In the digital-design-engineering world, at the foundation of innovation in advanced manufacturing, AI’s “deep learning” has the potential to transform how the world makes products—in highly positive ways.
There’s an urgent opportunity, right now, to fully exploit the tools of computer-aided engineering (CFD, FEA, electromagnetic simulation and more) using the capabilities of AI. Yes, we’re talking about design optimization—but it’s optimization like never before, automated with machine learning, at a speed and level of precision far beyond what can be accomplished by most manufacturers today.
We’re talking quantum leaps in efficiency and accuracy: AI tools can cut simulation times from hours to only seconds, employing deep learning to automatically evaluate, and then incrementally modify, the geometry of a part—within bounds that the user dictates—in order to create specific outcomes. The resulting, final design achieves the ideal combination of whatever attributes its makers have prioritized: lighter weight, stress and fatigue reduction, optimum fluid flow, heat exchange, conductivity, durability, part consolidation, and more.
How is this possible? With less number crunching, not more.
Now there’s no escaping the laws of physics; you have to do your design diligence, using whatever market-leading CAE tools are most compatible with your company’s product requirements. But what AI software can add to the digital-design platform is the ability to work within your existing simulation tools and reduce the need to calculate every single differential equation involved.
AI accomplishes this feat by solving the CFD or FEA equations in a non-traditional way: machine learning examines, and then emulates, the overall physical behavior of a design, not every single math problem that underlies that behavior. This uses far fewer computational resources while achieving an extremely robust evaluation of the design in every applicable environment. Hundreds of thousands of design candidates can be simulated and evaluated in less than a day. Bottom line: Applying AI amplifies the typical 10-20% performance improvements of simulation tools alone—up to 30% and higher. (Of course it follows that real-world testing of finished parts remains an essential task to ensure that all quality and performance metrics are met.)
PhysicsX
Additive manufacturing—a unique opportunity for AI
While machine learning can certainly benefit the design of products that are produced via any type of manufacturing process or technology, it’s with additive manufacturing (AM) that AI is perhaps the most complementary. Machine learning can fully explore the AM-design space, identifying the true limit of every type of physics that will apply to a specific component. This unleashes AM’s unique power to deliver whatever level of geometric complexity will enable the most creative and cost-efficient solution to a difficult engineering challenge.
This combination of additive manufacturing and AI has now been successfully applied to optimize and improve the performance of such disparate additively manufactured items as a 3D-printed heat exchanger used on jet engines, a championship-winning motorbike, the impeller blades of a cardiac pump for patients with heart failure and dozens of other applications in advanced industries.
PhysicsX
What’s more, certain AM-system makers have also recognized the value of this capability to improve their own machines--saving time, boosting performance, and fine-tuning the accuracy of their prints.
Here’s an interesting example of one AM equipment provider that used deep-learning software to optimize their 3D printer:
Now that advanced metal AM is producing parts certified for rockets, aircraft, and heavy industry (oil & gas, power, etc.), customer demand for larger-volume equipment has been surging. Several years ago, in anticipation of this, California-based Velo3D began designing its larger-volume Sapphire XC (extra capacity, with a 400% bigger build chamber) to include eight 1,000-watt lasers, four times as many as its original Sapphire machine.
Lasers produce soot when they melt metal powder material inside an AM build chamber. During this process some of the material vaporizes and condenses into very small particles that can occlude the lasers as they target the powder bed. The solution for this is to provide a constant flow of inert gas (typically Argon, but it depends on the reactivity of the material being melted) to sweep away the soot as it’s generated.
Sometimes, however, particles can escape this flow and land on the windows through which the laser light enters the chamber, causing contamination and heating that can distort the window itself. This creates what’s in effect an unintended “lens” in the optical path, bending the laser light from its intended direction and defocusing its spot size on the material bed. As this understandably affects build quality, it’s imperative that the laser windows remain clean throughout the build.
Solving the challenge of going bigger
Velo3D had already thought through the optimum gas flow for its bigger machines’ build chambers overall. But they knew that the longer powder bed, greater interior volume, and tighter packing of more lasers would be a challenge when creating optical window nozzles for their XC system. It was anticipated that the amount of soot generated by the new machines would be about four times as much as the original ones.
The company first tried some in-house computational fluid dynamics (CFD) simulations, then outsourced to a software provider as well—but the results fell short of their performance expectations. The time involved in setting up multiple CFD simulation iterations, while manually changing parameters like the diameters of the nozzle holes, was labor-intensive--essentially a lot of painful guess-and-check.
CAD images of a patent-pending Velo3D Sapphire AM system window nozzle (upper left full view, upper right cut view) designed and optimized by PhysicsX. The lower CFD analysis image shows a section view of the highly consistent flow distribution, the result of geometrical shaping and progression of the individual nozzle guide vanes achieved through machine learning.PhysicsX
Velo3D requested PhysicsX to design and simulate a solution. PhysicsX has deep experience in simulation, optimization and designing for tight packages (from considerable work in F1 racing and expertise in data science, machine learning and engineering simulation), plus proprietary simulation-validated tools that can automatically iterate on designs using machine learning/AI-based simulations. The PhysicsX approach involves creating a robust loop between the CFD, generative geometry creation tools and an AI controller to train a geometric deep learning surrogate. The surrogate’s speed, producing high-quality CFD results in under a second, is then exploited with a super-fast geometrical generative method in another machine learning loop, which deeply optimizes the design towards whichever multiple objectives the engineer decides are important. The fidelity of the deep learning tools and robust workflow enables a highly accurate solution for final validation of the results against the validated CFD model.
In the Velo3D window nozzle case, a number of metrics were used to automatically quantify the fraction of the recirculating flow within the argon curtain that was travelling upward towards the window. PhysicsX benchmarked the Sapphire window solution at the start of the project, then applied their proprietary AI/machine-learning software, and ran huge volumes of simulations to optimize the final design. This resulted in a nozzle design that produced the optimum Argon curtain flow, while working within the manufacturing envelope of the additive machine.
Different orientations of flow images of one-half of the eight-window Sapphire XC, showing the consistent gas flow quality achieved despite the extremely tight packaging volume.PhysicsX
The intricacy of the final turning-vane design would be a challenge to many conventional AM systems, but the Sapphire machine’s ability to 3D print extremely thin, smooth and low-angle vanes delivered the geometry that allowed the nozzles to function as intended. The final design was optimized for and produced on an original Sapphire and the first-ever-manufactured Sapphire XC was run successfully with the new window-nozzle parts in place—an example of an AM machine printing its own parts.
An opportunity to transform designs for additive manufacturing
This AM-nozzle-optimization example exemplifies the potential synergy between AI design optimization and 3D printing in a number of ways. No advanced-technology development can happen these days without computer simulation playing a role. Yet the simulation process still involves significant computing resources and hands-on optimization skills that slow process improvement—which is exactly what the AM industry is still working on.
Here is where AI can step in to intelligently accelerate and automate decision making for designers and engineers working in additive. In the case above, deep learning optimization not only transformed the geometry of a working 3D-printer component, but also improved the function of the key laser system that enables extreme acuity and therefore final product quality. These are the very attributes that the AM industry still needs to scale up and deliver on a global basis—what aerospace, automotive, science, medicine and other fields are looking for from the technology. Deep learning can be the accelerator that pushes the AM industry to achieve these goals.
