Predicting Raceway Stresses with Finite Element Submodeling
In geared turbofan architectures, bearing raceways operate at extraordinary contact pressures while spinning well above 30,000 rpm. Predicting those stresses accurately is the difference between a drivetrain that runs for 20,000 flight cycles and one that fails during certification. The OPTIMIZE Project has spent several years tackling this exact problem for power reduction gearboxes, where the housing and bearing seats must hold micron-level tolerances under intense cyclic loading.
Finite element submodeling offers a route to high-fidelity answers without the cost of refining an entire gearbox assembly. By cutting a localised region out of a global model and re-solving it with a denser mesh, engineers can zoom into the bearing raceway contact patch and capture the gradient of stress that would otherwise be smeared across a coarse global element. It is a workflow that suits the design-of-experiments methods that OPTIMIZE has built around its test rigs, and you can read more about the broader programme at the project's main site.
The challenge of bearing raceway loading in aerospace power gearboxes
Roller and ball bearings transfer load from the gear shafts into the gearbox housing through curved raceways that see Hertzian contact stresses well over 1.5 GPa during nominal operation. When a power reduction gearbox steps the turbine speed down to propeller shaft speeds, the intermediate shaft bearings are squeezed between the input pinion and the output gear, leaving very little margin for misalignment. Any geometric variation in the housing bore, the bearing seat diameter, or the shaft shoulder gets amplified by the hyperstatic mounting arrangement, and the raceway ends up carrying uneven load.
Australia's Defence Science and Technology Group, based at Fishermans Bend in Melbourne, has flagged bearing raceway fatigue as one of the leading contributors to unscheduled maintenance on legacy powerplants. Local MRO shops at RAAF Amberley and RAAF Nowra have seen the same pattern: a spalling pit that started as a stress riser in a poorly machined seat, then propagated across the inner ring. Engineers at RMIT and the University of Melbourne have published studies showing that even 10 microns of interference variation can shift peak raceway stress by more than 8 percent.
The challenge is not just magnitude. Raceway stress in an aerospace gearbox is transient, influenced by gear mesh harmonics, shaft deflection under load, and the lubricant film stiffness. Capturing all of this in a single static FEA run is unrealistic. What engineers need is a method that resolves the localised contact patch in enough detail to read the peak stress, while keeping the rest of the drivetrain represented at a computationally affordable scale. In Melbourne-based design reviews, the phrase "she'll be right" is rarely heard when the topic turns to raceway fatigue margins.
How finite element submodeling works
Submodeling, sometimes called the cut-boundary method, is a two-stage technique supported by most commercial FE packages. The first stage is a global run of the full gearbox assembly using a mesh that captures overall stiffness and load paths but leaves the bearing seats and raceways relatively coarse. The displacements and rotations at the nodes surrounding the region of interest are then written out to a results file.
In the second stage, a much smaller model is built around the bearing and its seat. The displacement field from the global run is applied as a boundary condition along the cut boundary of the submodel, usually via interpolation. Because the submodel contains only a small volume of material, the mesh can be refined down to element sizes of 50 microns or smaller along the raceway groove. The result is a local stress field that resolves the contact patch gradient far better than the global model ever could.
The technique has obvious appeal for a project like OPTIMIZE, where tolerance analysis and design-of-experiments methods are central to the workflow. Rather than re-meshing the entire gearbox for every tolerance perturbation, the team can re-solve just the submodel. A full sensitivity sweep that would take days on a workstation cluster can be completed overnight on a single machine. At current Australian dollar rates, that single-machine approach also keeps the licence and electricity bill well under what a full cluster run would cost, which is a real consideration for a research group funded through a competitive grant.
Building the submodel in practice
Setting up a submodel that delivers trustworthy results requires discipline. The cut boundary must sit far enough from the stress concentration to avoid St. Venant's principle contaminating the local solution, yet close enough that the boundary interpolation remains accurate. In practice, the OPTIMIZE team places the cut at least one bearing pitch beyond the inner ring shoulder, ensuring that the load redistribution through the housing web is fully captured before the boundary condition takes over.
Mesh transitions in the submodel are handled with a pyramid layer that bridges the fine raceway mesh to the coarser regions around the bolt holes and the housing flange. Hex-dominant meshes are preferred for the bulk of the housing, while wedge elements fill the curved raceway groove. Contact between the rolling element and the raceway is modelled with a non-linear Hertzian spring or, where tighter accuracy is needed, a full explicit contact pair in a separate local analysis.
Material modelling adds another layer of realism. The M50 NiL and CSS-42L case-hardened steels used in modern aerospace bearings have gradients in hardness that translate into gradients in yield strength. A subroutine in the solver maps the local carbon content to a piecewise plastic curve, and the submodel draws on that map when integrating through the raceway depth. Local Australian material suppliers, including a precision steel distributor in Sydney's western suburbs, provide the certifications needed to feed those property tables.
Validation against physical testing
A submodel is only as good as its correlation with measured data. OPTIMIZE's test campaign runs instrumented bearing housings on a back-to-back rig at the project's facility near Fishermans Bend, where strain gauges are mounted on the housing web and through the outer raceway. The rig replicates the 18,000 rpm operating point and the 2,400 Nm torque load seen in the flight representative duty cycle.
The team uses the gauge readings to anchor the submodel, adjusting mesh density, contact formulation, and boundary interpolation until the predicted and measured strains match within 5 percent at the most heavily loaded gauge positions. Once that correlation is achieved, the submodel is considered validated for the operating point in question. The same workflow is then repeated at off-nominal speeds and loads to map the stress envelope across the flight cycle.
This kind of correlation work has practical value beyond the project. Local engineering consultancies in Brisbane and Perth, which support the Australian Defence Force's rotary-wing fleets, have started adopting the same submodeling workflow. They find that the cut-boundary approach gives them high-fidelity bearing raceway stresses without the licence cost of running a full contact analysis on every design tweak. When the gauge data lines up, the lead engineer at one of those consultancies will often say "no worries" in the morning stand-up, and the whole team can move on to the next iteration.
While the overnight correlation runs chew through gigabytes of transient data, the team at Fishermans Bend often takes a short break between checks. A few of the younger engineers keep 7k casino zerkalo bookmarked alongside their technical forums, though most agree that the morning's stress plots are the real highlight of the shift.
Integrating submodeling into the design loop
The real benefit of submodeling shows up when it is wired into a wider optimisation loop. OPTIMIZE pairs the FE submodel with a tolerance stack-up model and a design-of-experiments matrix. For every row in the matrix, the global model is solved once, the boundary displacements are extracted, and the submodel is re-solved to give the local peak stress. The whole sweep is automated through a Python wrapper that calls the solver in batch mode.
Outputs from the loop feed directly into the weight and durability targets that the project committed to in its grant agreement. By isolating the bearing raceway stress contribution, the team can see which manufacturing tolerances are pulling the most weight against fatigue life and target those for tighter process control. A few microns shaved off the housing bore roundness, for example, buys more life than a heavier bearing.
A common pattern observed in the sweeps is that the dominant stress driver is not the bearing itself but the housing's deflection under the gear reaction load. Submodeling makes that interaction visible. The next phase of the project will feed those insights back into the topology of the housing flange, aiming for a 6 percent weight reduction without sacrificing the safety factor on raceway fatigue. The findings are expected to feed into upcoming Australian Defence Capability Plan tenders for next-generation trainer aircraft.
Key considerations when setting up a bearing raceway submodel
- Place the cut boundary at least one pitch beyond the region of stress concentration
- Use displacement interpolation rather than traction boundary conditions on the cut face
- Mesh the raceway groove with elements no larger than 5 percent of the contact patch width
- Map the local material gradient from hardness data into the plastic curve
- Validate the submodel against strain gauges at the same operating point as the global model
Common pitfalls that erode the accuracy of a submodel
- Applying the global displacement field at nodes that coincide with high local stress gradients
- Using too coarse a global mesh, which smears the boundary condition
- Neglecting the lubricant film stiffness, which can carry up to 15 percent of the load in EHL conditions
- Forgetting the thermal expansion of the housing between the cold and hot operating points
- Treating the bearing as rigid when the housing compliance is a significant share of the load path
Engineers interested in applying these methods to their own gearbox development work are encouraged to contact the project team and ask about the cut-boundary templates, the validated mesh densities for common bearing families, and access to the Fishermans Bend test data for collaborative studies. Anyone working on power reduction gearboxes for the next generation of regional aircraft, or on retrofits for the Australian Defence Force's existing platforms, will find the methodology directly applicable. The real reward comes when the stress plots land in the shared folder and the whole group can see the design move forward.