Abstract
Due to material multiplicity, rigidflexible hybrid and environmental variations, and so forth, the properties of the tracks will be inevitably changed due to the cyclic loads temporally and spatially. In this paper, a theoretical study is conducted to clarify the influence of the longitudinal inhomogeneity of track stiffness on system responses and then a general state estimation method is proposed to inversely predict the parametric distribution of the tracks. To achieve these purposes, a 3D nonlinear vehicletrack coupled model is firstly constructed, where track systems are established by the finite element method with random parameters. Based on this dynamic model, the distribution characteristics of system responses due to random parametric excitations are properly revealed. The numerical studies show that the longitudinal inhomogeneity of track supporting stiffness has significant effects on system responses; besides, the response distributions are varied against different parametric distributions. Finally, a representative example is demonstrated to show the effectiveness of the proposed track state estimation method.
1. Introduction
Railway tracks serve as an essential element in supporting the running vehicles and guiding their dynamical motions. Compared to vehicles, the railway tracks are characterized by the complexity in material multiplicity, rigidflex hybrid, longitudinal large scales, environmental changes, and so forth. Apart from the construction deviations, the track status, generally regarded as the track geometric such as track irregularities, and parametric status, such as the stiffness softening, are constantly affected by the cyclic vehicle loads and are nonuniformly and randomly evolved against the railway operations.
In most of the vehicletrack dynamic simulations [1–8], track irregularities at time or frequency domain are treated as the main system excitations, and the vehicle and the tracks are considered as steady and constant systems. The longitudinal inhomogeneity and randomness of the tracks and their dynamic influence are neglected. In the field of vehicletrack dynamics nowadays, irrespective of it being in localmicro or in globalmacro, it seems to be a growing trend to meticulously take into account the parametric excitation in a way approaching the essences. Generally, the parametric excitation is originated from the inhomogeneity of system properties, especially with uncertainties.
In previous researches, the cognitions of the track parameter unevenness are mainly characterized from frequency domain. Wu and Thompson [9] investigated the wheel/rail interactions and system responses considering the spatially varying stiffness of a ballasted track using a quasistatic method. In their method, the timedomain model is equivalently transformed from the spacevarying receptance based on the linearity of the tracks and wheelrail interactions. Nélain and Vincent [10] carried out a deep research into the influence of sleeper periodicity, randomness and variance of the sleeper stiffness, and the foundation stiffness along the track on the parametric unevenness. Additionally, Verachtert et al. [11] investigated the relation between the unevenness of slab or track bed and the perceived unevenness derived by a train in frequency domain.
Specific to systematic dynamics, Mao et al. [12] conducted an interesting work in random dynamic analysis of a trainbridge system, where the random system parameters are considered, while the longitudinal inhomogeneity of the system is not involved; similar work can also be found in [13, 14]. Xu and Zhai [15] made an extensive research on vehicletrack stochastic dynamics, in which the randomicity of the vehicletrack system in excitations and dynamic parameters is wholly considered. Besides, the full view of track irregularities is depicted in matrix forms to construct a highly efficient probabilistic model. From the perspective of microanalysis, Corrêa et al. [16] regarded the granular medium as a stochastic heterogeneous continuum medium and calibrated a heterogeneous continuum model to model the railway ballast samples.
However, a universal method where the nonlinear wheelrail contacts and the random track system with parametric unevenness along the track subject to a vehicle loading are not fully considered for studies in threedimensional (3D) space. Besides revealing the dynamic effects of track inhomogeneity in vehicletrack dynamics, another challenge that derives from the influenc of track inhomogeneity on system response is how to estimate the track parametric status from the responses conversely. Although it might be possible to experimentally detect the track parameters or support stiffness within a small track section, it is difficult to measure the parametric state at large scales.
Au et al. [17] described a method for the vehicle parameters subjected to multispan continuous bridges using genetic algorithm, where the vehicle is modelled as a 2degreeoffreedom system. Similar work can be consulted in [18, 19]. These approaches are practical rather than simplified systems only. Recently, Onat et al. [20] proposed a method to estimate the maximum friction coefficient in wheelrail interfaces based on a sigma point Kalmanfilterbased scheme. Up to now, there have been limited reports in literature which provide some theoretical guidance in the estimation of track parametric status, especially the parameters randomly distributed along a large length of the tracks.
In this paper, we pay special attention to clarifying the influence of track parametric inhomogeneity in a spatially longitudinal direction on system responses. Based on the clarification of the dynamic effects of track longitudinal inhomogeneity, a further research on inverse approach of estimating the track parametric status will be proposed. The random theories have indicated that only statistical information is required to construct a random model; thus, the estimation of the track parameters is concentrated on estimating the parametric distributions.
Hence, instead of the experimental researches and validations, the work presented is mainly conducted from a theoretical view using numerical methods.
2. Track Longitudinal Inhomogeneity and Its Coupled Dynamics with the Vehicle
Figure 1 shows two commonly used track types: ballasted track and ballastless track.
In this paper, the railway tracks are modelled as a ballastless system with rails, track slabs, and a winkle foundation. The rails and the track slabs are connected by discrete rail pads, and the cement asphalt (CA) mortar is applied to continuously provide elastic supports. Both the rail pads and CA mortar are treated as linear springdashpot elements.
In the track slab system, the concrete track slabs including the rails are essentially continuum medium in longitudinal direction. Therefore, we will meticulously construct their dynamic matrices using finite element method (FEM).
2.1. Matrix Construction for the Tracks with Inhomogeneity
Here a “Lateral Finite Strip and Slab Segment” method [21] is applied to describe the shape function of the track slab. For arbitrary point within the track slab element (TSE), as shown in Figure 2, its vertical displacement can be expressed aswithin which and denote vertical displacement and angular displacement around axis for point and and denote vertical displacement and angular displacement for point ; is the local coordinate of point ; is the length of TSE along axis.
Meanwhile, the vertical displacement and angular displacement in can be further expressed as
within which and are, respectively, the local coordinate for points and ; is the width of TSE along coordinate. is the angular displacement around axis.
Accordingly, by combining equations (1) and (3), the interpolation function for vertical displacement of o can be written by
It is assumed that the dynamical parameter, for example, vertical stiffness of the cement asphalt mortar, follows a function . Obviously, the stiffness matrix induced by the vertically elastic interactions between the track slabs and the subgrade can be calculated:
Certainly, the matrix induced by the elastic deformation energy of the thinplate track slab element can be obtained bywithin which denotes Young’s modulus of the track slab; denotes Poisson’s ratio; denotes the height of the track slab.
To other matrices of the track components (e.g., the rails) with continuum materials, the same manner as above can be applied, in which one needs to replace the specific parameters with spatially varied function (discrete).
The detailed expressions have been listed in [6]; no more illustrations are therefore given for brevity.
Two other issues for the construction and computation of the random track system in large length will be encountered:(i)Computational consumption for large matrices(ii)Boundary condition in longlength calculation
For solving the above matters, a cyclic calculation method presented in [22, 23] will be applied in the computations.
2.2. Construction of Coupled Dynamics with the Vehicle
Following the work presented in [1], the track vibration accompanied by its coupling effects to a moving vehicle or train can be depicted mechanically and physically, as shown in Figure 3. In this work, the vehicle is considered as a multirigidbody system, including one body, two bogie frames, and four wheelsets. Each body has five degrees of freedom, that is, bounce , transverse , roll , yaw , and pitch .
(a)
(b)
Without loss of generality, the dynamic equations of motion for the vehicletrack interactions can be expressed byin which the subscripts “” and “” denote the subsystems of the vehicle and the track, respectively; , , and denote the mass, damping, and stiffness matrix, respectively; denotes the loading vector.
In equation (9), the details of dynamic matrices in the equations of motion have been presented in [6].
Besides, the interactions between the vehicle and the tracks are mainly transmitted through the wheelrail interfaces, where the wheelrail contacts/creepages execute complex motions.
In this work, the wheelrail nonlinear contacts at threedimensional (3D) space are depicted. A wheelrail coupling model [1] is introduced, as shown in Figure 4. In this model, the relative motions between the full degrees of a wheelset and the vertical, lateral, and angular motions of the rail are accurately constructed from geometries and algebraic equations with nonlinearity.
The dynamic loads derived by this wheelrail coupling model have been listed in [15].
2.3. Simple Comparisons with FASTSIM
Wheelrail interaction forces have been known as the key technique of vehicletrack dynamics, which is also an aspect that is notoriously difficult to quantify theoretically. In this part, a comparison between this dynamic model and FASTSIM by Kalker [24] is presented with respect to wheelrail forces, where track irregularities are used as system excitations.
Figure 5 shows the comparisons of wheelrail longitudinal and lateral forces, respectively. It can be clearly observed from Figure 5 that the wheelrail forces derived by this model agree well with the results of FASTSIM. Accordingly, the timevarying process of wheelrail interactions including amplitude distribution and frequency components can be wholly demonstrated. Further validations about this vehicletrack interaction model have been presented in [6].
(a)
(b)
3. A General Method to Estimate the Status of the Tracks
The reality of track parameters can be expressed aswith , where the subscript “” denotes the time point, “” denotes the mth track parameter, and there are spatial values for each track parameter vector , .
The statespace equations of the vehicletrack system can be further written by
In the above equation, denotes the timevarying process of the system responses; denotes the measurement function, which will be substituted by the vehicletrack model here. Thus, will denote the system responses. denotes other system factors; represents noise signals.
Obviously, it is a difficult work to measure the real status of tracks, especially with large scales. One is therefore expected to inversely and approximately estimate track status (the distributions of track parameters) from the conventional dynamic responses and estimated status, namely, and .
The system responses are mainly induced by two factors: wheel/rail irregularities and inconstant mechanicalphysical properties longitudinally and evolved. Therefore, the noise in equation (11) contains both the measurement error and the effects of irregularities.
In this present study, an efficient vehicletrack interaction model, denoted by , had been fully constructed, where the physicalmechanical behaviors and interactions of the vehicletrack systems can be accurately depicted. The system responses derived by the theoretical model can therefore be calculated bywhere denotes other system parameters including wheel/rail irregularities.
Now a problem will be firstly encountered: we have no idea what the longitudinal distribution of the track parameter looks like. By applying a multicorrelatedvariable Monte Carlo method [15], we can randomly generate a pseudoparameter matrix:with , in which the superscript “” denotes the simulated value; denotes the total number of samples; and denote the minimum and maximum correlation coefficients for the th parameter, respectively; denotes the correlation coefficient between the th and the th parameter columns.
Loading these parameter samples into the vehicletrack interaction model, one can obtain the system response matrix consequently; that is,
In reality, the real track parameters follow a specific probability distribution, namely, , and is the probability function of the parameter . Thus, the key work lies in building up the relationship between , , and and estimating .
In stochastic analysis, we more likely intend to clarify the distribution of a variable in probability rather than attempting to estimate the deterministic values. Following the basic principle, the closer the correlation of system responses between simulation and measurement is, the more approachable to reality system parameters used in simulation are.
The steps below are therefore designed to achieve the estimation of : (1)To increase the precision of estimation, the comparison between responses of calculation and measurement, that is, and , are conducted within multiple time segments, , .(2)The uncertainty of the responses will be estimated by the correlation between the calculation and the measurement at time portion . where the subscript “” denotes the type of the responses, calculation, or measurement: , respectively; denotes the system responses, ; denotes the estimator of covariance and denotes the standard deviation. Obviously, is a matrix with order of , while is a matrix with order of .(3)The weight coefficients evaluating the contribution of the parameters on system responses can be obtained by with , , where is the variance of the measurement noise, , and is the posterior distribution of the response under certain track conditions.(4)Based on equation (16), the estimated distribution of the parameters will be assessed bywhere indicates the probability distribution of the parameters in the parameter matrix.
Finally, the estimation of can be approximated by assuming .
Since has been estimated approximately, an acceptancerejection algorithm, as illustrated in Algorithm 1, can be introduced to randomly simulate the spatial series of track parameters .

4. Numerical Examples
In the numerical examples, it is assumed that the vehicle runs with a constant velocity of 300 km per hour on the tracks. The main parameters of the vehicle and the tracks are listed in Tables 1 and 2, respectively [1].
4.1. Influence of the Longitudinal Inhomogeneity of Track Stiffness on System Responses
To clearly reveal the dynamic influence of the longitudinal inhomogeneity of track stiffness on system responses, track profile irregularities will not be taken into account in the vehicletrack interactions in this subsection. It is assumed that the values of a track parameter satisfying specific probability characteristics are randomly distributed along the railway line instead of a constant.
Three types of probability distributions are considered: uniform distribution , normal distribution , and lognormal distribution . To guarantee the comparability for different conditions of property inhomogeneity, the random parameter values should be within almost the same ranges. The equations below are therefore satisfied:where denotes the cumulative probability function and the superscript “” denotes the inverse function.
Take the response distribution of car body accelerations, wheelrail forces, displacements, and accelerations of the rail and the track slab as examples. For normal distributions, the coefficient of variation (CV) is denoted bywhere is the standard deviation and is the mean. in this example.
Figure 6 illustrates the statistical mean PDF of wheelrail vertical force, rail vertical displacement, and trackslab vertical displacement with 200 samples simulated by different forms of distributions of rail pad vertical stiffness. From Figure 6, one can observe that the longitudinal variability of rail pad vertical stiffness shows a significant influence on the probabilistic distribution of system vertical vibrations. Besides the variation of divergence and response amplitude of a specific index accompanied by the changes of the parametric distribution, the longitudinal uniform distribution exerts more effects on the variance of system responses compared to the other two, normal distribution and lognormal distribution.
(a)
(b)
(c)
For quantitatively characterizing the uncertainty of a random signal, the concept of entropy can be introduced bywhere denotes the entropic value, denotes the mathematical expectation, denotes the logarithm with bottom , and denotes the probability function.
As an example, Figure 7 shows the entropic distribution of the wheelrail vertical force, from which one can observe the following: first, the uncertainty of the wheelrail vertical force excited by the uniformly distributed rail pad vertical stiffness is significantly higher than those excited by normal and lognormal distributions; second, it can be observed from Figure 7(b) that the statistical uncertainty of system responses fluctuates slightly, which means that once the parametric distribution is confirmed, we can approximately estimate its influence on the distribution of system responses.
(a)
(b)
Figure 8 further plots the PDF of wheelrail lateral force and rail lateral displacement affected by the deviations of rail pad lateral stiffness. In this example, for showing the influence of the magnitude of rail pad lateral stiffness on system lateral vibrations, the stiffness coefficients following lognormal distribution are given higher values, as shown in Figure 9.
(a)
(b)
It can be observed from Figure 8 that the longitudinal inhomogeneity of rail pad lateral stiffness also shows distinctive effects on the distribution characteristics of the wheelrail lateral force and the rail pad lateral displacement. Obviously, the larger the lateral stiffness of the rail pad, the smaller the wheelrail lateral force, and the larger the rail lateral displacement. It is similar to the results derived above where the uniform distribution shows higher influence on the uncertainty of system response compared to the other two distributions.
In Figure 10, the PDF of rail vertical displacement is exposed, from which one can observe that the longitudinal distribution of the rail pad lateral stiffness has little influence on the system’s vertical vibrations.
The results for the dynamic response induced by the longitudinal inhomogeneity of vertical stiffness of the cement asphalt mortar (CAM) (subgrade support layer) are shown in Figure 11. The vertical stiffness of the CAM shows great influence on the responses of track components, for example, the rails and the track slabs. However, its influence on the wheelrail interaction forces is relatively small.
(a)
(b)
(c)
4.2. Estimation of the Distribution of Track Parameters
From the numerical analysis above, one can recognize that the rail pad stiffness, especially the vertical stiffness, shows significant influence on wheelrail forces and rail vertical displacement. Hence, we therefore put an emphasis on estimation of the distribution of rail pad vertical stiffness.
The track irregularities used in vehicleslab track interactions from a highspeed line are shown in Figure 12. Besides, the longitudinally nonuniform rail pad vertical stiffness is assumed to follow normal distribution with CV of 0.05, 0.10, and 0.20.
(a)
(b)
In the estimations, the following computational conditions are considered:(1)Only track irregularities are loaded as the system excitations, where the system responses are denoted by (2)Track irregularities and randomly distributed rail pad vertical stiffness (which is assumed as the real distribution of “measurement” in the theoretical analysis) are loaded as the system excitations, where the responses are denoted by (3)The responses excited by a group of correlated variables of rail pad stiffness and the specific track irregularities are denoted by
Figure 13 plots the timedomain responses of rail vertical displacement with CV = 0.2, from which one can observe that the rail vertical displacement changes significantly against time once with the longitudinal inhomogeneity of vertical stiffness of rail pads, that is, versus . The response assemblage, that is, , is the result of a set of excitations (parametric and irregularity), which has properly covered the responses and .
According to the method presented in Section 3, we can estimate the probability density function (PDF) of the vertical stiffness of the rail pad, as shown in Figure 14. As seen from Figure 14, the estimated distribution of the rail pad vertical stiffness coincides well with that of assumption in various variations: CV = 0.05, 0.10, and 0.20. Through probability entropy calculations, the entropy difference between estimation and assumption is, respectively, 0.002, 0.01, and 0.008 for conditions CV = 0.05, CV = 0.10, and CV = 0.20. From the analysis in Section 4.1, one has recognized that once the longitudinal distribution of a track parameter is determined, the system responses it caused can be approximately estimated with a slight deviation.
(a)
(b)
(c)
Undoubtedly, one can also estimate the distribution of the rail pad vertical stiffness through the wheelrail vertical forces. The corresponding results are shown in Figure 15, from which it can be clearly observed that estimation of the stiffness distribution also approaches those of assumption.
(a)
(b)
(c)
From the numerical examples, it can be seen that the present method is highly effective. However, limitations have also been notified in its implementation to engineering practices; that is, the CV of track parameters is a prerequisite in the simulation. An abundant measurement data is therefore required to build the mapping relationship between system responses and track parametric distributions.
5. Conclusions
In this paper, the dynamic influences of the longitudinally distributed track parameters on system responses are systematically studied using stochastic and statistical approach, and a new approach for estimating the parametric distribution of the tracks was developed theoretically.
From the numerical studies, several conclusions can be drawn accordingly:(1)The longitudinal inhomogeneity of track stiffness shows significant influence on vehicletrack interactions. Besides, the distribution characteristics of the system responses are varied against the difference of distribution of track stiffness.(2)The mean entropy value of wheelrail vertical force is increased by about 8% when it is assumed that the rail vertical stiffness follows uniform distribution instead of normal and lognormal distribution. Besides, it is observed that the response dispersion degree tends to converge when the random samples are larger than 50.(3)The nonuniformity of track vertical stiffness shows larger influence on wheelrail forces and track structural deformations in comparison with the lateral stiffness.(4)It could be a practical way to assess the effects of random distribution of track parameters on the uncertainty of system responses.(5)The state estimation method newly developed in this work is fairly effective in estimating the track parametric distributions.
In the future, three aspects are urgently needed to study: the first is a more physically approachable model to simulate the vehicletrack dynamic performance with respect to spatial variability and nonlinearity of track properties, the second is the development of a highly accurate and efficient measurement method to detect the vehicletrack system responses, and the third is an exploration of practical methods to estimate the track parametric distribution through the vehicular responses in the condition of low signaltonoise ratio.
Data Availability
Data will be available upon request to the corresponding author.
Conflicts of Interest
The authors declare that they have no conflicts of interest.
Acknowledgments
This work was supported by the National Natural Science Foundation of China (no. 52008404 and 52008382), Earthquake Research Fund of CEA (2019B01 and 2019C09), Heilongjiang Provincial Natural Science Foundation of China (no. LH2019E094), and Program for Innovative Research Team in China Earthquake Administration.