Heavy-haul freight couplers are safety-critical joints. They connect rail vehicles, transmit longitudinal traction and buff forces, keep adjacent cars aligned, and absorb impact and cyclic loading during service. The hook head and knuckle contact region carry complex tensile, compressive, torsional, and wear loads, and their geometry produces stress concentrations as clearances increase. Casting defects such as shrinkage cavities, porosity, inclusions, and hot tears reduce section integrity and can become fatigue-crack initiation sites. In one service review of 20,000 t mixed freight formations, the crack rate of couplers—mainly knuckles—exceeded 60% within one depot repair cycle, with cracks concentrated at the hook head/knuckle joint and inside the knuckle.
Investment casting is well suited to such parts because an expendable pattern and ceramic shell can reproduce the curved hook head, internal passages, and thick-to-thin transitions with limited machining. The same geometry creates a difficult solidification problem: thin webs freeze early and close feed paths, while the hook head and knuckle joint remain liquid and require feed through a low-conductivity shell. Numerical simulation of filling, solidification, stress, and microstructure is therefore used to move the last-solidifying region into the gating system rather than into the coupler.
Coupler Duty, Material, and Casting Constraints
The coupler must transmit traction and buff loads without relative lateral or longitudinal movement between connected cars. Its load state is not static: impact, tension, compression, torque, wear, and random vibration act together. The hook head and knuckle contact surfaces are the main load-bearing regions, and their geometry is complex. Wear increases the mating clearance, lowers local stiffness, and raises stress concentration, while casting defects interrupt material continuity and reduce strength and ductility.
The case considered here is a large E-grade steel coupler casting with overall envelope 594 mm × 370 mm × 350 mm. The bottom circular hole has radius R1 = 50 mm, and the two side holes in the knuckle region have radius R2 = 30 mm. The minimum wall thickness is about 6 mm, the maximum is 32 mm, and the average wall thickness exceeds 15 mm. The casting volume is approximately 9,534,593 mm³, the surface area is about 871,430 mm², and the blank mass is about 70 kg. These proportions—thin webs combined with thick load-bearing sections—are the source of the feeding problem.
The production material is ZG25MnCrNiMo, an E-grade low-alloy cast steel used under specifications equivalent to AAR M201 and TB/T 2942—2015. It is selected for wear resistance, low-temperature impact toughness, and high strength. The nominal and actual heat composition are given below.
| Element | Specification | Actual heat |
|---|---|---|
| C | 0.25–0.32 | 0.26 |
| Si | 0.40–1.50 | 0.45 |
| Mn | 1.30–1.85 | 1.40 |
| Cr | 0.50–0.60 | 0.55 |
| Ni | 0.45–0.50 | 0.45 |
| Mo | 0.25–0.30 | 0.25 |
| Fe | Balance | Balance |
The quality requirement is not simply “sound casting.” Critical sections must be free of cracks and visible shrinkage; porosity may be tolerated only in non-load-bearing regions; dimensional accuracy and surface quality must limit machining; and the as-cast microstructure must fall within the specified ferrite–pearlite grain-rating range. After quenching and tempering, the matrix is expected to become fine martensite with a strength–ductility balance suitable for service.
Baseline Gating and the Defect Mechanism
The baseline gating system used a horizontal, single-side gate and a top open riser on the flange. The intention was to reduce metal consumption, simplify shell construction, and limit superheat loss. The casting lay on its side, and the riser was expected to feed the upper flange during solidification. The main simulation conditions are listed below.
| Parameter | Value |
|---|---|
| Pouring temperature | 1550 °C |
| Shell preheat temperature | 400 °C |
| Ambient temperature | 20 °C |
| Pouring time, baseline | 15 s |
| Shell thickness | 8 mm |
| Shell material | CBSMS-M chemically bonded shell molding sand |
| Casting–shell heat-transfer coefficient | 500 W·m−2·K−1 |
| Shell–air heat-transfer coefficient | 10 W·m−2·K−1 |
| Mesh size | 8 mm for gating/riser, 5 mm for casting |
Filling was initially fast: the lower part of the casting reached about 0.71 m/s, and slight air entrapment occurred. By t = 11.2 s the entrapment disappeared, the flow became stable, and the velocity fell below 0.36 m/s. The cavity was essentially filled at t = 14.7 s. From a filling standpoint, the baseline system was acceptable: it filled completely and without severe cold shuts.
Solidification was the failure mode. The thin outer walls, shoulders, and thin webs solidified first, while the thick hook head and knuckle joint remained liquid. At t = 1567 s, a large isolated liquid region had formed at the top of the casting and at the hook head/knuckle intersection. A YZ temperature slice at t = 2327 s showed the highest temperature below the joint, after the surrounding sections had already solidified. The joint therefore could not be fed.
Node tracking clarified the feeding-path closure. Node 31404 in the thin wall cooled first and formed a solid region early. Node 18397 near the gate solidified next, before the remote thick section. Because this node closed the gating feed path, the thick section received no sustained feed during the later stage of solidification. The casting followed neither sequential solidification nor simultaneous solidification.
Shrinkage prediction used the Niyama criterion, \(N=G/\sqrt{\dot{T}}\), where \(G\) is the local temperature gradient and \(\dot{T}\) is the cooling rate. A low \(N\) value indicates that the thermal gradient is insufficient relative to the cooling rate, so feeding is difficult and porosity is likely. The baseline simulation produced shrinkage at the top of the hook head, the lower hook body, and the release cylinder. More importantly, the hot-crack index at the hook head/knuckle joint exceeded 0.0115, and the base step exceeded 0.0053. These are section-change regions where thermal gradients and restraint concentrate strain.
The joint is not a good location for a local riser: removing a riser would interfere with machining and surface quality, and wrapping the large shell in insulation was impractical. The gating system itself had to change the thermal field.
Vertical Symmetric Gating and Thermal Management
The redesign changed the casting from horizontal pouring to vertical pouring. Six gates were arranged symmetrically on both sides, the runner volume was increased, a heavy runner was placed at the hook neck, and a top open riser was retained for exhaust and feeding at the hook head. The total gating-and-casting assembly mass increased from about 80 kg to about 150 kg. A 30 s pour was used in the first redesign simulation to keep the fill rate stable; other conditions remained as listed above.
Filling became smoother. The metal rose steadily in the cavity, filled the complex sections completely, and showed no entrapment. At 70% fill, particle tracking gave a maximum flow velocity of only 0.48 m/s. The low, uniform velocity reduced impact on the shell, limited turbulence, and favored gas escape and slag flotation.
The solidification pattern also changed. The hook head/knuckle joint no longer developed an isolated liquid zone. Because the gating system was thicker than the casting, it remained liquid longer and continued to feed the joint. The casting solidified before the gating and riser; the last liquid region was in the gating system. The hook head and the planar connection region solidified at nearly the same time, and both finished before the gating system. This is the intended sequential pattern: critical casting sections freeze while the feed path is still open, and shrinkage is transferred to the gating.
Stress tracking showed the same improvement. Node 59684 at the base reached a maximum effective stress of about 470 MPa at t = 6000 s, when its temperature had fallen below about 938 °C. Node 29772 at the hook head/knuckle joint solidified at about t = 3000 s with a stress of about 370 MPa. Node 35369 at the shoulder gate solidified at about t = 3500 s and was the slowest casting node, owing to the adjacent heavy runner. Node 51539 at the middle gate reached about 400 MPa at t = 6000 s. At that time, most of the casting had cooled to about 800 °C; the joint remained near 1000 °C, and the abdomen was the coolest region. Overall effective stress was low.
Deformation was concentrated at the shoulder edges in the X and Y directions, at about 2.5%, and at the release cylinder and base in the Z direction, also about 2.5%. Four reinforcing ribs below the shoulder restrained Z-direction movement. The hot-crack index remained highest at section changes, which is expected: if those regions do not crack, the probability of cracking elsewhere is low.
The optimized gating produced only small shrinkage at the top of the hook head and the middle of the release cylinder. The total shrinkage volume was about 1.13 cm³, located outside the critical load path. This is a substantial improvement over the baseline, where shrinkage and hot-crack tendency coincided at the joint.
Microstructure simulation used a cellular automaton–finite element (CAFE) model with the nucleation and growth parameters of ZG25MnCrNiMo. Three zones appeared: a fine surface grain zone, a columnar zone growing normal to the shell, and an equiaxed zone in the interior. The thin hook-head XZ section contained a higher proportion of columnar grains and a smaller central equiaxed zone. The YZ shoulder section showed columnar grains along the upper and lower edges. The XY section showed more developed equiaxed grains at the hook head/knuckle joint and the shoulder gate, where solidification took longer; these grains were coarser than in faster-cooled regions. For E-grade steel, excessive columnar structure can reduce formability and toughness, so the process should promote equiaxed grains in critical sections rather than maximize gradient everywhere.
Taguchi Optimization of Pouring Parameters
With the gating geometry fixed, the remaining process window was explored with a Taguchi orthogonal array. The objectives were to minimize the total shrinkage volume in the casting body, \(Y\), and the average effective stress at five selected nodes in critical sections, \(Z\). The factors were pouring temperature (A), shell preheat temperature (B), and pouring time (C). Each factor had four levels, giving an L16(4³) design.
| Level | A: pouring temperature (°C) | B: shell preheat temperature (°C) | C: pouring time (s) |
|---|---|---|---|
| 1 | 1530 | 350 | 28 |
| 2 | 1550 | 400 | 30 |
| 3 | 1570 | 450 | 32 |
| 4 | 1590 | 500 | 34 |
Because both responses are smaller-is-better, the signal-to-noise ratio was calculated as
\[
\eta=-10\log_{10}\left(\frac{1}{n}\sum_{i=1}^{n}y_i^2\right),
\]
where \(y_i\) is the response in run \(i\) and \(n\) is the number of runs. A larger \(\eta\) corresponds to a smaller response and a more robust setting. The ANOVA results are summarized below. At \(\alpha=0.05\), pouring temperature is significant for average stress (\(P=0.036\)); shell temperature and pouring time are not. For shrinkage volume, none of the factors reaches \(\alpha=0.05\), although pouring temperature has the largest \(F\) value. The shrinkage model therefore identifies a dominant trend rather than a statistically proven global optimum.
| Response | Source | SS | df | MS | F | P |
|---|---|---|---|---|---|---|
| Y | A | 0.046 | 3 | 0.015 | 8.838 | 0.102 |
| Y | B | 0.021 | 3 | 0.007 | 3.996 | 0.200 |
| Y | C | 0.028 | 3 | 0.009 | 5.467 | 0.155 |
| Y | Error | 0.005 | 3 | 0.002 | — | — |
| Z | A | 110.959 | 3 | 36.986 | 27.155 | 0.036 |
| Z | B | 3.056 | 3 | 1.019 | 0.748 | 0.572 |
| Z | C | 4.307 | 3 | 1.436 | 1.054 | 0.487 |
| Z | Error | 4.086 | 3 | 1.362 | — | — |
Range analysis gave the same ranking for both responses: pouring temperature first, pouring time second, shell preheat temperature third (A > C > B). For minimum shrinkage, the individual optimum was A3B4C1: 1570 °C, 500 °C, and 28 s. For minimum average stress, the individual optimum was A3B1C2: 1570 °C, 350 °C, and 30 s. The combined compromise was 1570 °C, 425 °C, and 29 s.
| Response | Factor | k1 | k2 | k3 | k4 | R | Rank |
|---|---|---|---|---|---|---|---|
| Y (cm³) | A | 0.982 | 1.015 | 1.119 | 0.999 | 0.137 | 1 |
| Y (cm³) | B | 0.987 | 1.006 | 1.041 | 1.081 | 0.094 | 3 |
| Y (cm³) | C | 1.097 | 1.027 | 0.985 | 1.006 | 0.112 | 2 |
| Z (MPa) | A | 345.738 | 345.468 | 352.049 | 347.748 | 6.581 | 1 |
| Z (MPa) | B | 347.280 | 348.328 | 348.018 | 347.379 | 1.048 | 3 |
| Z (MPa) | C | 348.582 | 347.317 | 347.796 | 347.308 | 1.274 | 2 |
A verification simulation at 1570 °C, 425 °C, and 29 s gave a shrinkage volume of 0.879 cm³, lower than the best orthogonal run for \(Y\) (0.887 cm³ in run 2). The average effective stress was 341.700 MPa, lower than the best run for \(Z\) (343.366 MPa in run 8). At the selected nodes, stress ranged from 186.886 MPa to 578.372 MPa. The maximum hot-crack index decreased from 0.0198 to 0.0188, indicating a lower cracking tendency. The combined setting is not one of the 16 orthogonal runs; it is a compromise verified by an additional simulation.
The dominance of pouring temperature is physically consistent. Superheat controls the liquid lifetime, the thickness of the mushy zone, and the time available for feeding. Too high a temperature increases shrinkage and grain size; too low a temperature reduces fluidity and may cause cold shuts. A value near 1570 °C is about 70 °C above the liquidus of this steel, which balances fluidity and feeding. Shell preheat modifies the cooling rate and thermal gradient: a hotter shell slows cooling, prolongs the feed path, and coarsens the structure. Pouring time controls the fill rate and the initial thermal profile; 29 s is a compromise between stable filling and excessive superheat loss.
Casting Trial, Microstructure, and Mechanical Response
The optimized gating and the combined parameter set were tested in production. Low-temperature wax was injected on a WZL-Ⅰ.Ⅱ horizontal wax press at 48–52 °C, 0.3–0.35 MPa, and 3 L/min; injection time was 50–55 s, followed by a 30 min hold. The shell used a water-glass binder. The face coat consisted of two quartz-sand layers dried at 20–25 °C and ≤60% relative humidity for 5–6 h, with an AlCl₃ solution of concentration 1.08. The backing coat consisted of five to six layers of quartz sand, refractory clay, and ceramic particles dried at 20–25 °C and 40–60% relative humidity for at least 12 h. Dewaxing was done in hot water. The shell was fired at 830 °C for 3–4 h, held for 1 h, and preheated to 400–600 °C before pouring.
The trial pour used 1570 °C, a shell preheat of 425 °C, and a 29 s pouring time. After cooling and cleaning, the castings were heat-treated by quenching at 910 °C and tempering at 590 °C for 2 h. Visual inspection showed smooth surfaces without macro-porosity or inclusions. A randomly selected casting was sectioned at the hook head; no internal shrinkage cavity was found.
Fluorescent magnetic-particle inspection and X-ray inspection were performed at the hook head, the hook head/knuckle joint, and the base. No cracks or shrinkage cavities were detected in these critical regions. This confirms that the redesigned gating shifted the last solidifying region away from the joint.
The as-cast microstructure consisted of ferrite and pearlite. Grain size and morphology met the grade 1–3 requirement of TB/T 2942.2—2018. SEM showed ferrite with lamellar pearlite distributed in the matrix. After quenching and tempering, the microstructure became fine, uniform martensite; SEM showed a dark ferrite matrix with dense fine martensite. This is consistent with the reported behavior of ZG25MnCrNiMo, in which quenched structures are lath martensite and the laths become finer with increasing austenitizing temperature.
EDS detected Fe, C, Mn, Cr, Ni, and Mo in both conditions. After tempering, the alloying elements were more uniformly distributed and no acicular segregation appeared. Absolute carbon values from EDS are not reliable in steel and are not used here; the nominal carbon content remains 0.26 wt%.
| Condition | Average tensile strength (MPa) | Range / maximum tensile strength (MPa) | Average elongation (%) | Maximum elongation (%) | Hardness |
|---|---|---|---|---|---|
| As-cast | 625.83 | — | 1.79 | — | 235–241 HBW (~22 HRC) |
| Quenched + tempered | 1139.17 | 850–1500; max 1450 | 2.73 | 4.5 | 352–375 HBW (38–40 HRC) |
| TB/T 2942—2015 | ≥830 | — | ≥14 | — | 241–311 HBW |
The tensile strength after tempering was well above the 830 MPa minimum, and the highest value reached 1450 MPa. However, the elongation was far below the 14% minimum in the cited standard: 2.73% on average and 4.5% at best. The hardness of 352–375 HBW also exceeded the 241–311 HBW range. A strict comparison therefore shows a strength–ductility imbalance: the trial achieved soundness and high strength, but did not satisfy the full ductility and hardness windows of TB/T 2942—2015. The tempering schedule should be adjusted—higher temperature or longer hold—to raise elongation and lower hardness, even at some cost to strength.
Fractography supports this interpretation. The as-cast fracture was relatively flat and transgranular, with little deformation, indicating low toughness. The tempered fracture showed quasi-cleavage with discontinuous propagation and tear ridges on the cleavage facets, a mixed cleavage–dimple mode. Toughness improved relative to the as-cast condition, but the ductility margin remained limited.
Static Strength Check
A static strength check was performed in ANSYS Workbench using a hexahedral mesh with 5 mm elements, 569,998 nodes, and 168,653 elements. Tensile and compressive loads of 1000 kN were selected from measured coupler load data on the Datong–Qinhuangdao heavy-haul line. The results are summarized below.
| Load case | Maximum stress (MPa) | Location | Maximum Z deformation (mm) | Maximum total deformation (mm) | Deformation rate |
|---|---|---|---|---|---|
| Tension, 1000 kN | 1052.9 | Hook head curved section | 0.463 | 0.645 | 0.078% / 0.109% |
| Compression, 1000 kN | 592.52 | Abdomen rib connection | 1.016 | 1.069 | 0.171% / 0.180% |
Under tension, the maximum stress occurred at the curved section of the hook head and reached 1052.9 MPa. This is below the measured average tensile strength of 1139.17 MPa, but the margin is only about 7.6%. Under compression, the maximum equivalent stress was 592.52 MPa at the abdomen rib connection, where the compressive load path passes through the hook body and local geometry produces stress concentration. Deformations were below 0.2% in both cases. Static strength is therefore adequate under the selected 1000 kN load, but the check does not cover fatigue, wear, or impact, which are the governing service modes of a heavy-haul coupler.
Process Window and Scale-Up Implications
The case shows that gating for large couplers should be designed by solidification modulus and feed-path continuity, not by minimum metal consumption. Symmetric vertical gating reduces lateral thermal gradients; a heavy neck runner and a top riser keep the joint fed; shell preheat and pouring temperature control the liquid lifetime; and pouring time stabilizes the fill. The last solidifying region must lie in the gating or riser, not in the hook head/knuckle joint.
The Taguchi results identify pouring temperature as the dominant factor for both shrinkage and stress. The recommended window is 1570 °C, a shell preheat of 425 °C, and a 29 s pour. This window is a compromise: the individual shrinkage optimum favored a hotter shell and shorter pour, while the stress optimum favored a cooler shell and a 30 s pour. Because the ANOVA for shrinkage was not significant at \(\alpha=0.05\) and the design did not include factor interactions, the window should be treated as a robust starting point rather than a global optimum. Production should control pouring temperature, shell preheat uniformity, and pour time with tighter tolerances than the experimental levels.
The trial also shows the limit of simulation and NDT. Soundness and strength were achieved, but the ductility and hardness did not match the full specification. For heavy-haul service, heat-treatment qualification, fatigue testing, fracture-toughness testing, and wear testing are needed in addition to static strength and X-ray inspection. The shell system, wax injection, dewaxing, and firing schedule must also be controlled at scale, because a large low-conductivity shell makes preheat uniformity critical. Investment casting can meet the geometric and integrity demands of large heavy-haul couplers when the gating system controls the last solidifying region and the heat treatment is tuned to the required strength–ductility balance.