Development of online wheel defect detection methods for high-speed trains

Abstract

The issue of wheel tread defects has become a major challenge for the health management of high-speed rail in recognising the fact that a defect with small radius deviation has the potential to give rise to severe damage on both train bogie components and track structures. It is therefore important to detect the defects soon after their occurrence and then conduct wheel re-profiling for the defective wheelsets. There have been a variety of methods for wheel defect detection, among which a promising solution is the trackside wheel-rail interaction detector which can assess the wheel condition when the train is in operation. This research aims to develop an online monitoring system deployed on rail tracks to detect wheel local defect and wheel polygonisation with the rail response data collected during the passage of in-service high-speed trains. For this purpose, two analytical models for analysing the dynamic response of the vehicle and track structure under the impact excitation of wheels with local defects and polygonal wears. The first model which is developed with accounting for the rotation of the flat-defect wheel, predicts the dynamic strain response of a periodically supported rail under the impact excitation of a wheel with flat-defect in the time domain. With this model, the features of localised anomalies caused by wheel flats are revealed in the time history of rail dynamic strain response. In recognition of this, a fibre Bragg grating (FBG) sensor array is devised for deployment on rail tracks to detect wheel flat defects through capturing the localised response features resulting from wheel flats. The devised monitoring system is then deployed on a railway to verify its capabilityin detecting single and multiple wheel flats. The second model, as a vehicle-track-bridge coupling model, is developed to analyse the wheel-rail interaction and the dynamic responses of car body, bogie frames, wheelsets, rail, trackbed and bridge deck under the input of user-defined wheel roughness spectrum in the frequency domain. With this model, a full understanding of the effect of wheel roughness on different structural components in vehicle-track system can be obtained and a method to assess wheel polygonisation is developed. On this basis, a new sensing paradigm is proposed with three layouts of FBG arrays for wheel polygonisation detection: 1) FBG rosettes installed on the rail web; 2) vertical FBG array deployed on rail web above the sleeper; and 3) longitudinal FBG arrays mounted on rail base. A defect detection algorithm is then developed to identify both potential wheel local defects and polygonal wears based on the online-monitored rail response data. This algorithm consists of four steps: (i) strain data pre-processing by a data smoothing technique to remove the response component which corresponds to the ideal response excited by round wheel; (ii) diagnosis of localised anomalies using Bayesian learning method and outlier analysis technique; (iii) identification of local defects by a further analysis of the localised anomaly features extracted in Step (ii); and (iv) assessment of wheel polygonisation based on the frequency domain analysis of the normalised response data. To verify the methods for both wheel local defect detection and polygonal wear detection, two blind tests are conducted by operating a train with potential wheel local defects and wheel polygonal wears on the instrumented rail, respectively. No information about the locations of defective wheel(s) is provided during the tests. The test results indicate that both the local defects and polygonal wheels can be identified with high fidelity, which are in good agreement with the offline measurements of wheel radius deviation taken in a depot.A comparison is also made in terms of polygonal wheel identification capability of the aforementioned three layouts of FBG arrays, and the results show that all three layouts can successfully identify the polygonal wheel with a roughness of 26 dB (re 1 μm, radius deviation = 0.05 mm) at low train speed. Particularly, the rail base FBG longitudinal layout can even identify the polygonal wheel with roughness lower than 20 dB (re 1 μm, radius deviation = 0.02 mm). Compared with the other two layouts, the rail base longitudinal layout, which can also be used in detection of local defects, is more cost-effective.