Finding the ideal gear root radius for a defined load cycle
The root radius of an aircraft gearbox gear tooth has a direct influence on bending strength, fatigue life, manufacturability, and power density. A larger fillet can reduce stress concentration, yet it may also affect tooth thickness, clearance, finishing operations, and the geometry available for meshing. Selecting the best value therefore requires more than checking a single maximum-load case.
Design of experiments (DOE) provides a structured way to investigate this compromise. Instead of evaluating a long sequence of arbitrary radii, engineers can vary the root fillet radius alongside the load-cycle characteristics and other influential design factors. Statistical models then reveal which variables control performance and where a robust design region exists.
For geared aircraft engines, this approach is especially useful because the gearbox may experience high rotational speeds, changing torque, thermal variation, lubrication transients, and manufacturing tolerances. The OPTIMIZE project applies related research and engineering methods to power reduction gearboxes, combining simulation, testing, and design optimization for demanding propulsion environments.
Why root radius matters in gear design
At the tooth root, the transition between the involute flank and the gear body creates a geometric stress concentration. Under transmitted torque, this region carries cyclic bending loads. A small root radius generally produces a sharper transition and a higher local stress concentration factor. Increasing the radius can smooth the load path and lower peak bending stress, provided that the surrounding tooth geometry remains compatible with the design.
The relationship is not unlimited or perfectly linear. A larger fillet may remove material from the dedendum, interfere with the mating gear, reduce backlash margins, or complicate tooling and inspection. In some gear forms, the radius is constrained by the cutter tip, grinding wheel, broaching method, or additive manufacturing process. The optimum is therefore a system-level result rather than the radius that produces the lowest isolated stress.
Root geometry also affects fatigue damage. Gear teeth can survive repeated cycles with stresses below the static yield limit, but local stress peaks, surface finish, residual stress, and material cleanliness influence crack initiation. A DOE study should therefore consider stress amplitude and cycle count together, rather than treating the highest instantaneous torque as the only meaningful response.
Defining the load cycle before running experiments
A reliable optimization begins with a representative duty cycle. This may include takeoff, climb, cruise, transient operation, descent, start-up, shutdown, and occasional overload events. Each segment should be described by torque, rotational speed, duration or number of cycles, temperature, lubrication condition, and the number of repetitions expected during the component life.
The load history can be converted into stress histories at the tooth root using analytical gear calculations, finite element analysis, or a validated hybrid model. For each design point, the resulting stress sequence can be assessed with a fatigue damage method such as Miner’s linear cumulative damage rule. A simplified damage estimate is:
[ D=\sum_i \frac{n_i}{N_i} ]
where (n_i) is the number of applied cycles at stress level (i), and (N_i) is the allowable life at that stress. A design generally requires damage below the project limit, with additional margins for uncertainty and model error.
The load cycle should include mission variation rather than a single nominal profile. For example, an aircraft gearbox may encounter a large number of moderate cruise cycles and a smaller number of severe transient cycles. The moderate portion can dominate accumulated fatigue damage, while the severe portion may govern peak deformation or tooth-root yielding. DOE makes it possible to examine these competing effects systematically.
Building a useful design of experiments
The root radius is the primary design variable, but it should rarely be the only factor. Other candidates include module, pressure angle, tooth thickness, face width, material condition, surface treatment, torque level, speed, temperature, and manufacturing deviation. Including every possible variable can make the experiment expensive, so preliminary engineering judgment and sensitivity screening are valuable.
A two-level factorial design can identify main effects and interactions efficiently during early screening. If the response changes sharply or curves across the radius range, a central composite design or Box–Behnken design can support a response-surface model. The selected method should match the number of simulations available and the expected nonlinearity of the gear-root response.
A practical DOE might use root radius as a continuous factor across a physically feasible range, with torque severity and manufacturing offset as additional factors. Replicated centre points can estimate numerical or experimental variation and reveal curvature. Randomizing the run order is useful when physical tests are involved, since it reduces the risk that temperature drift, tool wear, or equipment changes will be mistaken for a geometry effect.
| DOE element | Example application | Engineering value |
|---|---|---|
| Root fillet radius | Several manufacturable radius values | Measures the direct effect on bending stress and fatigue |
| Torque severity | Nominal, high, and transient load levels | Shows whether the preferred radius changes with duty intensity |
| Manufacturing variation | Radius and profile deviation | Tests sensitivity to production scatter |
| Temperature or lubrication state | Cold, nominal, and hot conditions | Captures changes in material and contact behaviour |
| Finite element response | Root stress, deflection, contact pattern | Provides detailed physics for each design point |
| Fatigue response | Damage fraction or predicted life | Converts stress results into service-relevant performance |
| Robustness metric | Mean response and standard deviation | Identifies designs that tolerate variation |
Combining simulation with physical evidence
Finite element analysis is often the most efficient way to generate responses for a DOE, particularly when the root fillet is small and local stress gradients are steep. Mesh refinement around the fillet is essential. A coarse mesh can underpredict the peak stress or produce unstable results as the radius changes. The model should also represent contact, tooth deflection, shaft and bearing flexibility, and relevant boundary conditions with enough fidelity for the decision being made.
A global gear-system model and a local tooth-root submodel can reduce computational cost. The global model supplies realistic loads and deflections, while the local model resolves the stress concentration. For high-speed geared aircraft engines, the analysis may also need to account for dynamic excitation, misalignment, thermal distortion, and lubrication effects. These factors can alter how load is distributed across the face width and along the tooth.
Simulation results should be checked against physical evidence. Strain-gauged gears, single-tooth bending tests, back-to-back gearbox rigs, and endurance tests can validate the predicted root stress and failure trends. The goal is not to test every DOE point experimentally. Instead, a carefully selected subset can confirm the model, expose systematic bias, and improve confidence in the predicted optimum.
Validation is particularly important when the best radius lies near a constraint boundary. A numerical optimum that depends on an idealized fillet, perfectly aligned shafts, or exact material properties may not remain optimal in production. Comparing simulation and test results also helps establish correction factors or uncertainty bands for the final design recommendation.
Interpreting the response surface
Once the DOE runs are complete, engineers can fit a response-surface model for outputs such as maximum root stress, fatigue damage, tooth deflection, transmission error, mass, and manufacturing cost. Analysis of variance can indicate whether root radius has a statistically and practically significant effect. Main-effect plots show the average trend, while interaction plots reveal whether the preferred radius depends on torque, speed, or another factor.
A common result is a diminishing-return curve: increasing the root radius sharply improves stress performance at first, then produces smaller gains beyond a certain value. If a larger radius adds mass or creates a clearance problem, the best design may be near the point where stress reduction begins to level off. This is more informative than selecting the largest allowable radius automatically.
The optimization should use multiple responses. A useful formulation may minimize fatigue damage and gearbox mass while imposing limits on root stress, contact stress, transmission error, clearance, and manufacturing capability. Weighted desirability functions, Pareto-front analysis, or constrained nonlinear optimization can identify several viable candidates instead of forcing every requirement into one arbitrary score.
Uncertainty analysis adds another layer of value. Root radius, profile form, material strength, torque, alignment, and surface condition all have distributions rather than fixed values. Monte Carlo sampling or polynomial uncertainty models can estimate the probability of exceeding a stress or life limit. A slightly less efficient nominal design may be preferable if it has substantially lower variation and a wider safety margin.
Designing for manufacturing and tolerance
The theoretical radius from a smooth optimization curve may not correspond to a practical production value. Gear cutting and finishing processes impose discrete tool sizes, minimum radii, inspection limits, and repeatability requirements. A useful DOE includes tolerances for the root radius and adjacent profile features, then evaluates how these deviations affect the fatigue response.
Tolerance analysis can be integrated with the design study through nested sampling. The outer experiment varies the nominal design, while the inner analysis samples manufacturing conditions. This separates controllable design effects from unavoidable production variation. It can also show whether investment in tighter process control is more beneficial than changing the nominal radius.
Hyperstatic gearbox structures make this issue more complex. Multiple load paths can redistribute torque as shafts, bearings, housings, and gears deform. Small geometric differences may change the load sharing between teeth or across the face width. Consequently, a root radius that performs well in an isolated tooth model may produce a different result in a full gearbox model.
The final design record should state the nominal radius, permitted deviation, inspection method, material and heat-treatment assumptions, and the validated load-cycle definition. Documentation available through the project’s members area can support collaboration around methods, results, and project-specific engineering knowledge.
Recommendations for an efficient engineering study
A disciplined workflow keeps the DOE focused while preserving the physical meaning of its results.
- Define the complete mission load spectrum before selecting radius levels, including torque, speed, duration, temperature, and lubrication state.
- Use a screening experiment to identify whether root radius, torque, alignment, or manufacturing variation dominates the response.
- Fit a response surface only across a geometrically valid and manufacturable radius range, with mesh convergence checked at critical points.
- Validate the predicted stress trend with targeted tooth-root or gearbox testing rather than relying solely on numerical rankings.
- Select the final radius using fatigue life, contact performance, mass, clearance, tolerance sensitivity, and production capability together.
The study should also distinguish between a screening model and a certification model. A fast analytical or reduced-order model can explore many candidate radii, while a high-fidelity finite element and test campaign can verify the shortlisted designs. Keeping these purposes separate prevents excessive computational cost at the beginning and insufficient validation at the end.
Turning the optimum into a defensible design decision
The preferred root radius is the value that performs reliably across the defined duty cycle and its credible variations. It may not minimize root stress in every operating condition. Instead, it should provide acceptable fatigue damage, adequate overload margin, stable contact behaviour, feasible manufacturing, and tolerance robustness within the gearbox architecture.
A clear decision can be expressed through a design space plot. One axis may show root radius, another fatigue damage, and a third mass or manufacturability score. Feasible regions can be marked using stress and clearance constraints. This visual representation helps designers see whether the optimum is broad and forgiving or narrow and highly sensitive to small changes.
The result should include the fitted model, confidence intervals, validation evidence, assumptions, excluded factors, and recommended controls for production. Reporting only a single radius value hides the reasoning and makes later changes harder to manage. Reporting the response trend and uncertainty allows the design team to adapt when the load spectrum, material, or manufacturing route changes.
Used this way, DOE becomes more than a statistical shortcut. It connects gear geometry with mission loads, fatigue mechanics, manufacturing variation, and experimental evidence. That connection is essential when improving aircraft gearbox efficiency and power density without exchanging a weight reduction for unacceptable durability risk.
Apply this workflow to your own gearbox development by defining the mission cycle, selecting a feasible root-radius range, and building a validated DOE around the factors most likely to affect tooth-root life. Use the resulting evidence to choose a radius that is efficient, durable, and manufacturable under real operating conditions.