Research On Estimation Methods Of Track Irregularities Based On Axle-Box Vibration

Measurement of track irregularities has an extremely important significance for track condition management and safety of rail traffic. Railway transportation is developing towards high speed, heavy haul and high density nowadays, therefore leading an increase in the wheel/rail impact forces, a curtailment of the variation period of track condition and a decrease in ride comfort.Current technology of track irregularities measurement has several flaws:high economic cost, the complication of system composition, the occupation of service time and low applying efficiency. Due to all the reasons above, it is necessary to develop a new measurement technology, which should be suited to the trend of railway development and economic efficiency.This paper proposes to estimate track irregularities are based on theoretical background of inverse problem of dynamics, comprehensive application related theory and technology of signal processing, system identification, load identification, vehicle dynamics and on-board testing. By taking the axle-box acceleration as the research object, this paper mainly discoveries among three aspects:extracting and separating impact signal caused by out of round of wheel from axle-box signal, identification of track spectrum in frequency domain by using IPEM,estimation track irregularities in time domain by using NARX neural network.The main conclusions of this article are discussing in following passages.(l)This paper illustrates the significance of measuring track irregularities, introduces track irregularities and its current measuring technology, concludes the application and development situation of domestic/foreign technologies under the measuring of track irregularities, pinpoints that inverse dynamics which composed by system and excitation identification is the theoretical basis, analyzes the relation between axle box vibration signals and track irregularities, and sets up the vehicles and track dynamics model.(2)This paper pinpoints the objectivity that out of round phenomenon exists in serving vehicles’wheelsets, proposing the extracting and separating methods based on signal atomic decomposition theory and improved matching pursuit algorithm. In the beginning, it classifies out of round of the wheelsets and briefly illustrates its mechanism of production, establishing the dynamics model of the vehicles’ out of round; additionally, it introduces some related theories about signal atomic decomposition including the sparse representation of signals, the definition of dictionary and typical composition of atomic and so on, investigating the improved matching pursuit algorithm and its combination with intelligent algorithm. In current track measuring theory, it always assumes that wheel tread is in an ideally even condition, but there are various uneven flaws existed during the measure of virtual servicing vehicles or even track measurement vehicles. Out of round is the most common wheel tread flaw; wheel flat and wheel polygon are the most representative out of round forms. Vehicles dynamics’ simulating calculation results indicate that vibrating signals causing by wheel flat possess the characteristic of periodic pulse decay while axle box vibration caused by wheel polygon possesses the characteristic of sine wave. Introducing the signal atomic decomposition theory to the extraction and separation of the signal of wheel out of round, we can structure the unanimous pulse decay atomics. By using matching pursuit algorithm we can search the pulse decay atomics or harmonic component existed in the signals one by one and effectively detect that whether the component extracted from axle box vibration signals was stimulated by out of round of the wheel by combining atomic parameter, wheel diameter and vehicle’s speed.In order to improve the efficiency of parameters global optimization, it proposes the improved matching pursuit method based on the Particle Swarm Optimization. By analyzing the superimposition of the atomics from extraction in a chronological sense and the atomics superimposition’s time frequency distribution figure by Wigner-Ville’s distribution, it can be seen clearly that the existence of wheel flat or polygon. The results of stimulation indicate that this method can effectively extract and separate the pulse and harmonic component signals introduced by wheel tread out of round. Signal allowance that without interference signals can maximize characterize the condition of the track, providing stable data for next step.(3)This paper proposes the track spectrum’s identification method based on inverse pseudo excitation method. Track spectrum, which is also named track uneven power spectrum, is the manifestation of track uneven, under frequency domain. Track spectrum can reflect the tracks’ holistic even condition from wave length and amplitude, possessing broad applying prospects in the maintenance of tracks. Conventional track spectrum method depends on the test data of track measuring vehicles, confining its application. This treatise puts randomly vibrating inverse pseudo excitation method into an application of the track spectrum’s identification. The results indicate that this method tremendously improve the efficiency of calculating, with higher identification accuracy, leading an easier way to get track spectrum and therefore laying the foundation of next step of popularization and application.(4) In the last chapter estimation of track irregularities in time domin based on GA-NARX neural network are discussed. Vehicle vibration is mainly caused by track irregularities through wheel-rail contact which has strong nonlinear features.For estimating track irregualtion from axle-box acceleration signal,it’s necessary to build inverse model of axle-box acceleration to track irregularities at first. The estimating track irregularities have a great error by using traditional linear ARX model. Therefore, this chapter presents a method to estimate track irregularities using GA-NARX neural network in time domin.Nonlinear ARX model is expanded from ARX model,and NARX neural network is one of NARX model which is using neural network to approximate nonlinear feature.NARX neural network can be regarded as BP neural network increased multi-step input output delay. This model structure can be used to match system dynamic characteristic betterly.Input-output order are important parameters for NARX neural network model,but there is no standard rules for determineing those parameters until now,so that this chaper propose an algorithm which is binding Genetic Algorithm to determine input-output order. Average relative variance is selected as Objective Function. Because the restriction of actual conditions,it’s difficult to obtain synchronous experiment data of axle-box acceleration and track irregularities.In order to close to actual situation,this chapter utilizate MSD.ADAMS/rail, the famous multi-bodies dynamics software to simulate and acquisite synchronous simulation data.Then,using this data to verification proposed algorithm.