Proceedings of the Institution of Mechanical Engineers Part F Journal of Rail and Rapid Transit · 2015 · 70 citations · 21 references
EngineeringElectrical EngineeringSystems EngineeringCondition AssessmentMachine LearningMaintenance EngineeringCondition MonitoringCondition-based MaintenanceFailure DetectionFault DetectionFault DiagnosisSupport Vector MachinesCondition Monitoring SystemsEdit DistanceEquipment Condition MonitoringFailure DiagnosisFault ClassificationOnline Fault DetectionOnline MonitoringRailway Systems Engineering
Failures of railway point systems (RPSs) often lead to service delays or hazardous situations. A condition monitoring system can be used by railway infrastructure operators to detect the early signs of the deteriorated condition of RPSs and thereby prevent failures. This paper presents a methodology for early detection of the changes in the measurement of the current drawn by the motor of the point operating equipment (POE) of an RPS, which can be used to warn about a possible failure in the system. The proposed methodology uses the one-class support vector machine classification method with the similarity measure of edit distance with real penalties. The technique has been developed taking into account specific features of the data of infield RPSs and therefore is able to detect the changes in the measurements of the current of the POE with greater accuracy compared with the commonly used threshold-based technique. The data from infield RPSs, which relate to incipient failures of RPSs, were used after the deficiencies in the data labelling were removed using expert knowledge. In addition, possible improvements in the proposed methodology were identified in order for it to be used as an automatic online condition monitoring system.
21
Estimating the Support of a High-Dimensional Distribution
Bernhard Schölkopf, John Platt, John Shawe‐Taylor et al. · Neural Computation · 2001 · 5.9K citations
Scaling and time warping in time series querying
Ada Wai-Chee Fu, Eamonn Keogh, Leo Yung Hang Lau et al. · The VLDB Journal · 2007 · 220 citations