Optimizing Gear Tooth Root Fillets With FEA And DOE
The root fillet of a gear tooth is a small geometric region with an outsized influence on transmission performance. It connects the involute flank to the gear body, carries repeated bending loads, and often becomes the location where fatigue cracks begin. A radius that is too small can create a sharp stress concentration, while a radius that is too large may interfere with the mating gear, reduce usable tooth thickness, or complicate manufacturing.
Finding the best fillet radius therefore requires more than selecting the largest practical curve. Engineers must understand how geometry changes local stress, tooth stiffness, contact behavior, mass, and manufacturability. These effects are coupled, and they can vary as operating speed, torque, material properties, and production tolerances change.
Finite element analysis (FEA) and design of experiments (DOE) address different parts of this problem. FEA reveals what happens inside a specific design, including the distribution of root stress and deformation. DOE determines which variables matter across a design space and identifies interactions that may be missed by changing one parameter at a time. Used together, they provide a more reliable route to a lightweight, durable, and power-dense gearbox.
The Root Fillet Controls More Than Peak Stress
Gear tooth bending is commonly assessed using a nominal root stress multiplied by factors for geometry, load distribution, dynamic effects, and manufacturing conditions. The fillet radius is embedded in the geometry factor because it changes the path through which force flows from the tooth into the rim. A smooth transition generally lowers the stress concentration, but the benefit depends on the complete tooth profile rather than radius alone.
A larger fillet can improve the local stress concentration factor by spreading load over a broader curved region. However, the result may be offset by reduced tooth thickness near the base, an altered involute start point, or a shortened dedendum. The optimum is often a compromise between a lower local stress peak and sufficient cross-sectional area to resist bending.
The fillet also affects tooth stiffness. A flexible root can increase deflection under load, changing the transmission error that contributes to vibration and acoustic emissions. In a geared aircraft engine, even small changes can matter because high rotational speeds amplify excitation frequency and dynamic loading. The relationship between geometry and noise is therefore relevant to the same optimization that targets fatigue life.
FEA Reveals Local Load Paths
FEA is valuable because analytical gear formulas simplify the tooth into idealized geometry. A detailed three-dimensional model can represent the actual root curve, rim thickness, web structure, adjacent teeth, and contact position. With suitable boundary conditions, the model shows how load spreads through the tooth and where the highest principal, von Mises, or bending stress occurs.
A robust model requires careful treatment of the gear mesh. Applying force to a single node can create an artificial singularity, while an overly rigid contact definition can suppress realistic deformation. Engineers may use surface-to-surface contact, distributed loading, nonlinear material behavior, and multiple mesh positions to capture the changing load path through the mesh cycle. Mesh refinement in the fillet is especially important because the stress gradient is steep.
FEA also supports fatigue assessment. Root stress can be extracted at the critical section, compared across torque levels, and combined with material S-N data or fracture mechanics methods. Results can reveal whether a candidate radius improves nominal bending strength but creates an unfavorable stress gradient or a new critical point near the fillet-to-flank transition.
Simulation quality depends on validation. Strain gauges, photoelastic methods, and instrumented test gears can provide evidence that predicted root strain reflects physical behavior. The OPTIMIZE project’s work on in-situ root stress demonstrates why measurements taken during operation are useful for checking models under realistic speed, torque, lubrication, and thermal conditions.
DOE Maps The Design Space Efficiently
Running FEA for every possible combination of gear parameters would be expensive, particularly when each simulation includes contact nonlinearity, transient dynamics, or thermal effects. DOE reduces this burden by selecting a structured set of design points. The selected cases can cover fillet radius, pressure angle, tooth module, profile shift, rim thickness, face width, material, torque, and rotational speed.
A factorial design can identify primary effects and interactions, while a response surface methodology can approximate the relationship between inputs and outputs. For a more complex geometry, Latin hypercube sampling or another space-filling method may provide better coverage. The important point is that DOE treats the design as a system of variables rather than as a single radius sweep.
The outputs should extend beyond maximum root stress. Useful responses include stress concentration factor, tooth-tip displacement, mesh stiffness variation, transmission error, contact pressure, mass, fatigue safety factor, and manufacturability indicators. A design that reduces root stress but increases transmission error or produces an impractical cutter path is not a robust solution.
DOE also helps identify interactions. For example, the effect of increasing fillet radius may depend on rim thickness. A radius that performs well in a solid gear may provide little advantage in a lightweight thin-rim gear because rim flexibility dominates the response. Similarly, the impact of profile modification can change with operating torque and center distance. Research on center distance variation illustrates how dimensional variation can influence transmission error and noise, both of which belong in a broader gear optimization study.
| Design or analysis feature | What FEA contributes | What DOE contributes | Value for fillet optimization |
|---|---|---|---|
| Local root stress | Resolves stress concentration and load transfer in detailed geometry | Shows how radius interacts with module, rim thickness, and load | Prevents optimization based on a single isolated stress result |
| Tooth deformation | Predicts deflection, stiffness, and mesh-cycle behavior | Quantifies sensitivity to geometry and operating conditions | Links fatigue design with vibration and transmission error |
| Contact behavior | Captures pressure distribution and contact shifts | Identifies combinations that create unfavorable mesh conditions | Helps avoid a root improvement that damages flank performance |
| Fatigue performance | Supplies stress histories or critical locations for life calculations | Establishes trends across torque, speed, material, and tolerances | Supports a safety factor that remains acceptable across the operating envelope |
| Manufacturing variation | Can model specific deviations in geometry | Propagates radius, profile, runout, and alignment distributions | Converts a nominal optimum into a production-capable design |
| Computational cost | Provides high-fidelity responses at selected points | Selects informative simulations and enables surrogate models | Reduces the number of expensive nonlinear analyses |
| Design decision | Explains the physical reason behind a result | Ranks influential variables and interactions | Creates a defensible basis for selecting the final radius |
Coupling Simulation With Physical Evidence
A DOE is only as credible as the data behind it. If the FEA model contains unrealistic constraints or inaccurate material properties, the statistical analysis can precisely describe the wrong physics. Model verification should therefore precede broad optimization. Element quality, convergence, contact settings, boundary conditions, and load application need to be checked before the results become DOE responses.
Validation should compare measurable quantities rather than relying only on a general visual match. Root strain at several torque levels, tooth deflection, mesh stiffness, and dynamic transmission error can each expose a different modeling weakness. Static tests may verify the stress field, while a geared test stand can reveal speed-dependent effects such as lubrication changes, centrifugal growth, and dynamic load amplification.
The physical test does not need to reproduce every design point. A carefully chosen set of validation cases can test the model at low, nominal, and high values of the most influential variables. DOE helps select those cases, while FEA helps determine where sensors should be placed. This creates a feedback loop in which experiments improve the simulation, and the simulation makes experiments more informative.
For aerospace gearboxes, validation is especially important because high power density leaves less room for uncertainty. A fillet that appears safe in a static analysis may experience higher alternating stress when misalignment, shaft deflection, gear eccentricity, and transient torque are included. Physical evidence helps distinguish a genuinely robust design from one that is merely favorable under ideal assumptions.
Tolerances Turn A Best Radius Into A Distribution
Manufacturing never produces one exact fillet radius. Tool wear, cutter geometry, heat treatment distortion, grinding strategy, measurement uncertainty, and local profile deviations all create variation. A nominally optimized radius may lose its benefit if its allowable range is too narrow or if adjacent dimensions shift at the same time.
Tolerance analysis can be integrated with DOE by treating uncertain dimensions as input variables. Monte Carlo sampling, statistical moment methods, or a design for six sigma approach can estimate the distribution of root stress and fatigue life. The analysis may show that the nominal design has the lowest average stress but a neighboring design has a better worst-case response.
This distinction matters in aircraft propulsion. Certification and reliability decisions are based on acceptable performance across specified conditions, not simply on the best result from a nominal computer model. A robust fillet design should tolerate variation in radius, tooth thickness, alignment, center distance, material strength, and applied torque without crossing critical limits.
The analysis should also account for how the radius is created. A mathematically smooth fillet may not be achievable with the selected cutting or grinding process. Tool access, cutter tip radius, finishing allowance, and inspection capability can constrain the design space. Manufacturing engineers should be involved before the optimization is closed, rather than asked to reproduce an idealized geometry afterward.
Selecting A Practical Multi-Objective Design
Fillet radius optimization is rarely a single-objective problem. Lower root stress is valuable, but it must be balanced against mass, size, mesh quality, contact fatigue, manufacturability, and cost. A Pareto analysis can show which designs offer meaningful trade-offs instead of forcing all performance goals into one arbitrary weighting equation.
A practical workflow begins with a parameterized CAD model and a verified FEA template. DOE then defines the simulation points, and response surfaces or surrogate models approximate the performance landscape. Candidate designs are screened against stress, fatigue, stiffness, contact, and dimensional constraints before a smaller group receives higher-fidelity dynamic analysis.
The final candidates should be assessed at the operating extremes that matter most: maximum torque, overspeed, low and high temperature, lubrication limits, transient loading, and relevant assembly misalignment. In a geared aircraft engine, the best radius may differ between the sun, planet, and ring gears because their geometry, load sharing, and structural support are not identical.
Recommended Engineering Practices
- Parameterize the fillet, adjacent involute geometry, rim thickness, and profile shift so every DOE point uses consistent geometry.
- Verify FEA mesh convergence and contact behavior before treating stress or deflection as optimization responses.
- Include transmission error, mesh stiffness, and contact pressure alongside root bending stress.
- Use tolerance analysis to compare nominal performance with worst-case and statistical distributions.
- Validate selected design points with strain measurement and test-stand data before approving the production geometry.
The strongest design decision comes from combining physical interpretation with statistical evidence. FEA explains why a radius changes the stress field, while DOE shows whether that change remains important when other variables move. Together they prevent narrow optimization, expose hidden interactions, and support a gearbox design that is efficient in simulation and dependable in service.
Explore the OPTIMIZE project’s research on simulation, experimental testing, tolerance analysis, and power reduction for geared aircraft engines. Applying the same combined FEA-and-DOE mindset to gear tooth root fillets can help engineering teams reduce weight and vibration while preserving the durability demanded by aerospace propulsion.