0:00 Hello everyone, hello, thank you, there we go, lovely to see you. Right, so I’m Alexander Pluke, I’m the GM and CTO at Additive Flow, division of Nano Dimension, and today I’m going to talk about how we connect systems, how we couple manufacturing and design to get a prediction of real part performance, and take that insight from design into decision you can make.

0:31 So where do we sit? So we’ve had a lot of information today about CAD design, that initial section, also some on the simulation, right, different simulation tools that we know about. Additive plow sits after both of those and interacts with them, communicating to your production. And so what we do is we bring your production decisions up front into your design step, and we communicate with Cam to make that happen and take in machine feedback. As part of N Dimension as well, we operate across a whole bunch of different manufacturing systems, we have also the Deep Cube Division, and they’ve got about 80 patents in industrial process AI and so on.

0:51 And so what we do is, with Ative Flow, we go from this CAD to Cam interface, we look at the process parameters, so any decision that you could be making in that production step, which will influence your quality, which will go into your simulations, which will then drive your metrics on cost, your metrics on production success. And so if we have been inside through de printing, we know that things often fail, right, parts can collapse, it could take a long time to understand actually how will my production process work, and we really help to do that, increasing yield and decreasing cost of production.

1:54 And so once you’ve done that kind of offline design production decision, how do you take that real time data back into your design process? And that’s where we use a whole bunch of machine monitoring technologies, and I’ll talk about that a bit later. What does this mean for the market? So we’ve really really driven around value, around business case, right, and you can see here some of the results that we’ve been able to deliver. Power results like cycle time improvements by up to four times with no geometry change, or taking in an existing geometry that’s already been worked on. Work with companies like Tarta Steel, taking scrap down 44% to zero, so very very interesting kind of work, and I’ll cover how we’ve done that now.

2:41 And so I’m going to talk through the story of your design, your production, to material characterization, and how we actually make that happen in the real world, bridging these silos of engineering that are often disconnected. And so if we take like an example here, this is ceramics workflow for additive manufacturing, we start out hair with the material input, so how will your material properties influence your formulation of a material, so like a slurry that’s going to be centered, how that actually works within your additive manufacturing process, after it’s coming out the build, how are you washing that, and then how you putting that into a Fess, and we do all of the decision- making and the interaction in this workflow, taking into account the real engineering data, your requirements for physics, and connecting a system together.

3:35 And so for the engineers in the room, hopefully you aware of, you know, process structure property relationships, but effectively in layman’s terms it’s saying if you change a production parameter that will cause a material to form differently in your process, and whether it’s more or less solidified, it’s going to be stronger or weaker. So these relationships are what addtive flow optimizes. Again, and so we can take in your existing FAA, your physics, we can take in some of these machine learning models that are also predicting things, great work at a companies like Nastro and others, and pull that back into your decision- making. Once you’ve got your experimental data coming out your machine, we can take that back in and help to really really improve the the outcomes.

4:23 This is the work we’ve done with Airbus, so you probably recognize this case, this is the TMTC case, it’s public, and so working very closely with AirST and Space on this for a few years. And here you can see the build time challenge that their traditional existing topology optimized component was having, it going to take 22 hours to make, right, it’s far far too expensive, that’s 8,000 per part, and that cost really needed to calm down without changing the geometry. So taking the existing to component, we looked at the process parameters in that 3D printing step, and we brought that cost down to 2 1/2k.

5:04 And so that’s us volumetrically allocating, I’ll show you in a moment, where to place the different cres parameters, with their ability to predict the success of the production and predict the requirements being met for that particular component. We then took it further, and then we also brought topology into the loop, and working with a whole bunch of people in this room on using their topology code in kernels as an interop, but we initially founded around the simultaneous optimization of a whole bunch of parameters, including topology and lising and the rest of it, but now very much on the production. And so with scal Maloy, which is an even stronger material, we did a much more aggressive topology optimization, and with our additional parameters we brought the cost down further to 1.8k, right, so it’s a better material meeting the requirements and significantly commercially viable.

6:01 And so actually how did we do that? So we taken the model, we’ve got here a connection of all the physics that are coming in, we can connect to, in this case they’re looking at Fe map and other kind of physics, we’ve got internal inh house, and we’re looking at the metrics here for production and cost simultaneously that are being weighted as objectives. And by weighting these objectives differently, we can trade off effectively against the internal volumetric composition, so you can see here in the blue and the gray regions are solver allocating volumetrically different process parameters in different locations based on the engineering requirements, making sure the parts to successfully print and that cost of the part.

6:40 And then we can trade off against different distributions here of parameters, in this case production speed versus cost versus the stress that’s arising, and one can actually explore that map as as the engineer, instead of wasting a whole bunch of time randomly iterating against parameters, you can volumetrically change your production parameters with confidence on the success.

7:02 Let’s look at another example, this is a heat exchanger, this is something we did with Zeiss, inside this cavity goes a CT scanner, that when it gets hot the CT scanner gets very unhappy, stops working, right. And so what Ze really wanted to do is bring down the cost and increase the thermal efficiency of this component. And so here the same story, we were allocating the different production parameters of this particular alloy, allowing Zeiss to trade off without changing that geometry the cost versus the performance.

7:31 So what happens about how do you produce these components, right, and this is where interrup formats are really really important, right. So this is where the 3MF Consortium Foundation has been very helpful to us, and so what we’ve done is at attive flow we’ve created and included that volumetric extension within 3MF. And so what it means is that our optimized parameters volumetrically within that component, together with the geometry being handed over, in this case to aod desk product, for both a discrete and a gradient transition to these parameters.

8:06 And so you don’t have a messing around with different meshes, with different manual allocation of where to place different parameters, we’re going to take a lot of trou and error, right. I know of companies that are spending weeks manually cutting up different sections of a component to apply faster parameters in certain areas, slower parameters than others, mixing around layers heights manually, that’s very retirement expensive to do. We cut all of that out and you get prediction how the parts actually going to work.

8:32 So just very very quickly, validation, right, we’re changing the micr structure in a component, how do we know that those different components are going to perform upfront, right, and that’s what our solvers do, that’s what our unique algorithms are able to do. And so here what we’re doing is, you’re seeing on the on the left here a buildplate of an existing doe, we’ we’ve got different samples with different parameters that we’re printing, we’re getting this process window map, and we’ve done a lot of work around doing this cheaply and efficiently, so this is about 20K of work that we did in about a month and a half of active time.

We then down select those those parameters, and then we’re now getting the stress strain curves from this break point, the performance material properties of those different parameters, that’s an input into our model. We then optimize based in this case on a notched tensile sample, and you can see here the real Digital Image correlation from National Physical Laboratory validating our data, and we’ve got here on the left the standard parameter, this is an EOS M290, and it was a scalo parameter, this is the standard one, dense in the middle we’ve got the fast parameter, this is much much faster to print, but has reduced material properties. And then on the right you got the optimized, this is the composite, effectively the mixture that our software has allocated different presence parameters, a mixture of A and B based on improving that performance.

9:59 And so it’s 60% faster to print, 1.39% marginal stress, as predicted by our software, that was selected, and here are actually the results of that component, you can see the DSC set up at NPL, and, as predicted, 1.39% marginal stress more than the dense, better strained performance, and since is effectively for very very little stress sacrifice you’re getting 60% faster builds, which is incredibly powerful.

10:27 Up front fatigue, right, if you’re changing ltis structures, if you’re changing micro structure, how are your parts going to perform in real world engineering applications, this is a massive challenge industry has facing them, to be able to adopt and qualify AM components. We did that upfront within our solvers and Itive Flow, and so here we did the same story, aluminium EOS M4 400, and you can see the bill plate, we did a whole bunch of fatigue testing, residual stress testing on these different parameters, that’s the input to our software.

11:02 We then, on the left, optimized volumetrically in the blue and the gray, those different process parameters, so different micr structure, different material properties, volumetrically in those pH samples, we manufactured of a whole bunch of them, you can see them being broken there in the fatigue test, high cycle fatigue. And so our results predicted versus experimental were 95% plus accurate for high cycle fatigue, taking into account that micr structure, and that’s really really powerful, because it means you can trust the 3D printed results of our solvers.

11:37 But what happens when things go wrong, right? So here is the optimized components on the top, X14 and X51, you can see a nice tight interface within our software. When we did this testing we sent it to this third party bureau, we sent them our lovely integrated 3MF file, for for the printing, but they ignored it and they just extracted the meshes and they put an offset in place, and so you can see this disgusting cavity in the middle, and it’s about a million dollars of work that we invested into this program, and it was four or five days.

12:23 Anderson, how many days was it before the presentation? Five days. To the UK government, to get our funding back, we were started about that time, pre-acquisition, and our our chief principal research scientist at NPL, our close friend, said, God guys, I’m sorry, you probably ruined, now I’m sorry, the parts weren’t printed successfully, there’s no way your models now can be accurate. I said, hold on, hold on, hold on, let’s take the NDT data. So they sent us this data, they did a polish of exactly what was printed, we took that data back into adid flow software, and it took, was it half a day or something, Anderson, to rerun, how many cases, 24, something like that? 50, even better, even better. So 50 cases we run, and we blind sent our updated predictions back to NPL, and that was the first graph.

13:12 I told you so, that 95% producted accuracy was taking into account the actual defects included in the print. And so what does this tell you? It shows that if you have that data in production, our solvers will be accurately able to predict op up front what’s possible. What else does it tell you? Interrupt really matters, right, and so here you can see the actual interface that we’ve got, we’re doing this with a whole bunch of other machines and vendors, but here’s with Autodesk, we’ve been great, and Alex Oster’s been pioneering that, and you can see here export on the left with the not technology oneclick export, import into Autodesk Net Faab, and you can see including the contour, all parameters allocated nice and cleanly. We’ve also got Tony saying something really nice about us too.

14:05 So what are we doing now? So we have a cool software, it’s getting lots of traction around organizations decreasing cost of AM, getting confidence in AM. Al other engineering, we’re doing a lot of bidirectional integration, so we’re aware that the positioning of our capability is potentially very important to mature industry, and so we’re working with others to put it inside their softwares and their workflow, bunch of people who presented already, and we’re also taking those technologies and bringing them to art as this integrated platform to connect all your systems together, and that’s really looking to partnership together with our customers to mature the market.

14:42 And a lot of our customers come to us and say, we’ve got this fancy physics that’s doing magnetic here, or we’ve got, you know, highly complex radio frequency, I don’t know how to bridge my engineering requirements with the outcomes that I need, to use AM with confidence, and that’s, so we can integrate, and we integrate those technologies again and again to make our clients happy.

15:03 And for fun N Dimension also has this addtive manufacturing of electronics, and so we’re also working on the prediction of how different conductive and dialectric inks can be included successfully, with all the different evaporation parameters to print electronics, and these are real 3D printed components with different materials, and and and conductivity, you can get ED embedded RF, you can do kind of embedded sensors and flexible electronics. And this flight top here is the design system for the design of different multimaterial ocations in the electronic space that we’re working on, so maybe we can show that next year.

15:39 D cool, so I think that’s it for now, but I think kind of the key takeaway is trust, predicting that AM really works, we can help solve that commercial viability, making sure it doesn’t take you 100 times of failed to get to what you want, and partnership. Let’s work together, thank you.