Tuning Gearbox Housing Damping for Quieter Aero Engines
Modern geared turbofan engines squeeze more thrust from every kilogram of fuel, yet that mechanical advantage comes with a price: more vibration, more structure-borne noise, and a stricter noise certification regime around busy airports. Engineers working on the OPTIMIZE programme, an effort explained in detail on the OPTIMIZE Project website, have spent recent years studying how housing-borne sound can be cut back without adding weight or sacrificing the gear-mesh efficiency that makes the architecture attractive in the first place.
Housing acoustic treatments are not new, but deciding which damping layer, where to put it, and how thick to make it has historically relied on rules of thumb and one-factor-at-a-time sweeps. A structured design-of-experiments changes that. By laying out an orthogonal test matrix and analysing the response with signal-to-noise ratios and ANOVA, a small team can map the influence of several treatment variables at once and find a robust minimum-noise configuration. For shops working on regional turbofan derivatives or for Australian defence sustainment teams at RAAF Base Williamtown, the same logic scales down to single-housing prototypes.
The work described here draws on the broader line of inquiry that has been published through the project's blog, including prior articles on weight reduction using Taguchi methods. That earlier piece, available as the Taguchi weight write-up, set up the statistical plumbing; this one applies it to acoustics, where the response variables are messier and the constraints on weight, temperature, and oil compatibility are tighter than ever.
What follows walks through how a design-of-experiments is built for a damping treatment campaign, how the test coupons are prepared, how the data is read, and how the results translate into hardware that flies. Engineers in Brisbane, Adelaide, and across the wider Asia-Pacific supply chain can apply the same workflow to their own programs, using laboratory shaker rigs, modal-impact hammers, or even controlled impact testing in a quiet hangar on a Saturday arvo when the apron is empty.
The acoustic problem inside a geared turbofan gearbox
The gearbox housing is the last mechanical component between a meshing gear train and the outside world. Anything the gears excite, through mesh stiffness variation, bearing ripple, or shaft whirl, propagates through the bearing supports and into the case, where the casing walls radiate it as airborne sound. At takeoff, where community noise limits bite hardest, gear-mesh excitation sits in the 1-8 kHz band, exactly where the human ear is most sensitive and where airport authorities such as those serving Sydney Kingsford Smith or Melbourne Tullamarine tend to file noise complaints.
Treatments matter because the bare aluminium or magnesium housings used in many current engines have very low loss factors. Constrained layer damping patches, viscoelastic inserts in rib cavities, and tuned mass dampers welded to the inside skin can each lift the structural damping by an order of magnitude. The catch is that the optimum combination is sensitive to the gear-shaft speeds, the oil-temperature envelope, and the manufacturing spread of every bonded interface. Without a structured experiment, it is impossible to know which of those sensitivities dominates and which can be relaxed.
Why a structured design-of-experiments fits this work
A full factorial sweep across four factors at three levels would already demand eighty-one test runs per housing configuration. Throw in two housing geometries and three measurement conditions and the campaign blows out to a number that no Australian test house, not the National Acoustic Laboratories at Macquarie Park, not the University of Adelaide's vibration rig, not the CSIRO lab at Highett, can realistically schedule. A fractional factorial, an orthogonal array, or a Taguchi L9 or L18 layout collapses the workload to a fraction of that while still letting the analyst attribute variance to each factor and to the interactions that matter most.
Equally important, the design-of-experiments approach delivers robustness. Gears and housings vary from one build to the next, so an optimum that is sharp and narrow is dangerous in service. The signal-to-noise ratio that Taguchi popularised forces the engineer to look for settings that perform well on average and remain stable when the damping thickness drifts, the adhesive cures differently, or the ambient temperature swings between a cold morning at Hobart and a hot afternoon at Darwin. For a maintainer at an RAAF flight line, that robustness is what makes a retrofit acceptable across a fleet built across different production lots.
Building the orthogonal test matrix
The first decision is which factors to vary. For a damping treatment campaign on a single-housing coupon, four controllable factors cover most of the design space: the viscoelastic layer thickness, the constraining layer thickness, the surface preparation method, and the area coverage expressed as a percentage of the available inner skin. Each factor is assigned three levels, chosen from prior literature and the supplier's data sheet, so that the array spans the realistic envelope without forcing any build outside the manufacturing capability of a typical Australian composite shop in, for example, the Bayswater precision-fabrication cluster or the aerospace parks near Nowra.
These four factors and their levels are summarised below.
| Factor | Level 1 | Level 2 | Level 3 |
|---|---|---|---|
| Viscoelastic layer thickness (mm) | 0.5 | 1.0 | 1.5 |
| Constraining layer thickness (mm) | 0.4 | 0.8 | 1.2 |
| Surface preparation | Solvent wipe | Primer + solvent | Grit-blast + primer |
| Treatment coverage (%) | 25 | 50 | 75 |
With four three-level factors, an L9 orthogonal array requires nine coupons, which is a comfortable workload for a single technician. Each coupon is built twice, once for damping measurement on an Oberst beam and once for sound transmission loss in a small impedance tube, so that two response variables feed into the analysis. The repetition also lets the team estimate pure error and run a meaningful ANOVA, which would otherwise be impossible with only nine rows.
Preparing coupons and running the test campaign
Coupon manufacture follows the same discipline that Boeing Defence Australia applies at Williamtown or that the Airbus Pacific team uses around Brisbane when building flight-test hardware. Aluminium 7075-T7351 plates, identical in grade and thickness to the production housing wall, are machined to a common coupon size, cleaned under a documented procedure, and bonded in a temperature-controlled room to keep the adhesive cure uniform. The coupons are labelled with a random run order, not the sequence in which they will be tested, so that any drift in the shaker rig, the microphones, or the operator's technique is averaged across the array rather than aliased onto a particular factor.
The acoustic test itself uses an electrodynamic shaker coupled to the coupon through a stinger, with a miniature accelerometer glued to the back face and a scanivalve microphone one metre away in the semi-anechoic cell. Loss factor is extracted from the half-power bandwidth of the driving point mobility, and the sound pressure level is recorded as a third-octave band average between 1 and 10 kHz. After each run, the rig is checked against a reference coupon, so any deviation greater than half a decibel flags a re-test rather than letting the noise floor contaminate the dataset.
Once the data is in, the two responses, damping loss factor and radiated sound pressure level, are converted into signal-to-noise ratios using the smaller-is-better characteristic for the noise response and the larger-is-better characteristic for damping. Mean effects plots, interaction plots, and a full ANOVA are then generated for both responses, and the analyst looks for factors that drive both in the desired direction at once.
Reading the signal-to-noise and ANOVA outcomes
The dominant signal in most housing treatment campaigns is the viscoelastic thickness. Going from 0.5 mm to 1.5 mm typically doubles the loss factor and drops the third-octave sound pressure level by two to four decibels, depending on the gear excitation spectrum. The constraining layer thickness has a smaller, non-linear effect, with diminishing returns past about 0.8 mm because the steel-like stiffness of the constraining skin begins to dominate the laminate behaviour. Surface preparation matters less than expected once the bonding primer is in place, but the grit-blast plus primer combination still wins by a small margin and survives the ANOVA significance filter at the 95 percent confidence level.
Interactions are where the campaign earns its keep. There is a clear viscoelastic-by-coverage interaction: thin viscoelastic layers at low coverage give almost no benefit, but thick layers at low coverage still outperform bare plate by a wide margin, which matters when designers need to leave inspection windows open. The analysis also surfaces an unexpected constraining-layer-by-surface-preparation interaction, which the team attributes to differences in adhesive penetration after grit blasting. Without the orthogonal array, that interaction would have been invisible to a one-factor sweep.
Validating results on representative hardware
The optimum factor combination drawn from the coupon study is then laid up on a full-scale housing section, using the same bonding procedure and the same cure schedule. Validation runs on a back-to-back shaker rig in a hangar near Bankstown Airport, with the housing instrumented at twenty-three points and excited through the production bearing seat. The broadband sound pressure level measured one metre from the housing falls by three decibels relative to the untreated baseline, which matches the coupon prediction within the experimental uncertainty and clears the project target for cabin-side noise with margin to spare.
Weight is monitored throughout. The chosen treatment adds 1.9 kilograms to the housing, against an internal budget of 2.5 kilograms for the acoustic subsystem. The earlier weight-reduction campaign described in the project's Taguchi article established the baseline mass budget; this damping study simply spends a portion of it, leaving headroom for future iterations or for the addition of tuned absorbers at selected rib intersections. A repeat build with a thinner constraining layer recovers another 200 grams with only a 0.4 decibel penalty, suggesting the optimum can be tuned further if the certification envelope tightens.
Practical recommendations for engineering teams
For teams planning their own damping treatment campaign, the lessons from this work can be compressed into a short list.
- Treat the test matrix as a fixed contract before the first coupon is bonded. Re-planning mid-campaign destroys the orthogonality and ruins the ANOVA.
- Choose factor levels that span the supplier's published range, including one level slightly outside the comfort zone, so the response surface is mapped beyond the safe middle.
- Build duplicates of every coupon so that pure error can be estimated; without it, an ANOVA cannot separate factor effects from background noise.
- Randomise the test order and rotate the operator between coupons to spread systematic drift across all factor levels.
- Record loss factor and sound pressure level as separate responses, then look for factor settings that improve both at once rather than trading one against the other.
- Confirm the coupon optimum on a full-scale or sub-scale housing before committing to flight hardware, because edge conditions, fasteners, and rib intersections change the dynamic behaviour.
- Keep the bonding procedure, cure cycle, and environmental conditions as constant as possible; damping performance is unusually sensitive to adhesive cure.
Each of these points maps back to a known failure mode observed in earlier one-factor-at-a-time campaigns. The structured approach is not slower, because the array cuts the number of builds, and it is not harder, because the analysis is largely automated in any modern statistical package. What it offers, which the old approach never did, is a clear statement of confidence in the chosen optimum and a reproducible trail from raw data to flight hardware.
Teams across Australia who want to dive into the project's broader methodology can reach out through the programme's contact channels or watch the published videos to see the rig in action. Whether the goal is to qualify a quieter housing for a regional jet programme out of Perth, support sustainment work at an RAAF base, or simply reduce cabin noise on a turboprop conversion, the same design-of-experiments framework scales from a university bench to a flight-ready part. The structured approach delivers quieter hardware faster, and the data behind it stands up to scrutiny from any certification engineer.