CDFAM NYC 2023 · New York · 14–15 June 2023

Predicting Fatigue Life in AM Metal Parts Using DigitalClone

Abstract

Process Module: A macro-level simulation that uses finite element-based analysis to simulate the layer-by-layer build process, considering inherent strain methodology and the material’s transition from solid powder to liquid metal pool to final solid part.

Microstructure Module: An in-depth exploration of the microstructure at the grain level, predicting the formation of the microstructure and porosity, and validating the model with experimental data.

Fatigue Module: Utilizing a patented fatigue engine based on damage mechanics, this module predicts crack initiation and life prediction without the need for an initial crack, providing a more accurate and efficient analysis of high-cycle fatigue.

Validation and Calibration: Jalalahmadi discusses extensive validation through government-funded programs and commercial projects, mainly in the aerospace industry, and the calibration process for specific combinations of materials and machines.

Efficiency and Cost-Effectiveness: Emphasizing the efficiency and cost-saving aspects of computational design and testing, the presentation highlights the potential to explore multiple scenarios computationally before experimental validation.

Transcript

From YouTube’s automatic captions, lightly cleaned; expect some errors. Each timestamp opens the video at that moment.

Read the full transcript · 3,471 words

Foreign, good afternoon, I hope you’re not tired yet, and I’ll try to give something that is, it’s a little bit different than what we’ve seen in the past couple of days. Past couple of days we’ve seen mainly let’s make a better design, and specifically in terms of the geometry, and, you know, geometrical topology, of course, you know, better feed, better materials, but here I’m going to show a tool that we’ve developed in Sentient that is focusing more on the performance of the part. Once, of course, you know, we made a good part that has good tolerances, how about its performance while it is in the field, it is in the service, how much is gonna last, you know, how long it’s gonna last, is it gonna work for like two days, one year, ten year.

0:50 And with that I want to emphasize that again, as you can see here we are working only on poker with fusion, and again as we are talking about mechanical fatigue we are mainly working on metallic alloys, not on plastic or, you know, ceramic applications. Before doing that, Sentient Science is a small company, I don’t have an introduction, a slide for that one, it’s a small company, we are about 40 people now, my name also, I don’t think it’s up there, actually, sorry for that, well I actually did it, it’s on there, Lamadi. I’m currently the VP of ID manufacturing for the company, and I’ve been with the company for 12 years now.

1:34 So again, we are a software company for mechanical engineering applications, the the software solutions that they are providing, we are branding them under name of digital clone, which is representation of, you know, we are making a digital clone of the market structure of the part, and basically how it’s going to behave under the loading, and, you know, how it’s going to fail. We have a few software solutions in that family, today I’m going to talk about the digital clock for item manufacturing, which again is designed for AM part made by powerbed, and again for metallic alloys, it’s ICME solution, of course. So for you, any of you that are familiar with the with the ISME solutions that are working on, you know, mechanical performances, you should be very familiar with this.

2:21 You know, modules that are included in our software, they are well accepted, you know, by anyone that’s working in that, you know, industry, that first of all we are talking about the process module, then we have the more construction module, and a final fatigue module. I have a couple of slides on each of them, of course, in the following coming slides, but in short, as you can see, it’s a multi-scale model, starting from the full level, in the first one, process module, is basically working with the entire geometry of the part at the macro level and simulating, you know, the layer by layer build process in the in the chamber. Second one, now focusing on a much smaller dimension, another, you know, scale, focusing at the green level, and see how the market structure is basically forming.

3:09 And then finally using those information learned from the two previous two ones, combining it with the load that is applied, in the withdrawal properties to predict fatigue life for the part that is, you know, under the investigation, and then of course considering all of the randomness and stochastic features that are involved in that process. The first one, process module, again it is at the macro level, it is working with the with the entire geometry of the part, it has finite element based, and it is working, it is using inherent strain methodology to basically simulate the layer by layer built of the of the part, and simulating the the change in the set of the material going from the solid powder to the liquid metal pool, and then, of course, you know, back to the final solid part through the solidification.

3:59 It needs, of course, similar to any other computation model, it needs a quite significant input parameters that can be divided to a few category, of course we need input parameter as the geometry, like STL file, then we need the set of process parameters that are set on the machine, like, you know, laser power, laser speed, and also we need metro properties, and again the list of the metro property that we are using are including both macro level properties, and which are, you know, very typical, like a belt pool or, you know, elastic modulus that you’re all familiar with them, and then also we are using some micro level, green level, natural properties that might not be very familiar for a regular engineer, but then of course the help, or users, you know, to identify them, and of course, you know, input them and use them.

4:47 And then once we do that simulation, of course, what is the output, the output, of course, would be the part that comes out of the machine, out of the bell chamber, what is the visual stress that is left in the part, what is the deflection that is left in the part, and of course using those to to make a better design, right, what we have been talking in the past couple of days, that of course how to make the part to make sure that, you know, we are in the acceptable tolerances in terms of the geometry of the part, and then of course not having too much zero stress in the part, to not breaking it off before even sending it to the field.

5:24 Then the second module, mock structure module, one output of the first module is the thermal history during the best process, so basically we have temperature gradients for each location, each time, during the entire build process, and then of course we can use those information, those thermal information, along with again metal properties to predict what is the market structure at any location of that part. As you can see here, it is, we all know it is impossible to predict the market structure for the entire part, right, because with the current computational powers that we have, that mother is going to run for a few centuries, so hopefully soon, I don’t know, Intel, AMD, come up with some much much better CPUs to be able to do that one.

6:08 But for now, instead of that one, we have to focus on a very small part of the the part, so basically users select that a specific location, that hottest spot, which of course, you know, it is determined, dictated by the geometry, by the loading of the part, you know, where is the spot that I like to focus on, that, front, make my mock structure for there, and for there, and see how it’s going to behave under the loading and how it’s going to fail. Again, we are predicting the market structure in 3D, and then of course we are making 3D, as you can see here in the movies here, you know, along the bell direction, perpendicular to that one.

6:43 Also, what you can see on the left hand side in our market search generator, we are also able to predict the porosity that happened in the market structure, and and actually predict the processary creation for both reasons, not enough energy or too much energy, basically, you know, the porosity that happens because of lack of fusion, or it is happening, you know, begin, you’re applying too much energy because of, you know, boiling and the keyhole effect. Two, of course, the very first question that comes into mind, what is the validation, so here I’m showing a couple of use cases to show validation, some level of validation, basically for that specific module.

7:23 As you can see here, I’m listening for two different materials, increment 718 and aluminum silicon alloy, both of them very well known metallic alloys used in power bet fusion across the world. So the first one here, on the top left corner, I want to show a flexibility of our model to input parameters that are set by the user, so I’m showing the comparison of all prediction versus the experimental ebst for the income 718, and the the people in that literature that did that experimental work, they use the same powder, but they use two different set of parameter on the machine, and as you can see the macro structure significantly changed because of those different set of parameters.

8:08 And in the lower row you can see that basically I did the same thing, that, you know, if I use the same metro property but I changed my process parameters, as you can see the market structure that also we are predicting, it is behaving differently, very similar to what is seen under there, again ebsd. On the on the right top hand side, I want to show you some quantitative values of the validation, so again, as you can see, the market structure that we are predicting for a aluminum silicon alley alloy, as you can see both qualitatively and also quantitatively on that normal distribution, is very similar to what was seen, you know, again under ebsd.

8:50 In the lower spot, I, again, as I mentioned in the previous slide, I’m showing responsiveness of the model on predicting the porosity when I’m changing my process parameters, of course OEM is a machine, OEM as providing a default optimum set of parameters that if you do that one you get the best market structure, but then what happens when we deviate from those for any reason. And again you can see the comparison, or prediction, in the lower right hand side, the predictionable model versus experimental build of those aluminum silicon alloy part, and how close, or in some cases not that close, comparing, you know, to those to those experimental porosities.

9:35 Coming to the fatigue, and fatigue actually is the the core value, our company Sentient has been established about 22 years ago, and mainly, even before, you know, coming to idea manufacturing, it was about market structure and its effect on the fatigue, and that’s where the company has the strongest, you know, technology, there, or fatigue engine is even U.S patented for both, again, 9 AM metallic alloys, and then also extending it to the IIT manufacturing, especially power bed metallic alloys. As you can see here on the left hand side, I’m showing the one of the features of the software, which is providing the statistical analysis through the viable distribution, to be able to investigate the statistical and the randomness of the data, and of course, you know, getting L5, L10 values.

10:28 And then also on the right hand side, a very familiar curve for any mechanical engineer, beautiful old school SN curve, and the biggest value here, of course, is that if you want to create that SM curve experimentally, we are talking about a few months of work, and of course, you know, million dollars of money, you know, to spend on, you know, making those samples and testing the fatigue. Fatigue testing is not cheap, and while computationally, of course, you know, I can create that SM curve in a week, and of course much cheaper than, you know, you have to do it on the test street.

11:06 As I mentioned, you are using, or fatigue engine, we are calling it CLP, it has company life prediction technology, it has again US patented, and why it is unique, in a sense it has, because it is based on damage mechanic, fatigue damage mechanic, I’m not sure how many of you are familiar with that farm, because most of the, excuse me, fatigue software that are out there are either based on extensive empirical data that they do some form of, you know, interpolation and making decision out of those, or they are based on some standard, like ISO standard, with a lot of collecting factors based on your conditions, or even if, if they are computationally, they are based on linear fracture mechanic.

11:53 This is based on damage mechanic, and its biggest advantage compared to fracture mechanic is that you do not need an initial crack to start your simulation and see how that crack is going to basically get to the failure, what which is against what you need in fracture mechanic, right, in fraction mechanic you need initial crack, and then basically you’re simulating how long it’s going to take for that crack to become critical. What is missing in that is that when you assume that initial crack you are overlooking your correct initiation stage, and most of our customers are in a race, airway space, and in aerospace crack initiation stage is all they care about, so you cannot do that when we fracture mechanic, right.

12:43 This one can predict the crack initiation stage, which, when you’re talking about high cycle fatigue, could be all the way up to 90 percent of the entire life of the part, and that’s the value right there. I’m showing a couple of use cases here, first one is on the simplest case that, you know, we can do on fatigue, is just a simple dog bone under axial loading, the simplest case of, you know, fitting testing that you can do. This one was made again by powerbit for stainless steel 17-4, and as you can see, again we did the prediction of the macro structure, and then the fatigue testing, and also physics simulation.

13:27 Excuse me, and what you have to notice on the right hand side image is that the prediction that you’re making, compared to the real data, is pretty wide, and the reason for that one is that basically, perpetually, when you are doing a simulation, we are considering the worst say, first case scenario, to the best case scenario, that basically, in terms of fatigue, but that means, means worst case scenario is when you have a bad flaw, you have a big inclusion, big void, and it happened that that bad flaw is also in a worse location, that it could be right under the load, right.

14:07 And then, the word, in the best case scenario is that either you have a very clean microstructure, or the void that you have is not in a critical location, so we are predicting that range to make sure that they consider all the possible scenarios, and then, of course, as you can see, experimental data, right, are right in the middle of that behavior. However, if you go to a little bit more complex part, like this one, a single gear 2, which is under gear to it bending fatigue, now you can see that even the experimental data start to become vital, because now we have more material, more possibility for the failure, so even experimental data are wider than a simple toggle.

14:50 And again, this one was made from C64, which is the highest ranked aerospace grade steel, and again we did the market structure prediction, we did the fatigue testing, and you can see the comparison of all prediction versus the experimental data. Let’s making it even more complex, this is a high pressure manifold, again from aerospace application, and in this manifold the loading is not mechanical loading, it is through the dynamic pressure, so the pressure inside the manifold, on the texture, on test trick it was going, you know, to some predefined value, and coming back to the zero, and that was the fatigue cycle, in term, in terms of the loading, the the changing pressure inside the manifold.

15:30 And again, same thing, as you can see on the top there, on the right hand side, the experimental ebst and our prediction of the macro structure, and then you can see the SN curve, and that’s another value right there, that because the fatigue testing was expensive, the customer only did it for one specific load level, those red dots, and versus our simulation, that of course we did that load level, and also repeatedly for four more load levels, because again coming back to the original point that I made, critique testing is expensive, it is time consuming, and let’s making it, you know, more efficient, you know, cheaper, wiser, through the, of course, you know, computational design and computational testing.

16:18 In terms of timing, again, running the simulation, similar to any other computational package, depends heavily, you know, on different parameters, you know, how big is your model, how many simulation you want, how complex is the loading, how complex is your market structure model, but to give you some, you know, nominal values, you can see that the first module takes a couple of hours, the second module we are talking about half a day, and then the last module, as we are running it, you know, on HPC servers, we are talking about a few hours, all the way to 24 hours, again, you know, based on the size of your model and the complexity of the model.

16:52 A decent slide kind of summarizes different type of outputs that we are providing through those different modules, again, all the way from deflection, there’s your stress, it coming right out of the machine, the mock structure coming out of the machine, fatigue test result completely generated, of course, and then also, in the bottom there, for if you are a mechanical engineer in the in the audience here, this is our processity map, that basically user can set a couple of input parameters and ask this simulation to provide a processity map for that, for them to show where they choose their process parameter will be the most beneficiary to produce the least porosity.

17:31 We are, this is our current metro library, whenever we are working with the new material or new machine, of course we are going through the calibration, that calibration has both experimental and computational part, you are doing it, you know, with collaboration with our customer, and the way that we are doing calibration, we are calibrating material slash machine, as you can see here, those x dot x marks are for a combination of material and machine, that means that right now, as you can see, for example, we have done the calibration of titanf64 on your Renaissance machine, but if you want to do same titanium 64 on es290, we have two decalibration for that specific machine.

18:13 So in summary, I, in the past few minutes, I tried to introduce you to one of our software solutions called Dragon for ID manufacturing, which is designed for mechanical application, focusing especially on the market structure and mechanical performance, especially fatigue, of the metallic components made by powerbit, and producing different type of outputs out of which those modules, and again it has been validated through extensive spr government funded programs, and also quite a few commercial projects that we have had, mainly in aerospace industry.

18:54 And the value again, emphasizing on that one, that yeah, doing the qualification, doing the mechanical testing, it has cumbersome, it is time consuming, it has expensive, let’s make that process, you know, more efficient. To add the end of it, instead of spending one million dollar on a couple of scenarios, of course, let’s spend that one million dollar, 100 different scenarios, and let’s do the first, first few iterations computationally to see how many of those hundreds even worth, you know, do the experimental exploration, and, you know, validation, and of course, you know, again make the most out of your money.

19:35 I’m gonna finish with our brag slide, I’m not sure how many of you are familiar, again, with the NIST, National Institute of Standard and Technology, here in U.S, they are holding a competition for AM excellence every other year. We attended their competition last year, they had 26 topics, we attended in six of those topics, based on, you know, the scope of the work that we do, and out of those six topics that we attended, as you can see here, we were awarded with four first places, on prediction of the market structure and residual stress and deflection behavior of England 718, and in class 625.

20:16 So by that, if you see something that you know might be interested to your work, I would love to talk to you more, please come and find me later, you know, in the break, in your desk. Thank you very much.

More from CDFAM NYC 2023

Design Computation Human

Design Computation Human

Onur Yüce Gün · New Balance

Mud y Robots: Keynote Presentation

Mud y Robots: Keynote Presentation

Ronald Rael · UC Berkeley / Emerging Objects

Roadmap to Replicators

Roadmap to Replicators

Neil Gershenfeld · MIT Center for Bits and Atoms

Scaling & Standardizing Patient Specific Design

Scaling & Standardizing Patient Specific Design

Adam Wentworth · Mayo Clinic

Computational Design Ops

Computational Design Ops

Ronnie Parsons · Mode Lab

Overcoming the Design Dilemma

Overcoming the Design Dilemma

Ryan Carter · Johns Hopkins APL

Generative Design & Digital Manufacturing at NASA Goddard

Generative Design & Digital Manufacturing at NASA Goddard

Ryan McClelland · NASA Goddard Space Flight Center

Digital Materials – Geometry and Data Combined

Digital Materials – Geometry and Data Combined

Marek Moffett · General Lattice

Ten Years Teaching Additive

Ten Years Teaching Additive

John Hart · MIT

10 Years of Applying DFAM

10 Years of Applying DFAM

Tim Simpson · Penn State University

MfAM Walks so DfAM Can Run!

MfAM Walks so DfAM Can Run!

Chelsea Cummings · The Barnes Global Advisors

Informatics and AI for Materials Development

Informatics and AI for Materials Development

Wes John Alder · Citrine Informatics

Designing Material Properties with Advanced Toolpath Generation

Designing Material Properties with Advanced Toolpath Generation

Alex Roschli · Oak Ridge National Laboratory

Design Automation Using Machine Learning & Computational Design

Design Automation Using Machine Learning & Computational Design

Rhushik Matroja · Cognitive Design Systems

Learning to Generate Shapes

Learning to Generate Shapes

Karl Willis · Autodesk Research

Design for Artificial Intelligence

Design for Artificial Intelligence

Chris McComb · Carnegie Mellon University

Register for Updates and Discounts on CDFAM events.