The influence of vehicle system dynamics on rail foot heat transfer. (2024)

1. Introduction

Modern railways are always looking for new ways to increaseefficiency and drive down costs. Track maintenance is an important issuefor both above and below rail operators. Below rail operators need toidentify and monitor faults as early as possible so intervention cantake place to repair rail flaws before they escalate to an unsafe andcostly level. Above rail operators need guarantees on track availabilityto move their product in an efficient manner. Any fault in the track maydamage rollingstock and, in the worst case, derail and potentiallydestroy parts of a train and that becomes extremely costly for allparties.

A recent trend in track maintenance has seen a move towardsdetection technologies that are integrated into revenue raisingrollingstock. Some examples of this include the Centre for RailwayEngineering's Infrastructure Wagon (McClanachan et al. 2001) andLBFosters' IntelliTrack Technology (2016). There are a number ofbenefits to this approach with the end result being an efficiencyimprovement for both above and below rail operators. The first benefitis the reduction in need for specialised rollingstock equipment fortrack monitoring. This in turn leads to a throughput improvement astrack time is not taken up specifically for maintenance activities. Thesecond benefit is that a solution integrated into general rollingstockis generally cheaper, and may be rolled out across a wider variety ofrollingstock types to further increase the rate at which the track ismonitored.

The majority of research conducted into track maintenancetechnologies is focused on the rail head and Rolling Contact Fatigue(RCF) flaws. While there is significant justification to this research,another part of the rail, the foot, continues to also experience flawsthat lead to broken rails and potentially to derailment. From 1992 to2002 rail foot flaws were the 4th highest occurring flaw type accountingfor 4.6% of the total defects identified by the FRA with a damage billexceeding $US49 million. Although these statistics are over 15 years oldthe lack of development in detection technologies during this time wouldindicate that the occurrence of rail foot flaws has likely not reduced.In Australia, heavy haul rail operators still struggle with rail footflaw detection and are the primary funding source for research into theproblem presented here in this paper.

Rail foot flaws occur for a number of reasons that includemanufacturing defects, incorrect installation and use, and therail's inability to withstand any further fatigue (Cannon et al.2003). There is limited research or products available for foot flawdetection, and even less for integration of detection systems ontogeneral rollingstock.

1.1. Thermography

A common technique in flaw detection for Non Destructive Testing(NDT) is Thermography. A specific type, Infrared Thermography (IRT), isthe process of analysing an infrared response from an object anddetecting unexpected temperature differentials that may be indicative ofa fault. There are two approaches to Thermography, passive and active.Passive Thermography relies on natural temperature differentials betweenthe object being measured and its surrounds. Passive Thermography iscommonly used in medical sciences and surveillance (Bagavathiappan etal. 2013).

Active Thermography on the other hand uses excitation to generate aheat flux in the object being measured. There are three common types ofexcitation used for Active Thermography, namely optical, mechanical andelectromagnetic (Maldague 2002). Active Thermography is most commonlyused for flaw detection where surface or sub-surface cracks in theirpassive state aren't distinguishable from the surrounds (Yang andHe 2016).

IRT has been chosen as a potential methodology for rail foot flawdetection as it presents a number of potential advantages over othernon-contact measurement techniques such as laser guided wave ultrasonicand optical. It is more immune to crack direction, such as longitudinalweb-foot separation. In this scenario, ultrasonic requires multiplesensors at different orientations for detection of transverse andlongitudinal cracks. Initial crack growth is typically very fine andalmost invisible, however, thermal resistance may still be high enoughto generate temperature variation around the crack. Like optical, IRTcaptures imagery and this allows for location, size and orientation ofthe crack to be detected through post-processing analysis, providingsufficient heat flux is present at the site of the flaw.

It is proposed that, to use IRT as a detection technique, an activeapproach is required; for this paper a Modulated Thermography (MT)approach will be explored. In MT, a periodic heat source is applied tothe surface of an object and the temperature rising process is capturedby an infrared sensor (Badghaish and Fleming 2008). Where the heat flowencounters variations in thermal diffusivity caused by flaws, it isinterrupted and either an increase or decrease in temperature isexperienced in this region. The temperature differential is imagecaptured by the infrared sensor and identified as a flaw.

The next section will detail different locations of heat generationin the rail and set the stage for investigating how heat transfer willinfluence an IRT detection technique.

1.2. Sources of heat

There are four sources of heat generation in the rail attributed tobending, wheel/rail contact, sleeper/rail contact and the naturalenvironment. A diagram of these interactions is shown in Figure 1.Bending and reverse bending of the rail occurs due to the normal andshear stresses exerted by the wheel/rail contact and sleeper attachment,causing strain in the rail and resulting in heat generation. A sideeffect of these shear stresses are stability problems in the vehiclethat may affect the system dynamics with the result being a change tothe temperature generated in the head of the rail due to fluctuations inwheel/rail contact. The influence of this may be captured throughadvanced multi-body modelling and simulation. Direct quantisation ofthis effect is not demonstrated by this paper.

Wheel/rail contact generates friction at the contact surface andconduction between the wheel and rail. The contact friction results inan increased temperature at the rail surface (Ertz and Knothe 2002) andpermits heat transfer vertically through the rail. The wheel/railconduction is generally known as rail chill (Crowe and Raj 1998) as therail is typically at a lower operating temperature than the wheel.

The sleeper/rail interaction results in longitudinal rubbing due tothe strain in the rail as it bends under load with heat generated by therubbing surfaces. There is also a conductive effect between sleeper,wear plate and rail. However, depending on the material composition ofthe sleeper, this effect is quite small and often ignored in this typeof model (Kesler and Zhang 2007).

The final heating effect is due to environmental factors such assolar radiation, convection and body radiation. Zhang presents an energyequilibrium model incorporating these factors for a rail weatherprediction system that considers the rail as a beam floating in space(Zhang and Lee 2008).

Of the four sources of heat presented, this paper will focus onlyon heat transfer resulting from wheel/rail contact friction. In terms ofa MT approach, the heat excitation is periodically generated at a pointon the surface of the rail by the wheel/rail contact friction of eachpassing wheelset. The key challenge in this approach is the effect thatVSD has on the periodic surface temperature excitation and if sufficientheat is generated to flow to the foot of the rail.

This paper shall present a train simulation and rail foot heattransfer methodology for the Australian Heavy Haul Industry where axleloads up to 40 tonnes are in service with 200 cars or more in a trainconsist. It shall determine the role VSD contributes in affecting thetemperatures experienced in the rail foot under different operatingconditions and discuss whether wheel/rail contact friction excitationmay be used in a MT detection technique. The paper will not considervariations in thermal diffusivity in a foot flaw itself or the heatself-generated due to its internal frictional rubbing. This will becovered in future work.

2. Methodology

The development of a MT foot flaw detection approach usingwheel/rail contact friction excitation first requires an understandingof the heat flux process from the running surface to the foot. Toachieve this, a theoretical approach for the modelling and simulation ofthe heat flux vertically through the rail foot is developed. Initialfield tests conducted internally by the Centre for Railway Engineering,CQUniversity have measured variations of foot temperature in the rangeof 0-2 [degrees]C after a heavy haul train of approximately 100 vehicleshas passed. Due to these low temperature increases, it is concluded thata full train simulation methodology is required for modelling the heattransfer through the rail.

The simulation approach combines full train system dynamics usingadvanced VSD simulation techniques, calculation of wheel/rail contacttemperature excitation and a vertical rail heat transfer model. A blockdiagram of the overall methodology is shown in Figure 2. For a MTapproach to be valid, the wheel/rail contact friction excitation mustgenerate a temperature increase greater then 15milli-Kelvin in the railfoot. This is equivalent to the minimum sensitivity for an uncooledmicro-bolometer-type infrared camera (FLIR Ex-Series Infrared Cameras2016).

2.1. Rail heat simulation

The rail vertical heat transfer problem is presented as aone-dimensional boundary value problem with the 'top' boundarycondition representing the running surface, or excitation surface of therail, and the 'bottom' boundary condition as the foot of therail. The temperature of the excitation surface is calculated using thesurface contact temperature algorithm presented by Spiryagin (Spiryaginet al. 2016) and the basic flow chart for the algorithm is shown inFigure 3. The bottom boundary condition is assumed to be insulated. Itis also assumed that the non-boundary elements are insulated, in otherwords, only the top boundary condition considers convection andradiation. As the demonstration of a simulation methodology this issufficient, however, future enhancements of the model should considernon-boundary convection and radiation. The heat transfer is currentlysolved using the Finite Difference Method (FDM) in Equation (1).

[u.sub.j.sup.m+1] = [u.sub.j.sup.m] + s([u.sub.j+1.sup.m] -2[u.sub.j.sup.m] + [u.sub.j-1.sup.m] and s =k[[DELTA]t/[([DELTA]x).sup.2] (1)

The rail is divided into arbitrary elements, with each representinga one dimensional heat transfer in the rail. The inertial frame, orabsolute position, of every wheel is maintained by the simulator and theexcitation surface temperature boundary condition for an element updatedwhen the wheel is determined to be within the region of that particularelement. It is a limitation of the surface temperature model that itassumes the initial temperature of the rail surface is at equilibrium,i.e. it has not been heated by any previous rolling contact. Therefore,before updating the element boundary condition, if the existingtemperature of the boundary condition is higher than the calculatedtemperature, the new lower temperature is ignored. Further modeldevelopment will focus on correcting this behaviour and considerationsfor wheel/rail conduction.

The surface temperature algorithm requires as input thelongitudinal creepage and velocity for each wheel in the train.Therefore, at the end of a train simulation time step, the currentvehicle velocity and longitudinal creepages are passed to the rail heatmodel. Each inertial frame wheel position can be determined by theinertial frame mid-body vehicle position and knowledge of the relativevehicle dimensions.

Once the excitation boundary condition for all wheel contacted railelements has been updated, convection and radiation can be applied tothe entire set of element excitation boundaries as a cooling process.The Finite Difference Method is then applied to update the heat transferthrough all elements in the rail model.

2.2. Train simulation

The train simulation captures the key elements of a heavy haultrain, including length and number of wheelsets. The rationale for anapproach simulating the entire train is to ensure that all kinematicfeatures that occur in long trains are captured and evaluated throughthe rail heat model. This is important for a MT approach asunderstanding the minimum number of periodic heating cycles required isessential for determining practical limitations of the detectiontechnique.

The outputs required from the train simulation for every vehicleinclude its position, longitudinal creepage of each wheel and velocity.These parameters are used as input to the rail heat model forcalculating the running surface excitation. To capture the kinematics ofa full train, a hybrid simulation approach is used comprising the Centrefor Railway Engineering's Longitudinal Train Simulator (CRELTS)(Cole 2006) and the Gensys Multibody simulator (Gensys.1607 n.d.).

CRELTS provides modelling and simulation of train dynamics for thelongitudinal degree-of-freedom as shown in Figure 4, with a major focuson non-linear draft gear modelling. Each vehicle in the train ismodelled as a rigid body connected by non-linear draft gear connectionswith bespoke force application depending on vehicle characteristics. Theforce applicators include propulsion resistance, curve resistance,tractive and dynamic braking effort, train braking and gravity. Thedraft gear connection models accurately capture the function ofdifferent draft gear configurations such as friction clutch and polymergears. Various control strategies are provided for driving the trainsuch as Fuzzy, feed-forward PID and locomotive logger input. CRELTS ishighly optimised and can execute at speeds up to 200x real-timedepending on consist configuration. Cole (2006) describes furtherdetails on the CRELTS system.

GENSYS is designed as a general multipurpose software package formodelling mechanical, electrical and/or multibody systems. Modelling ofrail vehicles using computers was begun by ASEA (Allmanna SvenskaElektriska Aktiebolaget or General Swedish Electrical Company) in Swedenin 1971 in the lead up to the development by that company of the X2000high-speed tilt train. After initially producing a linear programme inthe frequency domain to model a bogie frame with two wheel-sets calledLSTAB, a non-linear time-domain simulation programme called SIMFOquickly evolved to model a whole railway vehicle. In 1992, athree-dimensional general multibody dynamics analysis programme calledGENSYS was developed. At that time, the responsibility for the softwarepackage moved to a new company, AB DEsolver, which now has the sole taskof developing and supporting the package (Spiryagin et al. 2014).

2.2.1. Co-simulation

An approach described by Spiryagin et al. (2012) for thehybridisation of Gensys with an external simulator is throughco-simulation. Co-simulation is the process where two simulator entitiesshare information between each other in either a half or full duplexmanner. In a half-duplex model, one simulator acts as a provider. Anexample of this might be a simulator running in real-time that providesa timing tick to another non real time simulator. In a full duplexsystem, there is a bi-directional flow of information where bothsimulators are reliant on each other for data inter-dependence. In themethodology described by this paper and for the remainder of thissection, co-simulation is merged with a parallel computing architecture,redistributing the computational complexity and reducing the run-time ofa large multi-body simulation problem.

To use Gensys in a co-simulation environment, it is started as aTCP server listening on a port. An external process or simulator canthen connect in a peer-to-peer arrangement and share variableinformation with the Gensys simulator. An example of this is a Gensysvehicle model co-simulating with an external simulator that calculatesinfluences on the vehicle model such as connections, tractive efforts orresistive forces and transfers these to Gensys for application on thevehicle body per time step.

The simulation approach defined by this paper extends the singlevehicle co-simulation approach to model an entire train through a singleinstantiation of CRELTS and multiple Gensys servers representing eachvehicle in the train. The challenge presented by a full train simulationapproach using MBS is computation time. If each Gensys vehicle instanceexecutes in sequence, the computation time for the train simulator isrepresented by Equation (2).

[t.sub.comp] = (n[C.sub.CRELTS] + n[C.sub.Genys]) x [t.sub.sim] (2)

Where n is the number of vehicles in the train and [C.sub.CRELTS]and [C.sub.Genys] are the times taken to run one numerical integrationstep of CRELTS and Gensys, respectively, for one vehicle. Time profilingof the two simulators, although heavily model dependent, shows Gensys isat least three orders of magnitude longer in execution time compared toCRELTS. In the profile comparison conducted for the unit case of 1vehicle simulating for 1s, CRELTS takes 0.0076s whilst Gensys takesapproximately 10s. Using the time Equation in (2) for a 250 vehicleheavy haul train executing a 500s simulation would take around 2 weeksto complete. In this case, it is clearly evident that the Gensyssimulation component heavily dominates the total computation time. Inthis form a full train simulation is unachievable due to the excessivecomputation time and an algorithm reduction in complexity is requiredfor the Gensys component of simulation. A solution to this problem is toexecute the Gensys vehicle instances in parallel with each other. Thecomputational time function then reduces to Equation (3), where theGensys component now executes in constant time.

[t.sub.comp] = (n[C.sub.CRELTS] + [C.sub.Genys] x [t.sub.Sim] (3)

In comparison, the computation time reduces from 2 weeks to 1.65 hwith a reduction in Gensys simulation complexity from linear to constanttime. The remainder of this section shall detail the parallel simulationapproach.

This type of parallel simulation activity is becoming verypractical with access to high-performance computing facilities andmulti-core CPU workstations now prevalent throughout university andindustry. The proposed Gensys vehicle model, simulation step rate andexecution time result in non-volatile memory consumption into the manygigabytes for a single-vehicle model. When multiplied out over a fullheavy haul train, this obviously accounts for hundreds of gigabytes ofRAM usage for a full train simulation. These types of RAM configurationare still not commonplace in typical engineering work stations. Theblock diagram in Figure 5 shows the flow of operation andsynchronisation for the simulation approach used within this paper.

2.2.2. Train co-simulation methodology

This section describes the general train co-simulation processshown in Figure 5. Before the simulation can commence, Gensys serversare started for each vehicle in the train. As can be appreciated, eachvehicle may require a different model type depending on theconfiguration. In a typical unit train environment seen in heavy haul,there will be at least two models, one for the locomotives and one forthe wagons. To ensure the correct data is transferred to the rightserver, the port listening address for each Gensys server must be knownand configured in advance. This is achieved by configuring each serverto a known address by using the convention of base port plus vehicleposition offset. As an example, in a 250 vehicle train consist therewould be 250 individual servers started with each listening on a uniqueport address.

The main thread, or system process, manages the CRELTS instance andcontrols the flow of execution between CRELTS and the Gensys vehicleservers. When it starts, it loads and initialises CRELTS andinstantiates a thread per vehicle in the train up to the number ofvehicles defined by the CRELTS consist. The purpose of each vehiclethread is to manage its co-simulation connection interface. Each threadopens a connection with its respective Gensys server according to theport convention defined for the Gensys servers and establishes theparameters to be transferred. Details on parameters transferred for thespecific methodology defined by this paper are discussed in theSimulation Parameters section. Once the per vehicle threadinitialisation is complete, each thread signals the main thread that ithas completed initialisation and then enters a wait state for furtherinstruction.

Once the main thread has received notification from all vehiclethreads that initialisation is complete it can then start the systemsimulation. The first step is to execute a time step of CRELTS tosimulate the current longitudinal state of the train. Upon completion ofthe CRELTS step the main thread broadcasts to all waiting vehiclethreads to wake-up and start their respective co-simulation step. Themain thread then enters its own wait state where it listens forcompletion notifications from its vehicle worker threads.

When each co-simulation vehicle thread wakes up due to the mainthread signal, it must collate the required simulator state informationfor its respective vehicle position from the CRELTS interface andtransfer it by TCP to the Gensys server. Each thread then commands itsGensys server to execute its simulation time-step. To maintainsynchronisation between the two simulators the Gensys server does notrespond to the command until it has completed simulation of its timestep. Once the response from the Gensys server is received, the vehiclethread can initiate the results transfer process from Gensys. How theseresults are then used by the co-simulation or main thread is applicationdependent. When the co-simulation vehicle thread completes the resultstransfer, it signals the main thread with a completion event and thenre-enters a wait state for the next simulation instruction.

When the main thread receives completion events from allco-simulation vehicle threads it can then execute any other user definedfunctions. In the case of this paper, it executes the rail heat transfermodel using result data transferred from Gensys. Further detail inprovided in the Rail Heat Transfer Simulation section.

2.2.3. Synchronisation

Although the two systems, CRELTS and Gensys, share as much data asthey can to ensure equivalence, there are still minor numericaldifferences that result in cumulative displacement and velocity error.There are three main sources of error that exist between the twosystems, numerical integration, track curvature modelling and wheel/railcontact mechanics. To mitigate these errors a synchronisation controlleris used to apply a small compensating force to the Gensys body centre ofmass as originally proposed by Spiryagin et al. (2017) in Equation (4).

CV = P([[??].sub.MBS] - [[??].sub.LTS])

where P is the proportional gain, [[??].sub.MBS] the Gensys vehiclevelocity and [??].sub.LTS] the CRELTS vehicle velocity.

This paper proposes a new approach that implicitly ties inSpiryagin's controller with an additional component to controldisplacement error. To ensure both systems are evaluated at the sametrack locale throughout the life of the simulation, the objective of theproposed controller is to minimise the displacement error between thetwo simulators. The standard form of a PID controller is shown inEquation (5) with the error function to control expressed in Equation(6).

CV = [K.sub.P]e(t) + [K.sub.I] [integral] e(t) +[K.sub.D][de(t)/dt] (5)

e(t) = [x.sub.MBS] - [x.sub.LTS] (6)

The proposed controller shall only use the proportional (P) anddifferential (D) components of (4) as the hysteresis provided by theIntegration (I) component is unlikely to be useful for this system.Further consideration for integration may be considered in the future.The resultant controller is shown in Equation (7) with Equation (5)substituted into Equation (4) for full clarity.

CV = [K.sub.P]([x.sub.MBS] - [x.sub.LTS]) + [K.sub.D][d([x.sub.MBS]- [x.sub.LTS])/dt] (7)

The proportional velocity control of (4) when compared to the newcontroller (7) can be seen to be equivalent to the differential onlycomponent. Differential only control provides good control on transientchanges, however, it is unable to manage any steady state error in thesystem. The effect on this type of system is that, over time, thedisplacement difference shall drift away in the direction of the steadystate error. An example of this is shown in the first series of Figure 6where, under a constant traction case, an increasing displacement erroroccurs. In contrast, the second series of Figure 6 shows that the new PDcontroller has control over the displacement error, allowing forindefinitely long simulations to be conducted. In this example theproposed controller uses the differential gain of 10kN/m/s and aproportional gain of 2kN/m.

3. Simulation parameters

This section describes the simulation configuration used for theCRELTS and Gensys simulators, the co-simulation data transfer and railheat transfer simulation.

3.1. Vehicle parameters

For simplicity, the wagon model used in the simulation shares thesame physical model parameters as the locomotives, with the exception oftraction. This allows for a single model type in Gensys with tractiveand dynamic braking effort always idle for wagon instantiations.

3.2. CRELTS configuration

Parameters describing the train consist and used by the CRELTSinstantiation are shown in Table 2.

3.3. Gensys configuration

The vehicle model created in and used by Gensys is shown in Figure7. Refer to the parameters in Tables 1 and 2 for more information aboutthe vehicle.

3.4. Co-simulation parameters

The CRELTS-Gensys co-simulation system executes on a 1 ms timestep. The data shared between the two simulators are shown in Table 3below.

3.5. Rail heat transfer simulation parameters

Rail heat transfer parameters and external input stimuli such asWheel/Rail contact friction and solar factors are shown in Table 4.

3.6. Test cases

The test cases to be demonstrated by this paper are shown in Table5. There are four different test case with the same initial runningspeed of 16.67 m/s (60 km/hr) and total simulation time of 200s. Thereare two track types, the first being straight track and the second beinga track with a single curve commencing 300 m after the lead locomotivestart position with a radius of 800 m, transition lengths of 50 m andarc length of 600 m. Both track instances are level (without anygradients).

4. Results

The results provided demonstrate the functionality of allcomponents of the simulation methodology. Figure 8 shows the couplerforces calculated by CRELTS and are a subset of the data transferred ona per simulation tick to each Gensys vehicle model. Figure 9 shows thecreep forces calculated from Gensys and these are transferred back tothe CRELTS process as input to the heat model. Figure 10 indicates thesynchronisation force calculated using the new PD controller presentedto maintain displacement and velocity control for the co-simulationenvironment. Figures 11 and 12 are heat maps for the four test cases attwo instances during the simulation, 60 and 160s, respectively.

5. Discussion

5.1. Rail temperature modelling

There is a variation in surface excitation temperature experiencedfor the four case studies. It is shown in tangent track that negligibletemperature excitation is generated, with a surface increase in lessthan 0.5 K for both the constant speed and accelerating trains. Thisresults in no heat flux vertically through the track anywhere under thetrain. For the two curve case studies there is a surface excitationgenerated primarily by the locomotives due to their higher creep forcesand this is shown in Figure 11(c and d). The wagons have a smallincrease in creep forces on the curve transitions as shown in Figure 9,and these are clearly identifiable as surface heat generated in Figure12(c and d). However, the excitation is sporadic and only generateslocalised heat flow through the rail. The curve track - increasing speedcase shows a heat rise greater than 1 K at a depth of 60 mm, but againthis is localised to the curve. No further increases in temperature weremeasured at depths greater than this.

5.2. Modulated Thermography

From the MT detection technique perspective, the excitationsgenerated have not been sufficient to generate heat flow through to therail foot area, resulting in no possibility for thermal radiation arounda crack and resultant temperature differentials. However, the rail heatmodel development is in its infancy, simplified and presented as ademonstration for further refinement as part of the greater modellingand simulation methodology.

Key areas for further development include the surface excitationmodelling as its limitation of using the initial rail temperature doesnot capture small gradual increases that may occur in the head of therail. As can be appreciated, MT relies on the surface excitationtemperature so this facet is crucial. The beam models vertical elementsare currently one dimensional and the next step of the model shouldinclude at least two-dimensional heat flow and possibly include coolingeffects on their vertical surfaces. Future development for theself-heating of the flaw area from mechanical excitation due to normaland shear forces in the rail may present an additional form ofexcitation not yet considered.

Useful information that can be gleaned at this early stage ofdevelopment is the effect that VSD has on surface excitationtemperatures. Running surface temperature modelling dependent onlongitudinal creepage and linear vehicle velocity will have largechanges in temperature generated at the surface of the rail. While itmay be possible under more extreme traction and braking scenarios thatheat flow will transfer to the foot of the rail, it is likely thatanother source of excitation is required for a general purpose IRT footflaw detection technique.

In the simulation scenarios presented, the curve radii isrepresentative of modern heavy haul rail corridors where curves are 800m or greater. In the case of older networks, or those with significanttopographical constraints, corners with radii down to only 300 m arefound in practise. In this environment, at low speeds and high tractiveeffort, the temperatures generated in the wheel/rail contact patch havebeen known to exceed 100 K above ambient track temperature. This posesthe most likely situation where the excitation is sufficient for MT tobe a suitable detection technique.

5.3. Synchronisation

Figure 10 shows the forces required for locomotive and wagonsynchronisation, respectively. The locomotive synchronisation controlexhibits greater hysteresis at the start of the run and this is surmisedto be a result of the differences in tractive effort application byCRELTS and Gensys. In the case of straight track and constant speed, thecontrol results in negligible synchronisation requirements. However, inthe increasing cases there continues to be a small amount ofsynchronisation force required and this is assumed to be caused by slipin the contact patch. The situation is similar in the wagon control;however, as there is no tractive effort, the control is able to maintainsynchronisation without significant force application.

The curving cases demonstrate the source of error described earlierin the synchronisation methodology section. It can be seen in both thelocomotive and wagon control cases that the synchronisation forcerequired changes at approximately the 15 and 30s time sequences,respectively, for the vehicles. This aligns with when each vehicleenters the transition to the curve.

6. Conclusion

This paper presented a full train simulation methodology forcalculating the heat transfer to the foot of the rail with considerationfor variations in vehicle system dynamics and wheel/rail contactmechanics. A parallelised co-simulation approach was used thatintegrated the CRE Longitudinal Train Simulator and Gensys. It was shownfor a MT detection technique that there are potential limitations forconsideration in using surface only excitation for the detection of railfoot flaws. There are significant further developments of the rail heatmodel planned and required.

Acknowledgements

The authors would like to acknowledge the Australasian Centre forRail Innovation (ACRI) for their support of this project.

Disclosure statement

No potential conflict of interest was reported by the authors.

Funding

This work was supported by the Australasian Centre for RailInnovation [grant number HH#1].

Notes on contributors

Chris Bosomworth is a PhD student with the Centre for RailwayEngineering at Central Queensland University. His thesis topic is MovingVehicle Rail Foot Flaw Detection using Infrared Thermography. Hegraduated in 2001 with a Bachelor of Mathematical Science and Computingwith Distinctions from Central Queensland University and has worked inboth industry and academia in research and commercialisationspecialising in software engineering for railway applications.

Maksym Spiryagin is the deputy director of the Centre for RailwayEngineering at Central Queensland University. His research interests arelocomotive traction, rail vehicle dynamics, contact mechanics,mechatronics, acoustics and real-time and software-enabled controlsystems. He received his PhD in the field of Railway Transport in 2004at the East Ukrainian National University. He has published four booksand has more than 100 other scientific publications and twenty patentsas one of the inventors.

Sanath Alahakoon received his BSc Eng (Honours) degree inElectrical and Electronics Engineering from the University ofPeradeniya, Sri Lanka in 1994. He received his PhD in Digital MotionControl from the Royal Institute of Technology (KTH), Sweden in 2000.From 2000 till the middle of 2009, he worked as a senior lecturer in theDepartment of Electrical and Electronic Engineering in the University ofPeradeniya, Sri Lanka. Currently he is a lecturer in ElectricalEngineering in the School of Engineering and Technology in CentralQueensland University, Gladstone campus. His research interests aredigital control, estimation and identification, non-linear control,electrical machines and drives, instrumentation, automation and hybridelectric systems.

Colin Cole is the director of the Centre for Railway Engineering atCQU. He has worked in the Australian rail industry since 1984, startingwith six years in mechanised track maintenance for Queensland Railways.Since then he has focused on a research and consulting career involvingwork on track maintenance, train and wagon dynamics, train controltechnologies and the development of on-board devices. He has beenextensively engaged with industry via the past nationally funded RailCRC programs, and has continuing involvement via the Australian Centrefor Rail Innovation and the new Rail Manufacturing CRC. His PhD was inLongitudinal Train Dynamics Modelling. He has authored and/orco-authored over 90 technical papers, one book chapter, two books,numerous commercial research and consulting reports and has developedtwo patents relating to in-cabin locomotive technologies.

References

Badghaish, Adel A., and David C. Fleming. 2008."Non-Destructive Inspection of Composites Using Step HeatingThermography." Journal of Composite Materials 42 (13): 1337-1357.doi:10.1177/0021998308092202.

Bosomworth C., Spiryagin M., Alahakoon S., Cole C. (2018)."The effect of vehicle dynamics on rail foot flaw detection."Dynamics of Vehicles on Roads and Tracks. Vol. 2. London: CRC Press.

Bagavathiappan, S., B. B. Lahiri, T. Saravanan, John Philip, and T.Jayakumar. 2013. "Infrared Thermography for Condition Monitoring--AReview." Infrared Physics & Technology 60 (Sep.): 35-55.doi:10.1016/j.infrared.2013.03.006.

Cannon, D. F., K.-O. Edel, S. L. Grassie, and K. Sawley. 2003."Rail Defects: An Overview." Fatigue & Fracture ofEngineering Materials & Structures 26 (10): 865-886.doi:10.1046/j.1460-2695.2003.00693.x.

Cole, Colin. 2006. "Longitudinal Train Dynamics,"Handbook of Railway Vehicle Dynamics. Vol. Ch 9. Boca Raton, FL: CRCPress.

Crowe, K. E., and P. K. Raj. 1998. Analyses of Rail Chill Effect.Final Report DOT-VNTSC-FRA-96-8. Washington, DC: Federal RailroadAdministration.

Ertz, Martin, and Klaus Knothe. 2002. "A Comparison ofAnalytical and Numerical Methods for the Calculation of Temperatures inWheel/Rail Contact." Wear 253 (3-4): 498-508.doi:10.1016/S0043-1648(02)00120-5.

FLIR Ex-Series Infrared Cameras. 2016. Accessed October 13.http://www.flir.com/instruments/ex-series/

Gensys.1607. n.d. Reference Manual. User's Manual for ProgramCALC. http://www.gensys.se.

IntelliTrack Technology. 2016. IntelliTrack Technology. AccessedNovember 7. http://www.lbfoster-salientsystems.com/Database_Management.asp

Kesler, Kevin, and Yu-Jiang Zhang. 2007. System and Method forPredicting Future Rail Temperature. US20070265780 A1, filed April 20,2007, and issued November 15, 2007.http://www.google.com/patents/US20070265780.

Maldague, Xavier P. V. 2002. "Introduction to NDT by ActiveInfrared Thermography." Materials Evaluation 60 (9): 1060-1073.

McClanachan, Mitchell, Ben Scown, Dudley Roach, Don Skerman, andBarry Payne. 2001. "Autonomous Detection of Severe Wagon-TrackInteraction Dynamics." In 7th International Heavy Haul Conference,Confronting the Barriers of Heavy Haul Rail Technology, Brisbane,Australia. Virginia Beach, VA.

Spiryagin, Maksym, Colin Cole, Yan Quan Sun, Mitchell McClanachan,Valentyn Spiryagin, and Tim McSweeney. 2014. Design and Simulation ofRail Vehicles. Ground Vehicle Engineering. Boca Raton, FL: CRC Press.

Spiryagin, Maksym, Qing Wu, Yan Quan Sun, Colin Cole, and IngemarPersson. 2017. Locomotive Studies Utilizing Multibody and TrainDynamics, April V001T10A002. doi:10.1115/JRC2017-2221.

Spiryagin, Maksym, Scott Simson, Colin Cole, and Ingemar Persson.2012. "Co-Simulation of a Mechatronic System Using Gensys andSimulink." Vehicle System Dynamics 50 (3): 495-507.doi:10.1080/00423114.2011.598940.

Spiryagin, Maksym, Wu Qing, Kai Duan, Colin Cole, Yan Quan Sun, andIngemar Persson. 2016. "Implementation of a Wheel--rail TemperatureModel for Locomotive Traction Studies." International Journal ofRail Transportation 5 (1): 1-15. doi: 10.1080/23248378.2016.1190306.

Yang, Ruizhen, and Yunze He. 2016. "Optically andNon-Optically Excited Thermography for Composites: A Review."Infrared Physics & Technology 75 (Mar.): 26-50.doi:10.1016/j.infrared.2015.12.026.

Zhang, Yu-Jiang, and Sung Lee. 2008. "Modeling RailTemperature with Real-Time Weather Data." In Fourth NationalConference on Surface Transportation Weather, 49-60. Indianapolis, IN.

Chris Bosomworth (a,b), Maksym Spiryagin (a,b), Sanath Alahakoon(a,b) and Colin Cole (a,b)

(a) Centre for Railway Engineering, Central Queensland University,Rockhampton, Australia; (b) Australasian Centre for Rail Innovation,Canberra, Australia

CONTACT Chris Bosomworth [??] [emailprotected]

https://doi.org/10.1080/14484846.2018.1457260

Table 1. Locomotive model parameters in both Gensys and CRELTS.Name Value UnitLocomotive model N/AMass 134 tLength 20 mNumber if axles 6 N/APowerTable 2. Key parameters for configuring CRELTS.Name Value UnitNumber of vehicles 130Locomotive positions 1 and 2Wagon positions 3 to 130Total wagon mass 17152000 kgTotal train mass 17420000 kgTrain length 2600 mSimulation step rate 1000 Steps/secondDraft gear connection SL76 N/ATrain braking system Standalone ECP N/ATable 3. Bi-directional data exchange between CRELTS and Gensys.Name Unit DescriptionCRELTS to GensysFront draft gear force (FDGF) N The resultant wagon bodyRear draft gear force (RDGF) N application force is -FDGF + RDGFRolling resistance force N Based on Davis equationTractive/DB effort N Locomotive models onlySynchronisation force N Force applied to Gensys vehicle body to maintain synchronisation with CRELTSGensys to CRELTS (and heat model)Vehicle position M Used by CRELTS inVehicle velocity m/s synchronisation controllerWheelset 1 left longitudinal mcreepageWheelset 2 left longitudinal mcreepageWheelset 3 left longitudinal mcreepageWheelset 4 left longitudinal mcreepageWheelset 5 left longitudinal mcreepageWheelset 6 left longitudinal mcreepageTable 4. Rail heat transfer and boundary condition cooling parameters.Name Value UnitGeneral parametersThermal conductivity 40 W/milli-KelvinW/R contact friction 0.47Specific heat 450 J/kgKRail density 7850 kg/[m.sup.3]Top rail surface area 0.0075 [m.sup.2]RadiationEmissivity 0.75ConvectionCoefficient 6SolarAtmospheric filtering factor 0.5Solar absorptivity factor 0.75Solar constant 1366 W/[m.sup.2]Table 5. Four simulation test cases.Name Power DescriptionStraight--constant speed Notch 5 Maintains almost constant speed for the given notchStraight--increasing speed Notch 8 Increases in speed by almost 2 m/s over the simulation time frameCurve--constant speed Notch 5 See Straight--constant speedCurve--increasing speed Notch 7 See Straight--increasing speed