Optimising oil nozzle placement for uniform planetary cooling with DOE
Modern geared turbofans route enormous power through planetary gear stages, where sun, planet and ring gears spin at speeds that push lubricant films into regimes where ordinary convection simply cannot keep up. In a typical power reduction gearbox, the planet bearings carry the heaviest thermal load, and a few degrees of difference between the hottest and coolest planet can dictate bearing fatigue life and scuffing risk. That difference is the reason so much engineering effort is spent on the humble oil jet, and the reason a disciplined approach to nozzle placement pays off in service. Australian aerospace engineers working on next-generation propulsion systems, including teams in Melbourne's Fishermans Bend precinct and around RAAF base Williamtown, know that even a small improvement in cooling uniformity translates directly into longer inspection intervals and lower cost of ownership.
The catch is that the spray pattern from a single jet interacts with its neighbours in non-linear ways. A jet aimed straight at a planet gear can shadow the next planet in the carrier, while a jet angled slightly outward may miss the planet it was intended for and instead soak the ring gear flange with hot oil. Without a structured method, engineers iterate by moving nozzles one at a time, chasing hot spots around the planet carrier like a game of whack-a-mole, and rarely finding a configuration that holds up across the full operating envelope. Trial-and-error placement wastes test hours and budget, and the underlying physics guarantees that intuition alone is unreliable. What the planetary set really needs is a statistical map of the design space, not another round of educated guesses.
The OPTIMIZE project, supported through CleanSky, treats nozzle placement as a variable to be explored systematically rather than a fixed design choice. By applying Design of Experiments to nozzle placement, the team turns what would otherwise be a long sequence of one-factor-at-a-time tests into a compact, information-rich campaign that reveals interactions between jets and installation constraints. The methodology also dovetails with Australia's growing emphasis on data-driven manufacturing, where local firms from Brisbane's aviation and space hubs to Adelaide's Lot Fourteen precinct are increasingly expected to deliver repeatable, traceable engineering decisions rather than artisanal prototypes. A well-designed DOE campaign is the foundation that turns nozzle placement from a craft into an engineering discipline.
Building a factor space for nozzle geometry
The first task is to translate engineering intuition into a finite set of controllable factors and measurable responses. In practice, the controllable variables typically include nozzle inner diameter, axial stand-off distance from the planet bearing, jet angle relative to the carrier, oil flow per nozzle, and the angular clocking position of each nozzle around the carrier. Continuous factors such as diameter and flow are often run at three or five levels, while categorical factors such as single-orifice versus slotted geometries are run at two. The responses of interest are the mean and standard deviation of planet bearing temperature, the peak temperature on the ring gear flange, the bulk oil temperature rise across the stage, and the pressure drop that the nozzle network imposes on the scavenge pump.
A screening or response-surface model then asks which factors dominate, which interact, and which can be held constant. For a planetary set with three or four planets, the number of geometric configurations grows quickly once nozzle clocking is included, so a fractional factorial or Taguchi orthogonal array is usually the right starting point. Engineers in Brisbane and Melbourne who run campaigns through CSIRO or partner universities such as RMIT and Monash often use these arrays to compress a hundred candidate geometries into sixteen or twenty-seven carefully chosen runs. The result is rarely a true optimum on the first pass, but it is always a map of where the response surface rises and falls.
Screening with CFD before cutting metal
Before any steel is cut, the factor space can be explored with computational fluid dynamics. A conjugate heat transfer model of the planetary set, fed with realistic boundary conditions at the gear-mesh interface and at the bearing housings, lets the team evaluate jets and droplets across the entire design space in a way that physical testing never could. A well-resolved oil spray simulation gives an early read on temperature uniformity and highlights blind spots in the spray pattern that physical tests might miss because of limited instrumentation access.
CFD also helps to set realistic factor levels. If the model predicts that a five-degree change in jet angle shifts the hot spot from one planet to the next, the team knows that jet angle belongs in the experiment with at least three levels and tight resolution. If the model shows that nozzle diameter has a near-linear effect, fewer levels are needed and the team can save those degrees of freedom for other factors. CFD-informed factor selection is one of the quiet wins of a screening campaign, and it is a habit worth keeping in any Australian programme where test slots are scarce and budgets are tight.
Building a test rig that mirrors the engine
Once the CFD-informed design of experiments has trimmed the candidate set, the surviving configurations need to be run on a rig that reproduces the thermal and acceleration environment of the real gearbox. A back-to-back or closed-loop planetary rig, instrumented with thermocouples in each planet bearing, slip rings or telemetry to bring the signals out, and an infrared camera looking through a sapphire window at the ring gear, gives the thermal and visual evidence that catches the first sign of temperature drift. Bulk oil temperature is measured before and after the planetary stage, and scavenge pressure is logged to capture the cost of each configuration in pump work.
The mechanical side of the rig matters as much as the thermal side. Gear geometry, bearing preload, and lubrication conditions all need to be held within tight limits, otherwise the response surface is contaminated with noise from sources other than the nozzle. This is where work on compact oil sumps feeds directly into the cooling study, because the sump geometry controls the oil supply pressure and the air-oil mixture that the nozzles see. A well-designed sump feeding a well-instrumented rig is the foundation that the entire DOE campaign is built on.
Analysing variance and identifying the optimum
The raw data from a DOE campaign carry variation from many sources, and the task is to separate the deliberate factor changes from the noise. Analysis of variance partitions the total sum of squares into contributions from each main effect, each interaction, and residual error. For a planetary cooling study, the dominant main effects are often nozzle diameter and jet angle, while the dominant interaction is usually diameter-by-angle, because a wider jet tolerates a wider angular error and a narrow jet demands precise aiming. The residual variance tells the team how much of the temperature spread is genuine signal and how much is measurement noise.
Once the ANOVA has identified the significant factors, a response surface model is fit to the significant terms, and the fitted surface is searched for the configuration that minimises the standard deviation of planet bearing temperature subject to a constraint on peak temperature and pressure drop. The result is rarely a configuration that an engineer would have picked by eye. It often involves a small nozzle aimed slightly off-axis, paired with a slightly larger nozzle on the opposite side of the carrier, in a pattern that looks odd but cools evenly. The standard deviation of planet temperature can typically be cut by a third or more compared with the baseline, and the peak temperature on the hottest planet falls by a similar margin. Heat treatment distortion and grinding variation are real risks to that gain, and the helix deviation study is a useful reminder that the gear geometry the nozzle sees in service is not the geometry that left the grinder.
Practical recommendations for production programmes
The DOE-driven approach pays off when it is carried into production, not when it leaves a report on a shelf. The points below summarise what consistently separates robust nozzle programmes from fragile ones, and they are written for the engineer who has to defend the configuration at the next design review.
- Start with a CFD-informed screening design rather than a full factorial, so that the factor space is mapped efficiently rather than exhaustively.
- Include nozzle clocking as a factor, because angular position interacts strongly with planet position in ways that intuition rarely predicts.
- Instrument every planet bearing, not just one, so that the standard deviation of temperature can be tracked as a response rather than inferred from a single reading.
- Hold bearing preload, oil inlet temperature, and scavenge pressure within tight limits, because uncontrolled variation in these parameters will mask the effect of the nozzle changes.
- Validate the optimum on a second rig or a second build, because the first optimum often shifts slightly once manufacturing variation enters the picture.
These recommendations are not a complete recipe, but they are a reliable starting point for any team that wants to move nozzle placement from a craft to a discipline. The recipe for repeatable cooling starts with a disciplined approach to the experiment that found it.
If your team is wrestling with cooling uniformity in a planetary stage, or if you would like to discuss how the OPTIMIZE methodology could be applied to a specific gearbox programme, the project team would be glad to hear from you through the contact page. A short note about your operating envelope, your current cooling configuration, and the pain points you are chasing is enough to start a conversation about how a compact DOE campaign could shave the hot spot off your hottest planet.