Modal analysis for resonance avoidance in gearbox structures
Geared aero engines deliver shaft power to propellers, rotors and generators at speeds and power densities that leave little room for waste. Every gram of housing, every stiffening rib and every gram of damping treatment has to earn its place. Modal analysis — the eigenvalue study of how a structure wants to vibrate — identifies natural frequencies and mode shapes so engineers can predict where external forces will be amplified, and change the design before a prototype cracks. For power reduction gearboxes, that prediction is not academic: it decides whether the unit reaches its rated output, or shakes itself apart on the way there.
The OPTIMIZE project treats vibration as a design variable rather than a test finding, using design-of-experiments methods, simulation, tolerance analysis and physical testing to improve efficiency, durability, weight and power density at the same time. That approach is particularly relevant in Australia, where light aircraft operate from aerodromes such as Moorabbin and Bankstown, rotorcraft fly survey and aeromedical routes through the outback, and summer heat in Brisbane pushes gearbox temperatures well beyond what a cool-climate test cell would reproduce. The following sections explain how a resonance review is built, from excitation spectrum through finite element modelling to measured mode shapes and the design changes that follow.
Why resonance threatens gearbox power density
Resonance in a power gearbox is a power-density problem before it is a maintenance problem. As a gear set rotates, teeth are forced apart and pushed back together at the gear mesh frequency, and that excitation is never a clean sine wave. When it lines up with a natural frequency of the housing, the gear carrier or the shaft system, the structure answers with an amplitude far larger than the force that produced it. The result is cyclic contact stress at the tooth flanks, sharply peaked bearing loads, accelerated spline fretting and a noise signature that reaches the cabin. Each of those effects consumes design margin, and the usual engineering response — heavier ribs, larger shafts, thicker flanges, extra damping coatings — adds mass directly to the power density calculation.
Australian operators recognise the symptom long before the failure. General aviation aircraft in Melbourne and Sydney generate vibration squawks that approved maintenance organisations must investigate under continuing airworthiness requirements, and rotorcraft used for aerial survey or medical transfer accumulate vibration hours at speeds and temperatures that ground rigs never see. An engine that passes a static acceptance test can still reveal a mode during acceleration, so designers need to know the natural frequencies of the assembled unit, not just of its parts. This is exactly the framing behind the project's stated objectives, where vibration control sits alongside efficiency, mass and durability rather than being bolted on at the end.
Turning the gear mesh into a usable excitation spectrum
The first step is to write down what actually excites the gearbox. The gear mesh frequency is simply the number of teeth on the driving gear multiplied by the rotational frequency of its shaft, so a 60-tooth gear running at 200 revolutions per second produces a 12 kilohertz excitation with harmonics at 24 and 36 kilohertz. Load modulation from torque ripple and index modulation from tooth-to-tooth manufacturing variation produce sidebands around each harmonic, and manufacturing scatter widens those sidebands into bands rather than lines. Add the bearing outer-race and inner-race fault frequencies, cage frequencies, the axial orders of the spline, accessory shaft speeds and the whirl orders of the rotor, and the excitation picture becomes dense enough that hand calculation is no longer credible.
That density is why a tachometer-synchronised order spectrum is used instead of a raw FFT. Engineers define the speed envelope that matters — start-up, idle, climb, cruise and overspeed — and then, for every significant order inside it, record the distance to the nearest natural frequency of interest. Resonance avoidance means holding that separation deliberately, not hoping it appears by accident. Transient events such as clutch engagement, gear change or a compressor surge produce broadband content, so the same margin argument applies to transients, and a speed band that is clear in steady state may still be crossed in a few milliseconds.
Modelling the gearbox as it is actually assembled
Finite element modal analysis only predicts useful results when the model resembles the hardware. A real gearbox is a hyperstatic assembly: several tooth pairs share load simultaneously, bearings are fitted into housings that are themselves located by bolted flanges, and splines connect a shaft to a hub through a clamped joint. The static solve converges on a set of contact pressures, but that distribution is not unique, and the stiffness matrix inherits the ambiguity. The practical answer is pre-stressed modal analysis, where eigenfrequencies are extracted about a converged loaded configuration, supported by a sensitivity study that reports how far each mode moves when the load case changes.
The inputs to that load case matter as much as the mesh. Shaft deflection and gear body wind-up decide which teeth actually carry load. Axial clamping force at the splines decides whether that joint behaves as a fixed boundary or a slipping one. Temperature decides the rest. Hot afternoons in Brisbane push oil and component temperatures upward, and the project's research into thermal growth and backlash shows how contact conditions, fit and gear backlash drift as components warm. A winter cold soak on a Melbourne apron does the opposite. Modal results are therefore quoted against a stated thermal state, and re-run when that state changes.
Reading natural frequencies and mode shapes
An eigenvalue solve returns a set of natural frequencies and the mode shapes that go with them, and the useful part is the pairing. A gearbox typically shows families of modes: housing bending and torsion, gear carrier motion on its bearings, an axial spline mode, and mounting modes for accessories such as pumps or generators. The mode shape shows where the energy concentrates, and that is where the design work happens. Thin webs between bearing pockets, fillet transitions at rib junctions and unsupported flanges are common hotspots, because a mode that puts strain there converts a modest excitation into a local fatigue crack. An anti-resonant mode can be useful too: placing an excitation at an anti-resonance keeps the response small, although it is risky if operating speed drifts away from the narrow band that delivers the null.
How far a mode shifts depends on how faithfully the model captures fit-up, which is where the project's tolerance work becomes directly applicable. The effort to optimise spline lead angle tolerance controls axial clamping force, and clamping force controls friction at the spline flanks. Friction decides whether the joint behaves as a fixed or a partly slipping boundary, and that choice moves axial modes by amounts large enough to change a separation margin. When physical testing is available, mode shapes from the model are compared with measured shapes using the modal assurance criterion, and damping ratios are extracted alongside frequency so that predicted amplitudes can be trusted rather than merely plausible.
Predicting forced response across the speed range
Modal superposition ties the two halves together. The measured or analytical frequency response function for each mode is combined with the order spectrum of the running engine, giving a predicted response amplitude per excitation order and a speed at which the response peaks. The graphical form of this is the Campbell diagram, in which the operating speed line sweeps past the constant-frequency modes, and crossings are the resonance conditions that a design review must address. Because the modes are fixed and the speed is not, the practical levers are frequency separation, damping and speed management. A mode parked two orders away from the mesh harmonic is safe; the same mode parked a few per cent from it is not.
The design response is usually local rather than global. Adding a rib to a bearing boss, thickening a web, changing a fillet radius, moving an accessory mount or adding a constrained-layer damping treatment can shift a mode or lower its amplification. Each fix costs mass, which is why optimisation studies weigh the frequency shift against the structural penalty and against fatigue life gained. The commissioning picture adds transients: a unit that is clear of every steady-state order can still be excited during clutch engagement or rapid acceleration, and a gearbox that has been tuned only at a single speed is not finished.
Verification routes and evidence worth keeping
Verification closes the loop between the simulation and the hardware. Physical modal testing typically starts with an impact hammer and small accelerometers on a stationary gearbox, or with a laser vibrometer where rotating parts and low damping make contact measurements unreliable. A tachometer or optical speed sensor converts the test into an order-tracked picture, so measured modes can be compared directly with the order spectrum. Model correlation is judged with the modal assurance criterion on mode shapes and with response function agreement across frequency, and the damping estimate matters as much as the frequency estimate because it sets the peak amplitude.
Useful records carry more than a table of numbers:
- Natural frequencies with the damping ratio and the uncertainty attached to each value
- Mode shapes or operating deflection shapes showing where strain concentrates
- Modal participation and effective mass percentages for the directions that matter
- Model-to-test correlation values for every mode carried forward into a resonance review
- Separation margins against each significant excitation order across the speed envelope
A resonance review is only closed when the operating reality has been checked:
- Tachometer-synchronised speed sweeps covering start, cruise and overspeed bands
- Oil and component temperatures recorded alongside every modal measurement
- Manufacturing scatter modelled as a band of possible frequencies rather than a single line
- Damping sensitivity reviewed for the modes closest to any excitation order
- Frequency, damping, mode shape and separation data filed in one traceable dataset
Australian capability is growing quickly. NATA-accredited calibration and vibration test laboratories operate in Melbourne, Sydney and Brisbane, and the relevant measurement standards sit in the ISO 10816 and ISO 20816 series adopted as AS/NZS documents. Universities including QUT, UNSW, the University of Melbourne and RMIT contribute to measurement, modelling and materials research, while the aerospace and advanced-manufacturing corridor around Brisbane and Adelaide supplies the machining and tolerance capability the work depends on. International benchmarking is equally valuable, and the European aero-engine community offers useful comparison for test practice and for how vibration data is handled in certification discussions.
Watch the modal tests and contribute operating data
The fastest way to understand how this work is done is to watch the physical testing rather than read about it. The project's videos show the rig, the instrumented housing and the excitation equipment in use, which makes the difference between a mode predicted in software and a mode measured on the assembled unit impossible to miss. Reviewing the objectives alongside those recordings shows how vibration results feed back into weight, efficiency and durability decisions rather than sitting in a separate report.
Progress in a vibration programme is shared. Australian operators and maintenance organisations hold tachometer traces, vibration histories and thermal logs from real aircraft, and that data helps prioritise the speed bands and orders that matter most in service. If your organisation maintains geared power units, bring speed and vibration data to the project team, attend a test session at the rig, or partner with a local laboratory to extend the measurement campaign. Every additional measured mode tightens the separation margins and makes the next generation of lighter, stronger, more efficient gearbox hardware a better bet for the skies it will actually fly in.