operation of a switch (e.g. representing current drawn during a swing of the points). Characteristic features (e.g. average current) representing a shape of the waveform are extracted 127. An operating behaviour of the switch during the event is classified 129 based on application of the extracted characteristic features to logic rules and a signal indicates the classified operating behaviour.
Diagnostic system and a method of diagnosing faults
2021-06-23
Patent
Electronic Resource
English
IPC: | B61L Leiten des Eisenbahnverkehrs , GUIDING RAILWAY TRAFFIC |
Automotive engineering | 1991
|Diagnosing munition faults - The EMMA approach
AIAA | 1989
|