Data-driven Condition Monitoring of Switches and Crossings

Pegah Barkhordari

Research output: Book/ReportPh.D. thesis

42 Downloads (Pure)

Abstract

Railway networks significantly rely on the correct functioning of switches and crossings (turnouts) to safely maximize the flexibility of traffic operation. Each year, railway infrastructure managers allocate vast amounts of money to maintenance activities in order to ensure the reliability and availability of turnouts. Switches and crossings are complex systems whose dynamic performance result from the mechanical interaction of several components. Failure of one or more components can severely affect the performance of the entire system. Due to their geometrical complexity, turnouts are subject to larger wheel contact forces compared to open tracks and thereby their components exhibit a faster deterioration rate and demand more frequent maintenance actions. Maintenance policies currently adopted by infrastructure managers are either periodic or reactive, which give rise to a high maintenance expenditure. The implementation of predictive maintenance policies could help reducing the maintenance cost; however this requires the availability of novel monitoring tools able to diagnose and prognose the state of health of turnout components. Timely detection of the deterioration processes taking place at the hidden elements of turnouts (i.e., railpad, ballast and subgrade) is a challenging task due to the difficulty in directly accessing information without disrupting train operations. The aim of this thesis is to develop novel data-driven condition monitoring systems for the continuous assessment of the quality of ballast and railpad as key elements of the track infrastructure. For this purpose, the thesis pursues two main objectives: the first is to develop a low-complexity behavioral model capturing the dominant dynamics of the turnout when excited by train passages; the second objective is to design a detection system able to provide early warnings of the development of the deterioration processes. Performance of a condition monitoring tool significantly depends on the availability of a reliable model. This thesis proposes low-complexity behavioral models describing the dynamic behavior of the ballast and railpad at different locations along the turnout. The modeling scheme is based on the combination of an advanced signal processing methodfor non-stationary signals (Empirical ModeDecomposition) and a subspace system identification algorithm (N4SID). The low-complexity models are continuously estimated exploiting the train-induced vibration data collected by a track-side measurement system. The obtained models are utilized to calculate the firstandsecondtrackresonancefrequencies representingthedynamiccharacteristics of the ballast and railpad. Statistical representation of the track resonances are then obtained to account for all sources of intrinsic uncertainties imposed by variations in operational conditions (train type, train speed, train load) and environmental conditions (temperature, amount of precipitation, quality of the soil). The statistical models provide the basis for monitoring both temporal and spatial changes in the first and second track resonance frequencies. Successful preventive maintenance requires timely diagnosis of fault inception by means of a continuously available detector to avoid unnecessary or delayed maintenance actions. Two novel detection systems are proposed for monitoring the quality of the ballast and in-service railpads. Based on physical insight and with the help of information gained from the analysis of the statistical models over time, proper features highlighting the development of deterioration processes in the ballast and railpad layers are identified. The first feature relies on obtaining a sequence, based on recursive estimation of the first track resonance frequency, which indicates changes in the properties of the ballast layer over time. The second feature for monitoring the deterioration process of the railpad relies on a load- and temperature-independent residual sequence extracted from recursive estimation of the second track resonance frequency. The generalized likelihood ratio test is utilized to design a detector able to distinguish changes in the distribution of the identified features. The performance of the designed detectors is validated using track vibration data collected over a 2.5 year period in a turnout of the Danish railway infrastructure. The validation results indicate that the designed detectors can successfully monitor gradual degradation of the ballast layer and the quality of the in-service railpads along the turnout. Last, the ballast degradation detector is operationalized to enable determination of the ballast quality class according to the European standard. This is carried out by determining a regression model that maps the detector output into a measurable indicator recommended by the European standard and in use by the track infrastructure manager. The findings of this research project have been disseminated through articles in leading international journals and peer-reviewed international conferences. These articles are enclosed in the thesis.
Original languageEnglish
PublisherTechnical University of Denmark
Number of pages282
Publication statusPublished - 2019

Fingerprint Dive into the research topics of 'Data-driven Condition Monitoring of Switches and Crossings'. Together they form a unique fingerprint.

Cite this