Taking Generative Design to the Next Level on a Large Jet Engine Component

NICK MARKOVIC: Hello, everybody. I want to thank you for joining this industry talk. We will discuss how we apply novel generative design technology to a real-life jet engine component. My name is Nick. I'm a Research Manager based in the London office. And my research area is design and make digital twins.

I was responsible for designing the component using research and product tools. My colleague Philipp is an Implementation Consultant from design and make solution based in Jena, Germany. He was responsible for the final 3D modeling, lattice creation, and builds simulation, plus preparation for large-scale printing.

So this is a standard safe harbor statement. But what I would like to highlight that some of our tools are not in the product right now. Therefore, we cannot guarantee if this will be productized in the near future. If you want to read more, please pause and then unpause and we'll carry on.

OK, so we have an action-packed agenda. And it is broken into five sections. I would give a brief overview of our motivation, then explain what part we are designing and making, and the critical project objectives. After that, I will demonstrate some workflows using generative design and numerical simulation to validate the design outcome.

Philipp will focus on a manufacturing side, how he applied and use the consultant new design guidelines to make a large-scale metal 3D component. Lastly, we will summarize the key takeaways for this project.

Project Monaco stands for manufacturing of a large-scale AM component. This is a three-year joint industrial and academic research program funded by the Clean Sky 2 program, Europe's foremost aeronautical research body.

Clean Sky aims to develop cleaner air transportation using novel and disruptive technologies. That means integrating demonstrated and validated technologies capable of reducing CO2, NOx, and noise emissions by 20% to 30% compared to modern aircraft entering service in 2014.

The Clean Sky program structure is outlined in the diagram below. It is comprised with three innovative aircraft demonstration platforms, or IADPs, and three integrated technology demonstrators, or ITDs.

In project Monaco, we have focused on a large passenger aircraft IADP, eco-design, and engine ITD. The goal and challenges set by Clean Sky allow us to partner with world-class universities and a world-class manufacturer to push the boundaries of engineering and explore new ways of designing and making new things. The consortium team has recently turned an ambitious vision into a viable solution.

Before we get into the technical discussion, I would like to give a highlight overview of the turbine center frame, or the turbine center frame assembly and its primary function. The assembly is broken into three main parts, the inner and outer cases and the struts that are housed between the cases.

The TCF has two main functions. It connects the high-pressure shaft rear bearing with the inner case housing and forms an aerodynamic transmission duct between the high-pressure and low-pressure turbine stages. In addition, oil lines and cooling manifold structures convey through the TCF struts to the turbine and the bearing.

This region is subjected to enormous amount of thermal and mechanical stresses because [? bearing ?] loads are transferred to the outer casing and through the TCF structure. The structure has to withstand up to 600 kilometers of radial loads and up to 400 degrees Celsius thermal loads. The TCF is partly made out of using subtractive and forming processes, but it is hugely energy intensive, costly, and 90% of materials are wasted, which is common in the aerospace industry.

Our primary mission was to build a jet engine with more thrust while burning less energy, emitting less CO2 and NOx, and ultimately created less harm to our planet. We have demonstrated using next generation engineering tools and manufacturing technologies to produce cleaner, leaner, greener, gas turbine engines.

We have done this by focusing on a real-life component, the turbine center frame assembly. The consortium oversaw the requirement and manufacturing of experiments to substantiate many critical processes. These include three designs and making iteration loops. We led the design creation of the turbine center frame assembly using multiphysics generative design technologies and supported the manufacturing and experimental processes using our own CAE and CAM tool, simulation tools.

The aim is to converge the design and manufacture using GE's -large scale AM machine. It is a laser power bed fusion machine that can make parts up to one meter in size using Inconel 718 material. The objective was to reduce the assembly mass, increase performance, add more functionality, and reduce the overall supply chain costs for large-scale manufacturing.

During the three year project lifespan, we have managed to design five TCF concepts. From what you can see left and right, we apply different design philosophies and learn what worked and what needed improvements. In the first design, we applied design-- so in the first design, we applied skin optimisation.

However, it did not meet the stiffness and mask requirements because we couldn't apply particular design objectives. We have played around with the fluids-- the generative fluids set up to get the best result for performance and manufacturability.

In this second attempt, we focus more on the manifold design. It was making it more manufacturable. However, we ran out of time to optimize the structure and applied manual structural reinforcements. Needless to say that, we still failed the mass and stiffness requirements.

In the third design attempt, we have used a better fluid topology optimization tool and utilized the structural generative design outcome as an inspiration to create stiffness and as reinforcements. We managed to meet the stiffness requirements. However, this made a bulky design and the ribs may have compromised the external flow characteristics, which is not desirable.

We created the fourth design, met all the design requirements, used new design guidelines, and improved the design further for large-scale metal additive manufacturing. In the following slides, I will present a high-level overview how I applied [? multi ?] [? Fusion ?] generative design to design 4.1 and 4.2.

In the design phase, we use [? Inventor ?] [INAUDIBLE] Topology Optimization tool to help us create the primary structure. This tool enabled complex model setups and complex objectives on a significant [INAUDIBLE] to meet GE's stringent requirements.

For example, we use cylindrical and symmetric boundary conditions, including large preserve bodies, and assign different objectives for different load cases. Currently you cannot do this in Fusion 360 Generative Design technology.

For example, we apply displacement limits in some regions for load case one. And we maintained stress levels above a certain threshold whilst reducing the starting shape mass. The outcome produced an organic shape highlighted in orange, but undesirable cavities and large undercut regions were observed.

Even though this didn't create the design we wanted, it gave us new ideas. It gave us new ideas, introducing a lattice structure to fill the undesirable areas which is so much more feasible for additive manufacturing. We use the organic shape as an inspiration, and many created the reinforcement regions using the form module in Fusion 360 to create the [INAUDIBLE].

In addition, we integrate the GE manifold structure into the reinforcement regions to further enhance the structural integrity with no mask penalties. With the benefits of using large-scale additive machine, the TCF assembly was reduced from over 100 parts to one monolithic part. Parts consolidation can reduce the overall assembly cost, reliability issues, and supply chain.

All right, the lattice was a sandwich structure that is housed between the inner and outer skins. If you can imagine we have peeled off the top skin, you would see a double conformal lattice structure with variable cell density using the latest Lattice Volumetric Kernel available in Fusion 360.

The implicit lattice creation was a fanta-- was effec-- sorry-- the implicit lattice creation was fantastic technology for creating, viewing, and modifying large-scale lattices at a rapid rate before generating the print files.

We reduced the overall assembly mass by a whopping 35%, which exceeded the overall target by 5%, and met the stiffness requirements where the maximum displacement had to be less than a human hair thickness. They significantly increased the stiffness to weight ratio and manufacture stability. In addition, high-beam densities were applied at high-stressed regions, and low-beam densities were applied in the remaining areas to save additional mass.

After generative design in the [? manifold ?] and TCF structure, our next step was to numerically substantiate the hardware by undertaking a thermal mechanical fatigue assessment. We need to understand if the hardware can survive extreme loadings during its surface life.

Therefore we applied finite element analysis using events in [INAUDIBLE] to calculate the hardware stresses, displacement, and temperatures. The image above shows the aerothermal loading profile when the engine starts and when it switches off, or during the engine cycle.

The image below shows the transient [INAUDIBLE] stresses. You can see a lot of stress fluctuations leading to fatigue damage. Some damage is OK as long it passes the design criteria. Off this simulation, we undertake a fatigue screening assessment to calculate the overall fatigue damage.

Using the Inconel 718 SN data reference from literature, and extracting the maximum amplitude and mean stress envelopes, we created the good-- the modified Goodman diagram to work out the fatigue damage at $90,000 cycles.

Creep effects were assumed to be negligible because the maximum temperature was significantly less than material melting temperature. And, finally, a fatigue-damaged console was created where the warmer colors show severe damage, and the cooler color show low damage. Fatigue damage was in high stress concentrated regions such as holes and fillets. However, the overall damage was less than [? unity. ?] Therefore, the hardware has passed this screen assessment.

Let's talk about how we apply generative design to a fluid problem. We have used generative fluids to design the manifold cooling lines to effectively minimize the resistance of fluid travel. This beautiful render was created by VRED and shows the velocity streamlines created by Autodesk CFD, where the flow starts in one inlet and exits at three outlets.

The model setup is similar to structural generative design where you have to divide the initial starting shape, preserve regions, and key power regions. Also the main objective was to reduce the pressure drop while minimizing the fluid volume during the topology optimization.

This technology was a game changer, as the design outcome reduced the system pressure dropped by 91% compared to the current design. And it also balanced the flow outlets. The algorithm created a smooth organic flow transition around the branch region and minimized the overall turbulence.

I would also like to mention that the lattice structure created extra functionality by insulating between the hot and cold flowing gases. The design feature acted as a thermal barrier, where the [INAUDIBLE] volume was 90% air and 10% metal. We predicted around 16 gigajoules of heat energy was saved during the design life compared to the original design, thus positively affecting the engine-specific fuel consumption.

Initially, we didn't believe the CFD results. Therefore, the consortium printed a 1-to-1 scale manifold that is made in halves. And we could form physical airflow experimentations. We then hooked up many sensors, performed several test conditions, and compared the data with our simulation results.

The bar chart shows the experimental inlet and maximum outlet mach number per load case. And the green and yellow curves show the pressure readings discrepancy in percentage between the CFD and the experimental results.

In summary, there was a 9% discrepancy at low speeds where the mach inlet number was less than 0.3. At higher mach numbers, the flow speeds were incompressible subsonic state, and the maximum pressure [? reading ?] sensor result was around 15%.

Unfortunately, the simulation results did not account for compressibility effects and variable air properties. However, we were happy that the simulation model and assumptions were in good agreement compared to the experimental results. Therefore, we were confident that the outcome created from generative fluids proved to work on real-life problems.

In this chapter, I have demonstrated how generative design was successfully applied to a multiphysics problem and leveraged the design freedom for large-scale additive manufacturing. We have managed to consolidate over 100 parts into one part. We have reduced the overall mass by 35% and reduced the manifold system pressure by 91%.

This part also meets the safe life requirements. The lattice structure created extra system functionality by significantly increasing the heat energy efficiency. And, lastly, we have demonstrated that we can generative design on a large-scale AM structure over a meter in size, which is significant of its kind. I will now pass this talk to Philip, thank you very much.

PHILIPP MANGER: Thanks, Nick. Now let's have a look at what we did to optimize the part for additive manufacturing on a large-scale [INAUDIBLE] system. Let me begin by showing you the evolution from design 2 until design 4.2 in terms of the manufacturability.

The first iteration was printed as 1/8 section of the entire turbine center frame was designed 2. Design 2 contained a simple [INAUDIBLE] structure with reinforcement around the manifolds. In this designed the first comparison between the printed section and the built simulation was done. And this showed similar defects, as you see on the pictures on the left.

In the following designs 3, the manufacturability and stiffness of the design were increased by adding stiffness ribs to reduce distortion errors. This design was also used to improve the build simulation results by comparing the printed and scanned part with the results of the simulation. More details about this will be showing on the following slides.

As my colleague Nick showed you, the final designs 4.x met the stiffness and weight requirements by implementing a shell lattice design. Design 4.1 was then used to run a full scale print simulation to predict potential errors-- more details about this on the next slides.

And, finally, this time 4.2, which is very similar to 4.1, but here we applied some additional design for additive manufacturing techniques to optimize the part for large-scale printing. In the following slides I will now want to share more details about the steps.

To optimize the design for the laser power bed Fusion manufacturing process, the team started with some more fundamentals. While working on design 2, the project team also develop design guidelines for Inconel 718. By working on design 2, the project team also developed design guidelines for printing using Inconel 718 in a laser power bed system.

To do this, the team created and printed a range of standard design features to validate the limits and restrictions for Inconel. The results of this activity were published in the journal Laser Application volume 43 in 2022. This learnings were then applied to improve the manufacturability of design tool, and also then on of the final design 4.2, what you see here on the bottom below.

The next step was then to validate and refine the build process simulation. For this, an 1/8 section of design 3 was printed at the Technical University Hamburg, and then 3D scanned at the Autodesk Technology Center in Birmingham.

With this 1/8 section, we also printed a test part for additional validation. On the test part, the sidewalls tend to distort. Using the same process parameters and materials, we then put a build process simulation of design 3.

The scan results and simulation results were then compared. And the comparison results were used to refine the build process simulation in Netfabb by adjusting the laser absorption ratio of this material slightly. And this is a value which is hard to measure, and therefore can vary from powder to powder, and therefore could point to do these adjustments.

Our designed 4.1, the full-scale thermal and mechanical process simulation was undertaken as well to understand potential problem areas and build distortions. We used this information then to improve design 4.2 and its required support structures.

For example, in the simulation results you see the biggest distortion is happening at the support structures in the inside. And we prevented this by adding stiffness ribs. On the following slide, I will share some details about the design changes and the final lattice creation.

[? Pulse ?] activities had a better print quality and reducing the risks of failures during the print [INAUDIBLE]. Now let's have a look at design 4.2 and what we did to further improve its manufacturability. On this design, we are applying additional design for additive manufacturing techniques.

On the left and on the right, you see we are adding ribs to avoid part-to-part supports at the manifold openings. In the middle, you see we added bigger [INAUDIBLE] and the transitions to reduce the stress concentrations and improve its fatigue performance.

One of the most important step was to modify the manifold, which I will show you at the next slide. Like I mentioned on the last-- like just mentioned, an important step was to modify the manifolds to avoid any internal supports and improve the internal surface quality.

On the left, you see the down-facing areas that require supports on 4.1. Please focus on the orange areas. On the red, you see 4.2 where the manifolds were modified by reducing the entire radius at the top to bring this radius under the maximum [INAUDIBLE] diameter or radius that can be produced without the need of any additional support structures. The maximum value for this was validated within the design guidelines for Inconel 718.

Results of this undertaking can be seen by the 22% reduced predicted support volume. Last missing piece after modifying design was the creation of the final lattice. Please let me now show you the process of this lattice creation. The lattice was created using Autodesk's Volumetric Kernel.

The first production version of this tool is called Volumetric Lattice. And it's available in the product design extension in Fusion. For this process we first took the internal cavities and the outside shell, and then we defined position and created a lattice with this models.

The lettuce was specifically positioned, but it's not blocking any of the [INAUDIBLE] holds which you see here on the left of the screen. Here, Autodesk Volumetric Kernel was very helpful because it enables you to perform transformations and modifications of a lattice very quickly, even if you have such a good [? shell-out ?] lattice we had to use. The lattice we created is a conformal lattice wrapping around the main core structure, and has increased beam thicknesses towards the outer surface of the reinforcements.

One of the last steps was then to cutting this final lattice with a 2.0 millimeter overlap to the outside, or shell. And this is a manufacturing requirement because we wanted to use different build process settings for the outside and also for the lattice. And the reason for this was to reduce the overall build time.

Now let's have a look at the final build preparation which were done to optimize the part to be manufactured on a large scale laser power bed Fusion system. Here we worked closely together with GE Additive to create build support structures for their large-scale power bed system.

These supports were designed to minimize post-processing efforts on such a large-scale structure. For this reason, we only applied supports which can be broken off afterwards by hand, and which have defined breakpoints at the contact surffaces with the part.

In general, we only use two types of supports for use, mainly wall supports with reinforcement ribs for part to build [? 1/8 ?] supports. And for part-to-part supports, we implemented a third breakpoint in the middle, and/or we use three supports.

After the final build preparations on design 4.2 were done, we were able to print the part at GE Additive. We were able to print the part within 36 days without any build errors or issues. Sadly, we cannot show you the final design. But the public presentation will be at TSAS in 2022, so please have a look in there. Now I would like to hand it over to my colleague Nick to do the final project summary.

NICK MARKOVIC: Thanks, Philipp. So just to conclude, we have made an impact in design and make industries by creating the near impossible in the aviation industry through additive manufacturing and large-scale critical components. The TCF hardware size was approximately 1 meter, and was made out of Inconel 718. The hardware was built using General Electric's large scale powder bed Fusion machine.

During the overall process, only 0.5% of the material was wasted, which is very sustainable compared to traditional methods. This is one of the most giant machines and the most significant metal 3D printed part that Autodesk has made of its kind.

The consortium were exceptionally pleased with printing this part for the first time, as they did not witness any build issues or potential build failures. We worked on the clock to ensure we successfully designed the hardware to print for the first time, and we did it.

This is a monumental achievement for Autodesk and the rest of the consortium. The outcome of this project has proven to be technical readiness level 4, and the project has demonstrated solving tomorrow's aviation problems.

And, lastly, but at least, we would like to thank GE Aviation, the Universities of Hamburg and Dresden team. It has been an absolute pleasure working with you all. And this is something we can all be proud of, which was achieving the near impossible. Thank you for this talk, goodbye.