Desktop Metal’s Live Sinter: How simulation software is mitigating sintering distortion
Sintering distortion is a fact of life in the Metal Injection Moulding industry. However, through the combination of an experienced eye, the ‘trial and error’ iteration of a part’s design, and the use of sintering supports when needed, stable high-volume production is achieved. With the growth of processes such as metal Binder Jetting, however, the need to manufacture a much wider range of parts at lower production volumes and in a shorter time frame means that a more efficient and streamlined approach is required. Andy Roberts, VP Software at Desktop Metal and the inventor of Live Parts™, presents the simulation software along with a number of case studies illustrating its capabilities
With the ability to eliminate tooling, dramatically reduce manufacturing timelines and create more complex parts than ever before, Binder Jetting (BJT) is quickly revolutionising the way many metal parts are produced. To reach its full potential, though, there is one hurdle that is often overlooked – sintering.
As with Metal Injection Moulding (MIM), BJT parts must be debound and then sintered at temperatures close to the melting point of the alloy being used. The sintering process can cause parts to shrink by as much as 20% and, if not properly supported, there is a real risk of parts slumping and sagging. The result is parts that may emerge from the sintering furnace cracked or deformed beyond usability. Even when not cracked, such parts are typically produced to tight tolerances and any requirement to correct the dimensional accuracy of the parts through post-processing creates additional overhead and expense.
For MIM suppliers, the low-tech solution has for decades been simply a mix of having an experienced eye for appropriate part design and basic trial-and-error: testing various combinations of part geometry, ceramic setters and rafts, then locking in the combination that works for mass production. However, what if, instead of fighting against sintering based deformation, we joined forces with it?
The need for a sintering simulation tool
A first-of-its-kind software application, Live Sinter is capable of simulating, in just minutes, the deformation parts undergo as they sinter, allowing manufacturers to predict how parts will change shape as they densify. Developed over a year in collaboration with Desktop Metal’s materials scientists, the software uses iterative simulation operations to create ‘negative offsets’ – proactively deforming parts by specific amounts in specific directions that allow them to achieve their intended shape as they sinter. Importantly, though, these negative offsets are not simply inversions of the deformation that parts experience during sintering. On the contrary, once the offsets are created, they represent an entirely new simulation challenge.
Fig. 2 shows three parts – the original CAD part (dark grey), a scan of the sintered straight part (purple), and the negative offset part gener- ated by Live Sinter (light grey). The geometry of the sintered straight part matches the shape that Live Sinter produces after its sinter simulation. The software uses the red vectors to derive green vectors that transform original points A to negative offset points C, thus forming the negative offset part. Importantly, the green negative offset vectors are not the negatives of the red vectors. Both the magnitudes and directions of the vectors are different. Live Sinter can generate the green negative offset vectors by using its sintering simula- tion engine. It is also noteworthy that, in performing a negative offset, Live Sinter not only transforms the geometry from points A to C, it also transforms the amount of material (metal powder) and the build orientations of cells. The negative offset process creates an entirely new physics problem that has its own sintering simulation results. The goal of these results is to produce a part represented by points C that sinters to the shape represented by points A.
The need for accurate, automatic, high-speed simulation tools such as Live Sinter is undeniable – in large part because a significant increase in part throughput is among the primary benefits that comes with Additive Manufacturing. If manufacturers hope to capitalise on the speed and agility of mass-production via AM, the rest of the manufacturing design process, as it ramps up to production, must be equally fast, and making sintering more predictable is a critical step.
Creating rafts and supports to hold parts during sintering is time- consuming, certainly, but the process is also expensive. In some cases, supports use more material than the parts themselves and, at scale, it can be the difference between a part that makes economic sense and one that does not. By predicting shrinkage and distortion during sintering, Live Sinter can reduce or fully eliminate the need for supports. The end result, despite sintering with minimal or even no rafts or supports, is parts that emerge from the furnace at near-net shape, reducing waste from failed builds and the time needed to post-process parts to meet specific tolerances.
While Live Sinter works across all Desktop Metal platforms, it is primarily targeted for use with the company’s Binder Jetting systems. Initially, Live Sinter will be available as a standalone application for download and local installation. In a future release, Desktop Metal may also offer a cloud-hosted version of the software.
Live Sinter may also be bundled with the sale of certain Desktop Metal Additive Manufacturing systems and will have features specifically tailored to Desktop Metal’s own technology and material offerings, but the technology is compatible with any sinterbased Powder Metallurgy process, including MIM.
Additive Manufacturing presents new sintering challenges
As the adoption of metal Additive Manufacturing, and, in particular, binder jet systems which deposit liquid binder onto metal powder to build parts layer-by-layer, has grown in recent years, so too has the demand for the sintering of metal powder parts. Whilst MIM and BJT parts share many similarities, including the requirement for debinding and sintering at near-melting point temperatures, in many ways the comparison between the two processes is limited to this step of the process. Creating a MIM part begins by creating a mould. Metal powder and binder are then mixed and injected into this mould to create what are referred to as ‘green’ parts, which go through a debinding process before being sintered in a furnace.
Additively manufacturing parts, by comparison, eliminates the need for moulds and other tooling or fixturing, allowing manu- facturers to quickly create parts, opening the door to highly-complex parts as well as mass customisa- tion from one build to the next.
While AM reduces part turnaround time and increases new part throughput, the lengthy trial-and- error process of finding a sintering solution becomes ever more impractical. To keep up, manufac- turers need simulation tools that can quickly predict how parts will behave in the furnace.
In addition, AM enables parts larger than those typically supported by MIM, meaning distor- tion during sintering can have a larger effect on their final shape. These factors and many others point to the need for a product like Live Sinter – a powerful simulation engine capable of modelling the complex physics at work as metal parts reach temperatures as high as 1,400°C.
The challenge in modelling sintering behaviour
The notion of simulating how mate- rials respond to gravity, shrinkage, density variations, elastic bending, plastic deformation, friction drag and more is not a new idea, but it is an incredibly difficult one. Part of what makes sintering so difficult to model is the fact that it involves both thermodynamic and mechanical transformations that take place under intense heat, making them difficult to observe.
To monitor those changes, manufacturers have only two real options – either halting the sintering process mid-stream and examining parts after they cool, or installing windows in the furnace to observe distortions from images taken at high temperature.
With few other options, the goal has long been to find a way to simulate the process and, though attempts have been made to do just that, those models must replicate a host of factors, including material properties, density, stress, strain – both elastic and plastic, and friction contact, to name just a few.
Further complicating those efforts, simulating the process based on first principles means other factors such as the micro-behaviour of the material at the particle level, models of heat transfer, chemical reactions to heat and the mechanics involved in simulating the shrinkage and plastic deformation caused by factors like creep strain, must also enter the equation.
The difficulty of creating a model that incorporates all these factors means that, to date, most attempts to simulate sintering behaviour have come from academia and have relied on custom code. Ultimately, though, the vast complexity of the models, combined with a lack of data from inside hot furnaces, has made the process virtually impossible.
A novel, integrated approach to simulation
Live Sinter, however, takes an alter- native approach. Rather than working entirely from first principles, it uses a multi-physics engine borrowed from the gaming world which runs on NVIDIA GPUs – the same proces- sors found in high-end gaming PCs. Capable of modelling 700,000-plus particles with mass and radii, the multi-physics engine can simulate how particles collide with each other, as well as with the rigid bodies of arbitrary shapes. In addition, the engine models both body and direc- tional forces as they are applied to the particles.
The result is an extremely fast approximation – simulations are run in just minutes – of the physics inside the furnace, including shrinkage, plastic deformation, friction interaction and more. To refine the engine’s approximations, Live Sinter also employs a meshless FEA engine, which analyses the model at regular intervals to provide Von Mises stress based on data derived from the physics engine.
Complex physics, complex models
In order to simulate the complex behaviour of parts as they sinter, Live Sinter uses a number of approaches.
Simulation of the elastic behaviour of solid parts during sintering builds on a model developed by researchers at NVIDIA. By connecting a collection of simulated particles together with position constraints and dampers, Live Sinter can simulate behaviour such as stretching and compression, both of which are critical to under- standing how metal parts change shape during sintering.
At the same time, the software can model both static and dynamic friction, including the way in which the resulting reaction forces may change from part to part, either due to material differences or the presence of anti-sintering agents. By overlaying a model of plastic defor- mation on the elastic behaviour of the position constraints, Live Sinter can model how creep strain leads to non- uniform deformation of parts.
The system applies creep strain by relaxing the resting lengths of the position constraints over time, which indirectly changes the strain. In areas of higher stress and temperature, that change rate will be higher, leading to more deformation in some areas and less in others.
Fast simulation and excellent accuracy
Armed with its unique, dual-engine approach and highly complex models, Live Sinter can create a simulation of a furnace run in as little as three to seven minutes, as opposed to simulations using complex, dynamic physics which require hours to run. Based on that simulation, the software generates negative offsets in
fifteen to twenty minutes, something that other approaches to sintering simulation are unable to do.
The design of Live Sinter allows the system to strike a balance between speed and accuracy – the GPU-based physics engine provides a quick approximation of the sintering process, which can then be tuned to give more accurate results. The premise is that, while it may not be possible to know the coefficients for every property such as friction, compliance, grain size, diffusion rates or activation temperatures,
it may be possible to tune the physics engine to get the correct resulting shapes and produce successful parts.
To ensure the simulations are as accurate as possible, the first step in using Live Sinter is to tune the system using a series of test parts and scans of these parts after sintering. Once that tuning process is completee, an unlimited number of parts can be processed, simulating sintering distortion and producing negative offset geometry that results in straight sintered parts. Additionally, Live Sinter retains the high level of detail that makes metal AM an attractive manufacturing technology.
Simulating macro and micro distortion effects
Generally speaking, the factors that affect how a part might behave during sintering fall into two main categories: macro factors, which cause distortion to the entire part, and micro factors, which might only occur in a small portion of the larger part. Importantly, Live Sinter compensates for both.
The bulk of the distortions compensated for by Live Sinter are related to macro factors, such as gravity and friction drag, which typically affect the entire part. In the case of the drape bar shown in Fig. 4, parts built without negative offsets showed a pronounced droop in the
middle. This is caused not primarily by gravity as one might guess, but rather the friction drag that prevents the bar’s feet from moving – the bottom portions remain fixed to the setter while the top regions are pulled together, causing a pivoting of the feet.
Were gravity and plastic distortion alone responsible for the distortion, the bar would not have this shape. The two outside feet would be planted flat. Instead, those feet are cocked inward far enough to lift the outside edges. Rather, the distortion is the result of the middle portion of the drape bar shrinking as it sinters. As that middle section
moves, however, the part’s feet remain immobile – friction drag prevents them from sliding inwards as far or as fast as the top of the bar, resulting in its characteristic distortion. While much of this distortion is caused by elastic bending during shrinkage, plastic creep strain takes over and causes the release of stress and freezing of the part in this final drooped shape.
The negative offsets created by Live Sinter instead arch the top of the bar up and tip the feet out, allowing the part to return to straight as it sinters. Because friction drag prevents the feet of the drape bar from sliding very much, the negative offset compensates for this by allowing the part to pivot around its largely immobile feet as it shrinks.
Interestingly, this places most of the part in compression as it sinters, rather than subjecting certain regions, such as the underside of the cross member, to tension, which would likely cause cracks.
For other parts, such as the heater body component shown in Fig. 5 from Desktop Metal’s Fiber™ machine, it became important to compensate for other issues. When built without negative offsets and minimal supports on the sides, the cylindrical part either warps or – in extreme cases – simply collapses on itself during sintering. Rather than build the part as a cylinder, the negative offset generated by Live Sinter creates an oval-shaped part. As it sinters, the combination of gravity and unsupported sides causes the oval shape to drop slightly, returning the part to its proper, circular shape.
In the case of the part known as a ULA bracket (Fig. 6), however, both of these factors – friction drag and gravity – are working at the same time to produce different effects in different regions of the part. When sintered without negative offsets, the feet of the bracket, as with the drape bar, tend to tip inward due to the shrinkage of the upper cross member combined with friction drag of the feet. At the same time, gravity, combined with an uneven weight distribution on the feet, causes the part to warp into a distinctive ‘duckfooted’ posture.
To compensate for these deformations, Live Sinter creates a part whose feet are tipped inward and arches the middle of the bracket, allowing the parts to return to straight during sintering. These macro effects, though, are just one type of feature that can lead to deformation of parts. The second is far more localised and stems from subtle differences in the density of the metal powder used in certain BJT processes.
Due to their symmetrical geometry, parts like the fuel swirler shown in Fig. 7 are far less susceptible to problems such as friction drag. If they do experience drag, their symmetrical shape means the entire part experiences it, so warping seen from looking down on the part is minimised. The pull of gravity also causes little, if any, changes during sintering.
However, when parts do exhibit problems, they may be related to density variations due to powder spreading or compaction in the powder bed.
Though this phenomenon is not completely understood, it is believed that changes in part density can occur as the powder spreading mechanism applies layers of metal powder over the build surface. Slight changes in density, built up through a part layer-by-layer, can cause a part to warp because areas of lower density shrink more than areas with higher density. In MIM, an equivalent scenario is when density variations arise as a result of powder/binder separation during the injection moulding process. Live Sinter, however, can compensate for these density variations and create negative offset designs that, when sintered, will result in straight parts.
Though it already shows great promise as a tool for making the sintering process more predictable, additional improvements to Live Sinter are planned for the future. One project, which will be undertaken in collaboration with Desktop Metal’s software engineers and material scientists, will add a layer of machine learning to identify correlations between changes to certain input parameters and changes in the deformation results in certain regions of a part. For example, if changing the friction coefficients for a part like the drape bar could lead to the part’s feet tilting in to a greater or lesser degree, a machine learning algorithm could spot the association, allowing the system to automatically tune these parameters to correct it.
A second project would add the ability for users to calibrate Live Sinter for even more precise results. Parts produced with negative offsets should emerge from sintering with straight geometry, but there may be cases where the geometry is not perfect, due to the software’s inability to accurately model some aspect of the shrinkage and deformation.
In these cases, users would be able to scan the finished part and identify areas that require fine tuning and the software would recalibrate the negative offsets to produce even more accurate results.
While the ultimate goal of Live Sinter is to eliminate deviations from specified part geometries that necessitate additional machining steps for BJT parts, the software today is unable to model sintering behaviour that emerges from different heat or gas flow patterns in furnaces or other complex thermodynamic transformations that take place in the furnace environment.
To address these issues, future versions of Live Sinter will include tools designed to allow users to scan finished parts, in which any number of parameters (related to different furnace runs, different production runs and more) were altered. The software can then automatically parts with usable results.
While Live Sinter is compatible with MIM parts, initially Desktop Metal will support materials offered on Desktop Metal AM systems and will continue to develop support for new materials in-house. In a future release, material optimisation capabilities will be made available externally, so customers can use Live Sinter to improve the accuracy of parts manufactured using their own novel materials.
Answering a decades-old challenge
By making the sintering process more understandable and repeatable across multiple part designs, Live Sinter could offer benefits not just to individual manufacturers but to the Additive Manufacturing industry as a whole. For decades – even before the emergence of Binder Jetting technology – the Powder Metallurgy industry has struggled with questions of how to create supports that prop up parts in the furnace and, for decades, the answer has been to rely on the intuition of the relatively few engineers with years of hands-on sintering experience. With Live Sinter, however, the process becomes far more controllable – something that will likely help to assuage concerns of potential users.
For many companies, particularly those who have never used a furnace and have no experience with sintering, the notion of additively manufacturing and bulk sintering hundreds – or even dozens – of parts is a daunting one. While companies with MIM experience may have standard support structures they can turn to for Binder Jetting, many more are entering the process with limited exposure to Powder Metallurgy, so a product like Live Sinter is a critical tool that enables them to adopt metal Additive Manufacturing with confidence that they will be able to deploy the technology for mass production.
It has often been said that Additive Manufacturing will change the face of industry – Live Sinter is a crucial step in making it happen.
VP Software, Desktop Metal
From Pim International Magazine