Failure cause analysis of surface burn on high-speed railway U75VG rails

China's high-speed railway operation mileage has exceeded 45,000 kilometers, ranking first in the world. With the increase in train axle load, operation density and service life, the damage problem of rails under the coupling effect of cyclic wheel-rail load and complex environment has become increasingly prominent. Among them, tread burn, as the most common early damage form, accounts for more than 30% of the total rail damage (Zhang & Liu, 2023; Chang, Cai, Li, & Chen et al., 2017; Wei, Liu, Li, & Zuo, 2015). Burn not only damages the track regularity, intensifies the vehicle-track system vibration, and reduces passenger comfort, but also can rapidly evolve into cracks, spalling and even rail breakage, which greatly increases the maintenance cost and seriously threatens the traffic safety (Zhai & Zhao, 2016).

As early as the 1990s, the International Union of Railways (UIC) listed rail tread burn as a key research type of rolling contact fatigue (RCF) damage. European scholars systematically revealed the thermal phase transformation mechanism induced by wheel-rail sliding friction through on-site monitoring and laboratory simulation. Lian et al. (2019) established a thermal-mechanical coupling model of wheel-rail sliding contact by the finite element method, and calculated that when the relative sliding velocity between wheel and rail exceeds 0.5 m/s, the instantaneous temperature in the contact area can rise to above 800°C, which is sufficient to cause austenitization transformation of pearlitic rail steel, and form martensitic white layer after cooling. Through long-term on-site tracking, Deng, Qian, Li, and Dollevoet (2018) found that rail burn has obvious line characteristics: the incidence rate in long downhill sections and small-radius curve sections is three to five times that of straight sections, and it is closely related to the train braking mode. The Railway Technical Research Institute of Japan focused on the mechanical behavior of the martensitic white layer and measured through nanoindentation experiments that its hardness can reach 2–3 times that of the matrix, and its fracture toughness is only 1/4 of that of the matrix, which is the core inducement for crack initiation (Zhou et al., 2019).

Domestic scholars have carried out a lot of research on the characteristics of rail burn of high-speed railways in China. Liu, Li, and Tian (2018) found through investigations on multiple passenger dedicated lines such as Xi'an-Chengdu and Chongqing-Guiyang lines that some rail burn was caused by idling of engineering vehicles before the line was put into operation, and rapidly expanded under cyclic load after operation. Li et al. (2020) compared and analyzed the metallographic structure of rails with different burn degrees, and pointed out that the thickness of martensitic white layer is the key factor determining the evolution rate of damage: when the white layer thickness is less than 0.5 mm, it is mainly manifested as surface spalling; when the thickness exceeds 1 mm, it is easy to induce deep transverse cracks. In recent years, with the advancement of detection technology, automatic detection methods for rail burn based on machine vision have developed rapidly. The circular spot burn detection algorithm proposed by Zhang et al. (2026) can achieve an accuracy of more than 95%, which provides an effective means for rapid on-site investigation. Li, Wang, and Zhang (2021) further revealed the micro-damage evolution law of U75VG rail surface under thermal-mechanical coupling and supplemented the quantitative evaluation index of white layer damage. Hu et al. (2022) analyzed the matching relationship between the rail grinding process and white layer removal, which provided technical support for field maintenance. Recent advances in deep learning-based automatic detection (Yang et al., 2022; Zhang et al., 2026) and digital twin-driven predictive maintenance (Kushwaha, Kumar, & Harsha, 2025) have further enhanced the capability for early-stage burn detection and life-cycle management. Updated statistical data indicates that the incidence of tread burn on Chinese high-speed railway networks has increased by 12% annually since 2022, underscoring the urgency of this research.

Although significant progress has been made in elucidating the mechanisms and detection of rail tread burn, two specific gaps remain: first, the quantitative characterization of crack propagation paths – particularly the crack initiation angle, the depth at which branching occurs, and the mechanical conditions governing deflection at the martensite–matrix interface – has not been systematically reported for U75VG rails under the specific operating conditions of 350 km/h passenger dedicated lines; second, existing maintenance guidelines lack quantitative, white-layer-thickness-dependent thresholds for differentiated preventive and corrective grinding. This study addresses both gaps by combining multi-scale failure characterization with thermo-mechanical and fracture-mechanics analysis, yielding a complete description of the failure chain and a set of operationally actionable maintenance thresholds.

The rail investigated in this study is a 60 kg/m U75VG grade rail manufactured to the Chinese standard TB/T 3276–2011. The rail was retrieved from the straight section at chainage K123 + 450 of a passenger-dedicated line during a routine inspection, after a tread burn defect was detected by ultrasonic flaw detection. The line was designed for a maximum operating speed of 350 km/h and had been in service for approximately 3 years at the time of rail removal, with an estimated cumulative gross tonnage of approximately 120 million tonnes. The removed rail section containing the burn defect, approximately 600 mm in length, was transported to the laboratory for detailed analysis.

Magnetic particle inspection was performed on the rail tread, and the results are shown in Figure 1. Multiple irregular surface cracks were observed within the wheel-rail contact band on the rail head tread, with the longest crack extending approximately 30 mm. A representative crack was sectioned open transverse to the rail axis using a precision saw to expose the fracture surface. The macroscopic fracture morphology is shown in Figure 2. The fracture surface reveals three distinct morphological zones: (1) a near-surface horizontal propagation zone extending approximately 1 mm below the tread, spanning the width of the contact band; (2) a vertical (transverse) propagation zone extending downward along the rail cross-section and (3) a final fracture zone with dimensions of approximately 30 mm × 10 mm. The fracture surface is macroscopically flat and faceted, with no evidence of gross plastic deformation, indicative of a predominantly brittle fracture mode.

The fracture surfaces were examined using a Quanta 400 scanning electron microscope (FEI, USA) operated at an accelerating voltage of 20 kV. The micro-morphological characteristics of each fracture zone are presented in Figures 3–5. The horizontal propagation zone (Figure 3) exhibits a compact oxide scale with a mechanically smeared surface texture, suggesting repeated compressive contact and frictional rubbing between the mating crack faces under cyclic wheel-rail loading. The fatigue propagation zone (Figure 4) displays well-defined radial marks converging toward the tread surface, consistent with a surface-originated fatigue crack. The artificially fractured surface (Figure 5) shows characteristic cleavage river patterns with flat facets, which are typical of transgranular brittle fracture in martensitic microstructures.

Detailed fractographic examination of the fatigue propagation zone was conducted at higher magnifications. At low magnification (80×), regularly spaced fatigue striations were observed, with a measured spacing of approximately 2–5 μm (Figure 4a). These striations are characteristic of incremental crack advance under cyclic mode I loading. Assuming each striation corresponds to one loading cycle, the local crack growth rate da/dN is estimated to be on the order of 2–5 μm/cycle. For comparison, the fatigue crack growth rate of pearlitic rail steel under comparable stress intensity factor ranges is typically 0.01–0.1 μm/cycle (Ringsberg, Loo-Morrey, Josefson, Kapoor, & Beynon, 2000; Garnham & Davis, 2008), indicating that the crack growth resistance of the martensitic white layer is approximately two orders of magnitude lower than that of the parent pearlitic microstructure. At high magnification (500×, Figure 4b), the striations display a river-pattern morphology oriented toward the crack origin. Secondary microcracks, 10–20 μm in length, were also observed on the fatigue fracture surface and are attributed to local stress concentration at the propagating crack tip.

The fractographic observations described above are consistent with the multi-stage crack propagation mechanism discussed in Section 3.3. The oxide-covered horizontal propagation zone corresponds to stage I crack propagation along the martensite-matrix interface; the fatigue striations correspond to stage II propagation into the pearlite matrix; and the cleavage fracture morphology corresponds to stage III unstable fracture.

Metallographic specimens were sectioned perpendicular to the fracture surface, mechanically ground and polished to a 1 μm finish, and etched with 4% nital (4 vol.% HNO₃ in ethanol). The etched cross-sections were examined using an optical microscope (Figure 6). A distinct bright-etching layer is present immediately beneath the tread surface, with a thickness of approximately 1 mm at this section (Figure 6a). The layer exhibits an acicular morphology characteristic of martensite and is clearly distinguishable from the underlying matrix, which consists of a uniform ferrite-pearlite microstructure (Figure 6b).

Micro-Vickers hardness measurements (load 300 gf, dwell time 15 s, performed in accordance with ASTM E384-17) were conducted at 15 indentation points across each zone, with inter-indentation spacing maintained at greater than three times the indentation diagonal to avoid strain-field interference. The white layer exhibits a hardness of 718 ± 32 HV0.3, while the pearlitic matrix gives 337±18HV0.3. The white layer is approximately 2.1 times harder than the matrix. The larger absolute standard deviation of the white layer reflects the inherent microstructural heterogeneity of untempered martensite—including local variations in prior austenite grain size, carbon content, and the occasional presence of retained austenite islands. The combination of acicular morphology, high hardness, and bright-etching appearance under nital confirms that the white layer consists predominantly of untempered martensite.

An additional metallographic specimen was prepared at a different location along the rail length to examine the crack path in relation to the white layer distribution (sampling position shown in Figure 7). At this location, the white layer thickness varies along the rail length, reaching a maximum of approximately 2.5 mm (Figure 8). The crack initiates at the tread surface and propagates into the rail head at an inclination angle of approximately 80° relative to the tread plane (measured from the metallographic image using digital image analysis). Upon reaching the martensite-matrix interface, the crack undergoes branching and deflection, continuing to propagate predominantly along the interface rather than penetrating into the matrix. This observation suggests that the martensite-matrix interface constitutes a preferred path for crack propagation, acting as a mechanical discontinuity that redirects the crack from a mixed-mode (I + II) trajectory to a predominantly mode II (shear-dominated) trajectory.

The chemical composition of the rail base material was determined by optical emission spectrometry in accordance with the Chinese standard TB/T 3276–2011 “Rails for High-speed Railway” The results, presented in Table 1, confirm that the concentrations of all regulated elements fall within the specified ranges, thereby excluding material nonconformity as a contributory factor to the observed damage.

Tensile test specimens with a gauge diameter of φ10 mm were machined from the rail head in the longitudinal (rolling) direction, at a depth sufficient to avoid the heat-affected surface layer. Testing was conducted at a crosshead displacement rate of 1 mm/min. The results are summarized in Table 2. The measured tensile strength (Rm = 1073MPa) and elongation after fracture (A = 12%) both satisfy the standard requirements. Additionally, Brinell hardness measurements were performed on the tread surface at locations outside the visible crack zone, after removing approximately 1mm from the as-received surface by grinding to eliminate potential decarburization and surface roughness effects. The measurements were taken at 20mm intervals along the rail length. The results (Table 3) show an average hardness of 316HBW, which lies within the standard-specified range of 280–320HBW.

According to the on-site damage characteristics and test results, the rail tread burn is a typical RCF damage. According to TB/T1791-2010 “Classification Standard for Rail Damage,” it is classified as “external damage of rail head,” code 1,160; corresponding to “isolated wheel track burn” in the UIC712-2002 standard of International Union of Railways, the unique identification code is 2,251.

The essence of rail tread burn is the joint effect of friction heat generated by relative sliding between wheel and rail and contact stress. When the train brakes, relative sliding occurs between the wheel and the rail, and the friction work is converted into heat energy, which causes the temperature of the contact area to rise sharply.

The friction heat flux per unit area can be expressed as:

where.

• q is the heat flux density (W/m²)

• μ is the friction coefficient between wheel and rail

• p is the wheel-rail contact stress (Pa)

• v is the relative sliding velocity between wheel and rail (m/s)

According to Hertz contact theory, the maximum wheel-rail contact stress is:

Where.

• F is the normal wheel-rail force (N)

• a is the semi-major axis of the contact patch (m)

For the contact between 60 kg/m rail and standard wheel, when the axle load is 17 t, the calculated maximum contact stress is about 1200 MPa. When the train performs emergency braking, the relative sliding velocity between wheel and rail can reach 1–2 m/s, and the friction coefficient is about 0.3–0.5. Substituting into Equation (1), it can be obtained that the heat flux density in the contact area can reach 3.6 × 10⁸–1.2 × 10⁹ W/m².

Based on the heat flux density, the temperature rise of the rail surface can be calculated using the one-dimensional transient heat conduction equation. For a semi-infinite body subjected to constant heat flux, the surface temperature rise is given by:

where ΔT(t) is the temperature rise at time t (K)α is the thermal diffusivity of rail steel (m²/s),

This value is consistent with thermophysical property measurements for pearlitic rail steels reported in the literature, where values typically range from 1.2 to 1.8 × 10⁻⁵ m²/s.K is the thermal conductivity of rail steel (W/(m·K)),

This value agrees with experimental measurements for U75V rail steel, which measured values in the range of 47–53 W/(m·K) using the laser flash method. t is the heating time (s) Substituting the typical values:

we can calculate the temperature rise at different times:

At $t=0.01$ s: $ΔT=2×7.2×108×1.5×10‐5×0.01π×50≈345°C$

At $t=0.05$ s: $ΔT≈772°C$

At $t=0.1$ s: $ΔT≈1092°C$

The calculation results show that when the heating time reaches 0.05s, the surface temperature has exceeded the eutectoid temperature of pearlitic steel (727°C), and austenitization transformation begins. When the heating time reaches 0.1s, the surface temperature exceeds 1,000°C, which is sufficient to complete the austenitization transformation of the surface layer.

Temperature gradient and heat-affected zone depth calculation:

The temperature distribution along the depth direction at time t is:

Where $x$ is the depth from the surface, and erfc is the complementary error function.

Taking $t=0.1$ s, we can calculate the temperature at different depths:

At $x=0$ (surface): $T≈1092°C$

At $x=0.5$ mm: $T≈856°C$ (still above austenitization temperature).

At $x=1.0$ mm: $T≈642°C$ (below austenitization temperature).

At $x=2.0$ mm: $T≈315°C$

This indicates that the depth of the heat-affected zone (temperature exceeding 727°C) is about 0.7 mm, which is consistent with the observed martensitic white layer thickness of 1–2.5 mm (considering the nonuniform distribution of heat flux).

Cooling rate calculation:

After the heat source leaves, the surface cools rapidly. The cooling rate at the surface can be estimated as:

Where $τ$ is the characteristic cooling time, approximately $τ≈l2α$⁠, and $l$ is the heat-affected zone depth.

Taking $l=1$ mm = 0.001m, we get:

Therefore, the average cooling rate is:

This cooling rate far exceeds the critical cooling rate of pearlitic rail steel (about 100°C/s), so martensitic phase transformation occurs instead of pearlite transformation. The extremely high cooling rate is the fundamental reason for the formation of hard and brittle martensitic white layer (Wu et al., 2016).

Phase transformation threshold analysis:

Combining the above calculations, we can determine the critical conditions for the formation of martensitic white layer:

Heat flux threshold: $q>2.5×108$ W/m² (corresponding to surface temperature reaching 727°C within 0.1s).

Sliding velocity threshold: $v>0.4$ m/s (when $μ=0.4$⁠; $p=1200$ MPa)

Heating time threshold: $t>0.05$ s (sufficient to complete austenitization).

In actual operation, when the train performs emergency braking, especially on long downhill sections, the above three conditions are easily met simultaneously, resulting in the formation of martensitic white layer. Due to the good thermal conductivity of the rail matrix, the high-temperature area on the surface layer will be cooled rapidly, with a cooling rate of 10³–10⁴°C/s, which exceeds the critical cooling rate of pearlitic rail steel; thus, martensitic phase transformation occurs, forming a hard and brittle white layer structure.

It should be noted that the one-dimensional transient model adopted here entails three principal simplifications: (1) lateral heat conduction is neglected, which is reasonable for the central region of the wheel–rail contact band on tangent track but may underestimate the heat-affected zone width on curves; (2) thermophysical parameters (thermal conductivity k, density ρ, specific heat cp) are taken at room temperature, whereas in practice k decreases and cp increases with temperature – the net effect is an estimated +10% to +20% overestimate of the surface temperature rise and (3) the friction heat source is idealized as a spatially uniform surface flux, while in reality the contact pressure distribution is elliptical, which may introduce a further ±10% variation in local peak temperature. Considering these factors together, the model provides a conservative upper-bound estimate of the white-layer depth, with a cumulative uncertainty of approximately ±15–25%. This level of accuracy is deemed acceptable for engineering failure diagnosis and maintenance threshold definition. For scenarios involving significant lateral thermal gradients (e.g. flange contact on sharp curves), a two- or three-dimensional model is recommended in future work.

From an engineering practice perspective, the critical thresholds identified above should be applied with appropriate safety margins to account for parameter uncertainties and worst-case operational scenarios. Based on a sensitivity analysis of the input parameters, we recommend the following conservative safety factors for field application: (1) For the critical sliding velocity threshold of 0.5m/s for white layer formation, a safety factor of 1.5–2.0 is recommended, meaning that operational sliding velocities should be maintained below 0.25–0.33m/s during braking events; this accounts for uncertainties in friction coefficient, contact patch size and the stochastic nature of wheel-rail adhesion conditions. (2) For the critical heat flux density threshold of 3.6 × 10⁸W/m² for surface austenitization, a safety factor of 1.3 is recommended, corresponding to a maximum allowable heat flux of 2.8 × 10⁸W/m². These safety margins ensure that the thermal phase transformation threshold is not exceeded even under adverse conditions such as contaminated rail surfaces, reduced brake pad performance, or elevated ambient temperatures. The recommended values have been validated against field operational data from multiple high-speed railway lines in China.

The martensitic white layer has high hardness but poor toughness, and it is very easy to initiate micro-cracks under cyclic wheel-rail load. According to fracture mechanics theory, the stress intensity factor at the crack tip is:

where K_I is the mode I crack stress intensity factor (MPa·m¹/²), σ is the tensile stress at the crack tip (Pa) and a is the crack length (m). Under wheel-rail contact, the maximum tangential tensile stress occurs at the subsurface, which can be estimated as σmax≈0.3 pmax. With pmax = 1200 MPa, the maximum contact stress gives σmax≈360 MPa. Considering the residual tensile stress caused by phase transformation (about 200-300MPa), the total tensile stress at the crack tip is approximately 610 MPa.

Based on this stress level, we can calculate K_I for different crack lengths. When a = 10μm (micro-crack initiation stage), K_I ˜ 3.4 MPa·m¹/²; when a = 100μm (crack propagation stage), K_I ˜ 10.8 MPa·m¹/²; when a = 500μm (crack approaching interface), K_I ˜ 24.2MPa·m¹/². The fracture toughness K_IC of martensite is about 15\20MPa·m¹/², while the fracture toughness of pearlite matrix is about 50\60 MPa·m¹/². The calculation results show that when the crack length reaches about 200μm, K_I ˜ 15.3 MPa·m¹/², which reaches the fracture toughness of martensite, and the crack begins to propagate rapidly. When the crack propagates to the martensite-matrix interface (a≈1 mm), K_I ˜ 34.2 MPa·m¹/², which is between the fracture toughness of martensite and pearlite.

The initial propagation angle of about 80°to the tread is determined by the combined action of stress state, microstructure anisotropy and crack type. Under rolling contact, the stress field is asymmetric; the maximum shear stress occurs at an angle of about 45° to the surface, but due to the presence of residual tensile stress perpendicular to the surface caused by phase transformation, the direction of the maximum principal stress deflects to about 80° to the surface. Martensite has a typical acicular structure, and the martensite needles are generally oriented at a large angle to the surface due to the directional cooling during quenching. Cracks preferentially propagate along the martensite needle boundaries, which also results in an initial propagation angle of about 80°. This is a typical mixed mode I-II crack, dominated by mode I (opening mode) and supplemented by mode II (sliding mode). The stress intensity factor ratio K_II/K_I is about 0.2–0.3, which causes the crack propagation direction to deflect from the pure mode I direction (90°) to about 80°.

The crack growth rate in Stage II is estimated using the Paris law da/dN = C·(ΔK)m. For the pearlitic U75VG matrix, the material constants are taken as C = 3.2 × 10⁻¹³ and m = 3.5 (ΔK in MPa·m¹/²), based on fatigue crack growth tests on U75V rail steel reported by Xu, Liu, Wang, and Liu (2025). For the martensitic white layer, standardized FCGR data are not available in the literature; we therefore adopt conservative estimates of C ≈ 1.5 × 10⁻¹¹ and m ≈ 3.0, drawn from studies on high-carbon martensitic steels of comparable hardness (680–750 HV) (Murakami, 2002). At a representative ΔK of 10MPa·m¹/², these parameters yield da/dN ≈ 3.5μm/cycle for the white layer versus ≈0.04μm/cycle for the matrix – a difference of approximately two orders of magnitude, consistent with the fractographic striation spacing observed in Section 2.2. The uncertainty in the white-layer C and m values is acknowledged; direct FCGR testing on rail white-layer specimens would be a valuable subject for future investigation.

Based on the Paris formula for fatigue crack propagation:

For martensitic structure, taking ΔK≈10 MPa·m¹/² (considering the cyclic load characteristics), we get da/dN ˜ 0.1μm/cycle. Considering that the train passes about 100,000 times per year, the annual propagation distance is about 10mm/year. This propagation rate is extremely fast, which explains why burn defects can develop into serious cracks within a few months in actual operation.

The crack deflection at the martensite–matrix interface is driven by a synergistic mechanical process that can be described in three sequential stages. First, the pronounced mechanical incompatibility between the hard, brittle martensitic white layer (hardness\718HV, incapable of significant plastic deformation) and the ductile pearlitic matrix (hardness \337HV, yield strength - 600MPa) creates a strong elastic-plastic mismatch at the interface: under cyclic wheel–rail contact loading, the matrix undergoes plastic deformation while the white layer remains elastic, generating a residual tensile stress field at the interface boundary. Second, as the Mode-I crack tip approaches the interface from the white-layer side, this interfacial stress field superimposes with the applied crack-tip stress, elevating the local Mode-II (shear) component. The K_II/K_I ratio increases from less than 0.3 in the homogeneous white layer to approximately 0.8–1.0 within \50μm of the interface, indicating a transition from tensile-dominated to mixed-mode loading. Third, under this mixed-mode condition, the crack preferentially branches away from its original \80° trajectory and reorients to a shallow angle of -10–20°relative to the tread surface, following the mechanically weaker interface plane where the local fracture toughness is reduced by the combined effects of microstructural discontinuity, residual stress, and the hardness gradient. This unified mechanical narrative integrates the previously separate observations of modulus mismatch, hardness gradient and stress-state transition into a single coherent mechanism.

When the crack propagates to the martensite-matrix interface, the stress state changes significantly, causing the crack to branch and turn. The specific mechanism includes elastic modulus mismatch, hardness difference and stress state change. The elastic modulus of martensite (about 210GPa) is slightly higher than that of pearlite (about 200GPa). This mismatch causes stress concentration at the interface, and the stress intensity factor at the crack tip is redistributed. The hardness of martensite is 2.2 times that of the matrix. When the crack tip reaches the interface, the resistance to continue propagating into the matrix increases significantly, while the resistance to propagate along the interface is relatively small. Before reaching the interface, the crack is mainly subjected to tensile stress perpendicular to the crack surface; after reaching the interface, due to the constraint effect of the matrix, the shear stress parallel to the interface becomes dominant.

After reaching the interface, the crack changes from mixed mode I-II crack to mainly mode II (shear mode) crack, and the propagation direction changes from about 80° to the surface to about 10–20° to the surface (i.e. nearly parallel to the interface). At this time, the stress intensity factor ratio K_II/K_I increases to about 0.8–1.0, showing obvious shear fracture characteristics. Before steering, the crack tip is mainly subjected to opening force, and the crack propagates perpendicular to the maximum principal stress direction. After steering, the crack tip is mainly subjected to shear force, and the upper and lower surfaces of the crack slide relative to each other, which is why the horizontal propagation zone shows obvious rolling and extrusion morphology.

The crack propagation is also influenced by driving direction, vehicle type and traffic density. The side where the wheel enters the contact area first is more likely to form an 80° initial crack because the tangential force is larger at the entry side. EMUs with larger axle loads (such as 17t) produce higher contact stress, which is more likely to cause crack initiation. Heavy-duty engineering vehicles are more likely to cause burn defects due to their different braking characteristics. High traffic density means more frequent cyclic loads. On the one hand, it accelerates the accumulation of fatigue damage; on the other hand, it increases the probability of braking events, both of which promote the initiation and propagation of cracks.

When K_I exceeds the fracture toughness K_IC of martensite (about 15–20MPa·m¹/²), the crack begins to propagate. Due to the anisotropy of martensitic structure, the crack preferentially propagates along the boundary of martensite needles, so the initial propagation direction is at an inclination angle of about 80° to the tread. When the crack propagates to the martensite-matrix interface, due to the large difference in elastic modulus and hardness at the interface, the stress state changes and the crack branches and turns. At this time, the crack propagation direction changes from the direction perpendicular to the maximum tensile stress to the direction parallel to the interface, forming a horizontal crack. With the continuous action of cyclic load, the horizontal crack continues to propagate, eventually leading to the spalling of the surface material and forming spalling defects. If the crack further propagates into the rail, it may induce a transverse fatigue fracture, causing serious traffic accidents.

The failure mechanism established in Section 3 identifies several critical thresholds that govern the formation and propagation of rail tread burn damage: a critical heat flux of approximately 3.6 × 10⁸W/m²required for surface austenitization; a critical crack length of approximately 200 μm at which the stress intensity factor reaches the fracture toughness of martensite; and the martensite-matrix interface as the preferred path for crack branching and delamination. These mechanistic insights provide a rational basis for designing a targeted, two-level maintenance strategy: source prevention (avoiding the conditions that generate the martensitic white layer) and damage control (arresting crack propagation once the white layer has formed).

The thermodynamic analysis in Section 3.2 demonstrates that the formation of a martensitic white layer requires the rail surface temperature to exceed the eutectoid temperature of pearlitic steel (727°C). This condition is met when three factors coincide: high contact stress, a friction coefficient above approximately 0.3, and a relative sliding velocity exceeding approximately 0.5 m/s during braking. Source prevention therefore aims to keep at least one of these factors below its critical threshold.

Reducing wheel-rail contact stress is the primary approach, as the peak contact stress directly determines the heat flux density. This can be achieved through three complementary strategies. First, track geometry optimization requires strict control of vertical profile irregularities to limit the dynamic amplification of wheel-rail forces. For high-speed lines designed for 350 km/h operation, vertical irregularities in the 1–20 m wavelength range should be maintained within 2 mm, consistent with Class I management values. This reduces the probability of transient contact stress peaks exceeding the nominal Hertzian value. Second, wheel-rail profile matching through conformal profile combinations increases the contact patch area and reduces peak contact stress. Matching 60 kg/m rail profiles with LM worn-tread wheel profiles has been demonstrated to reduce the maximum von Mises stress by approximately 15%, from approximately 1,200 MPa to approximately 1020 MPa, thereby reducing the heat flux by a similar proportion. Third, axle load control requires limiting the maximum axle load of vehicles, particularly slow-moving engineering vehicles with different braking characteristics, to 17 tonnes on high-speed lines to avoid excessive quasi-static contact stress.

Controlling friction coefficient and surface roughness is equally important, as the friction coefficient is a direct multiplier of the heat flux. Maintaining a low and stable friction coefficient reduces the thermal input for a given contact stress and sliding velocity. Surface roughness control during grinding is essential, as rail grinding itself can generate sufficient heat to produce a white layer if parameters are not properly controlled. The post-grinding surface roughness should be maintained at Ra≤6.3μm; coarser finishes have been reported to increase the effective friction coefficient by approximately 30%, which directly increases the propensity for burn formation. A three-stage grinding procedure (rough grinding → semi-finish grinding → finish grinding) using progressively finer abrasive grits is recommended, with the final finish grinding pass employing a grit size of 120# or finer to achieve the target surface quality while minimizing thermal input. Based on the thermal analysis, excessive grinding pressure or speed can produce a heat flux comparable to that from wheel-rail sliding during emergency braking, so grinding pressure should be maintained below 2.5 MPa and grinding speed below 20km/h to avoid grinding-induced martensitic transformation.

Optimizing train braking strategy represents the most direct means of source prevention, since the relative sliding velocity is the primary driver of heat generation. Reducing the frequency and severity of wheel-rail sliding events during braking directly mitigates the thermal conditions for burn formation. Electro-pneumatic blended braking with priority given to regenerative braking reduces the reliance on friction braking and thus the probability of wheel-rail sliding, as electric braking can dissipate the majority of the train's kinetic energy without generating heat at the wheel-rail interface. The maximum service braking deceleration should not exceed 0.7 m/s² under normal operating conditions, and emergency braking should be avoided whenever operationally feasible. Increasing the braking deceleration from 0.5 to 1.0 m/s² approximately doubles the relative sliding velocity, which in turn doubles the heat flux density and significantly increases the risk of surface austenitization. Modern wheel slide protection (WSP) systems should be properly calibrated and maintained to prevent wheel locking and uncontrolled sliding during braking, as effective WSP is arguably the single most direct engineering measure for preventing the thermal conditions that produce rail burn.

For burn defects that have already formed, the objective is timely detection and appropriate intervention to prevent progression from a manageable surface defect to a safety-critical crack or rail break. The fracture mechanics analysis provides quantitative criteria for determining the appropriate intervention level based on white layer thickness and associated crack length.

Optimizing flaw detection and condition monitoring requires a combination of complementary nondestructive testing techniques. Ultrasonic testing is effective for detecting internal cracks extending beyond the white layer, while magnetic particle inspection provides high sensitivity for surface-breaking cracks. Eddy current testing can provide a qualitative estimate of white layer thickness based on the difference in electromagnetic properties between martensite and the pearlite matrix. Risk-based inspection intervals should be implemented based on the mechanistic understanding that white layer thickness governs crack propagation rate. For sections where burn has been detected and the white layer thickness exceeds 1mm, the inspection interval should be reduced from the standard 3 months to 1 month, as cracks in this thickness regime may propagate at rates exceeding 1 mm/year. For high-risk sections such as long descending grades and small-radius curves, quarterly inspections are recommended, as these locations experience more frequent braking events and higher wheel-rail forces. For general straight track sections, semi-annual inspections can be maintained as the baseline interval. Establishing a digital record for each identified burn defect, documenting its location, dimensions, white layer thickness and any associated crack length at each inspection interval, enables trend analysis and supports predictive maintenance planning.

Implementing a science-based grinding repair strategy is critical, as the quantitative relationship between white layer thickness and crack stress intensity provides a mechanistic basis for determining the appropriate grinding depth. A three-tier strategy is proposed based on white layer thickness. For preventive grinding when h_wl < 0.5 mm, the maximum subsurface crack length is typically below 200μm, with the stress intensity factor at the crack tip well below the fracture toughness of martensite. The crack is in the stable initiation phase, and preventive grinding to a depth of 0.3–0.5 mm is sufficient to remove the entire white layer with an adequate safety margin. For corrective grinding when 0.5 mm ≤ h_wl<1.5 mm, as the white layer thickness increases, the crack extends deeper and K_I approaches the fracture toughness of martensite. Crack propagation has entered the accelerated stage, and crack branching along the martensite-matrix interface may have initiated. Corrective grinding to a depth of 1.0–1.5 mm is recommended to ensure complete removal of both the white layer and any associated subsurface cracks, with post-grinding verification by metallographic inspection or surface hardness measurement to confirm complete white layer removal. When h_wl≥1.5 mm, the crack has likely propagated beyond the white layer into the pearlite matrix and has branched along the martensite-matrix interface. The stress intensity factor approaches the fracture toughness of pearlite, and the presence of undetected crack branches makes reliable removal by grinding uncertain. Immediate rail replacement is the prudent course of action. After each grinding intervention, surface roughness and hardness should be measured to verify that the white layer has been completely removed and that the grinding process itself has not introduced a new heat-affected layer. The interval between preventive grinding cycles should be determined based on the rate of white layer re-formation under specific operating conditions; for high-incidence sections, the grinding interval may need to be reduced from the standard 15 million gross tonnes to approximately 10 million gross tonnes of accumulated traffic.

Operational management considerations also play a significant role in damage control. Engineering vehicles are a significant contributor to burn defect initiation due to their higher axle loads and frequent braking, so their operations on active high-speed lines should be scheduled and supervised to minimize wheel idling, wheel sliding and unnecessary emergency braking events. While gauge face lubrication is an established practice for reducing wear and corrugation in curves, the application of lubricant or friction modifiers to the rail tread should be avoided, as reduced friction may increase the probability of wheel sliding during braking, paradoxically raising the risk of thermal damage. The effective implementation of the differentiated maintenance strategy depends on the ability of field inspection personnel to reliably identify and classify burn defects, so periodic training on the visual and non‑destructive testing (NDT)-based identification of white layer damage is recommended to ensure early detection and appropriate intervention.

The above thresholds and inspection intervals were derived from operational data of a passenger-dedicated line with a design speed of 350km/h and a nominal axle load of 17 t. For lines with different operating parameters, the following adjustments are recommended as engineering guidance: (a) For lines with design speeds of 250–300 km/h, where the frequency and severity of braking-induced thermal events are reduced, the inspection intervals for high-risk sections may be extended to semi-annual (instead of quarterly), while the white-layer thickness thresholds for grinding remain unchanged; (b) for mixed-traffic or freight lines with axle loads of 23–25 t, the white-layer thickness threshold for immediate rail replacement should be lowered from 1.5 to 1.2 mm, and the corrective grinding threshold from 1.5 to 1.0 mm, reflecting the approximately 30% higher contact stress; (c) for small-radius curve sections (R < 800 m), additional lateral forces and flange contact may accelerate crack propagation at the martensite–matrix interface, and a dedicated field validation program is recommended before applying these thresholds. These adjustment factors are conservative first-order estimates; line-specific calibration through statistical analysis of historical damage records is strongly encouraged.

1. The tread burn observed on U75VG rails in the straight section of a passenger-dedicated line is an RCF failure. The root cause is martensitic phase transformation in the rail head surface layer, induced by the instantaneous high temperature generated during wheel-rail relative sliding under train braking. The resulting hard and brittle martensitic white layer serves as the initiation site for surface cracks and governs their subsequent propagation behavior.

2. Thermal-mechanical analysis of wheel-rail contact shows that when the relative sliding velocity exceeds 0.5 m/s, the heat flux density in the contact patch can reach 3.6 × 10⁸ to 1.2 × 10⁹ W/m². Temperature calculations using a one-dimensional transient heat conduction model indicate that the surface temperature can reach approximately 1,092°C within 0.1 s of sliding, exceeding the eutectoid temperature (727°C). The subsequent cooling rate of approximately 1.6 × 10⁴°C/s far exceeds the critical cooling rate for martensite formation in pearlitic rail steel. The resulting white layer, with a maximum observed thickness of 2.5 mm, exhibits a Vickers hardness approximately 2.1 times that of the pearlite matrix.

3. Fracture mechanics analysis indicates that crack propagation transitions from the stable to the accelerated stage when the crack length reaches approximately 200μm, at which point the mode I stress intensity factor (K_I ≈ 15.3 MPa·m^1-2) approaches the fracture toughness of martensite (15–20 MPa·m^1-2). The crack initiates at the tread surface and propagates inward at an inclination of approximately 80° (a mixed-mode I + II trajectory), driven by the combined effect of contact stress, residual tensile stress from phase transformation and microstructural anisotropy of the martensite. Upon reaching the martensite-matrix interface, the crack branches and deflects to a predominantly mode II trajectory, propagating sub-horizontally along the interface at a shallow angle of approximately 10–20° relative to the surface. Beyond this stage, continued cyclic loading may lead to surface spalling or, if the crack penetrates the pearlite matrix, to transverse fatigue fracture.

4. Based on the mechanistic understanding developed in this study, a two-level maintenance framework is proposed: (1) source prevention through the control of wheel-rail contact stress, friction coefficient and braking parameters to avoid the thermal conditions required for martensitic transformation and (2) white layer thickness-dependent damage control, in which grinding depth and inspection frequency are determined by the measured white layer thickness. The quantitative thresholds proposed (preventive grinding for h_wl<0.5 mm, corrective grinding for 0.5 mm ≤ h_wl <1.5 mm and rail replacement for h_wl≥1.5 mm) are directly derived from the fracture mechanics analysis and provide a rational basis for differentiated maintenance planning.