Track Vertical Irregularity Detection Method Based On Inference Of Belief Rule Base

With the developing and maturing of high-speed rail technology, high-speed railreplaces gradually major traditional rail lines. However the requirements about thesmoothness of the rails become more and more strict, especially track verticalirregularity has a great impact on train vertical vibration, the greater amplitude ofirregularity, the more excessive vertical vibration of train, so the track verticalirregularity is an important parameter of railway track maintenance. Facing with therandomness of track irregularity, the corresponding information processing andmathematical models are a bottleneck to development. The paper presents a practicaland effective way to establish a precise mathematical model that analyzes therelationship between different data by inference algorithm, which can be used toestimate, forecast and decision-making. Belief rule-base inference methodology usingthe evidential reasoning approach (RIMER) is developed on the basis of D-S evidencetheory, decision theory, fuzzy theory and traditional IF-THEN rule base. It hasmodeling ability of data with incomplete, fuzzy, probabilistic uncertainty, subjectivityor objectivity. So the paper puts highest emphasis on modeling complex relationshipbetween the train vibration accelerations in the frequency domain characteristics andthe safety management levels of irregularity and amplitudes of the track, optimizingand testing model. The main contents of the paper are as follows:(1) We introduce briefly the development history of expert system, compositionstructure and feature, and focus especially on belief rule-base model, knowledgedescription, evidence reasoning algorithm and transformation method of inputinformation.(2) Belief rule based inference method for detecting vertical irregularity of railtracks is presented. The belief rule base is used to model the relationship betweenvibration signals in frequency domain collected in the Locomotive (inputs) and thesafety levels of rail tracks (outputs). Using the limited training samples, the originalBRB and its parameters given by experts can be optimized using the least squaremethod. As a result, the irregularity degrees can be estimated by the outputs of theoptimal BRB. Finally, the test using the real vibration data collected from a section ofa certain existing main line shows the effectiveness and feasibility of the proposed method.(3) Belief rule base constructed by the separability measure is presented fordetecting vertical irregularity of tracks. Reference value of BRB can be got bymeasuring separability of sample data. this method can determine the reasonabledivision of vibration signal features using the separability on various features betweenany two categories of amplitude of irregularity. The initial BRB based referencevalues is established and optimized. The test using the real vibration data collectedfrom a section of a certain existing main line in china shows the effectiveness of themethod.