Modeling Tribological Contacts to Calculate Time to System Failure in Wind Turbine Gearboxes

A modeling approach to wind turbine drivetrain components requires fundamental knowledge in tribology, which is the study of friction, lubrication, and surface wear between contacting and sliding components. In this talk, a background on interfacial tribology in gears and bearings is discussed, followed by a discussion on Sentient Science’s approach to modeling tribological systems. Phenomena such as interfacial lubrication, surface contact, and surface roughness modeling are discussed, followed by a description of Sentient’s technique for combining tribology modeling with material microstructure modeling to predict the expected life of components in wind turbine gearboxes.

Recording Transcription:

Natalie: Today’s presentation will be on Modeling Tribological Contacts in Wind Turbine Gearboxes. For those of you who don’t know me, my name is Natalie Hils. I am the Marketing Programs Manager here at Sentient Science. I will be happy to reach out if you need to reach out to me. My contact information is on the screen here.
I wanted to meet you all in San Diego and welcome you to join us. We will be out there on February 2nd and 3rd, that’s next Monday and Tuesday, for the AWEA Wind Project O&M Safety Seminar. Our CEO and President, Ward Thomas, will be speaking on Project Lifecycle Critical Financial Points alongside Use of Wind Energy and Upwind Solutions.So please if you are in the area, feel free to join us and send us over a message, we’d love to meet with you. We also have a booth in the registration area. So if you happen to be at that conference, please feel free to stop by. Before we get started, I just wanted to take you through our webinar series. Like I said, we will be continuing with the same series that we did last year going forward in 2015.We will be starting with the prognostic technology approach that we will be going over today and applying it to wind turbine gearboxes. Next month we have our Life Extension Application Webinar Series which is on remanufactured gearboxes and our last month on industrial internet. We’ll be partnering with the Industrial Internet Consortium and presenting on different industrial internet and what it does for companies in transforming them today.We welcome you to all of our sessions going forward in 2015. Again, we are going to continue to record the sessions and make the PowerPoint presentations available for you on our website. If anybody needs any from 2014, I would be happy to send you over the links to those, and I will be sending out invitations to these ones shortly.Again, we love questions throughout our presentation so on the right hand side of the screen there is a go-to-webinar control panel, and there’s a questions box. Please, throughout the presentation, feel free to input your questions. We will allow 10 minutes at the end to ask Dr. Elon Terrell these questions. And also if we do not get through all the questions, I will be happy to follow up with you before the end of the week.

At this point in time I’m going to pass it over to Stephen Steen to take you through some basics about Sentient Science and some different fundamental capabilities to give you kind of a high level of where we are coming from.

Stephen: Thanks, Natalie. I know everyone is eager to hear Dr. Elon Terrell talk a little bit about Tribology and Wind Turbine Gearboxes and why that’s important. Just an introduction to give you a little bit of information about Sentient itself. And so if we can just go to the next slide, just again as an introduction of Sentient, we are a prognostics technology company.

We are really focused on how can we bring technology to you that’s capable of predicting the risk of failure, that sort of thing, so you can better manage your assets and better design your products into the future. The way that we do that is drastically different than the traditional prognostics approach. We are really focused on multi-physics model based prognostics, and Elon is going to go into a little bit more deeper detail into one section of that on tribology as opposed to the more traditional data mining approach which would be trending and trying to understand what a failure looks like from the data as it’s already happened and project that into the future. So we’re really focused on true prognostics even to the point where you’d be able to predict it before those assets enter service not at the design stage.

We really do that so that we can extend the life of your asset, your machines. In this case, we’re going to use the example of wind turbines but also taking a look at other measures such as changes in duty cycle, upgrading machines versus downgrading those types of things, to see how that impacts life so you can make the best financial decision.

As we eluded to earlier as part of the industrial internet we do this on what’s called the industrial internet. We are able to connect to these assets out in the field, bring in the operational data and use that to further project into the future and identify individual asset failure. We gather that data, make our predictions to then allow you to do trade-off studies on top of the industrial internet.

This is born out of about 10 years of federal funding and other government funding to put together this. Just recently we were named as one of the top contenders of commercialization of federal funds, just recently here the past week or two. You can kind of see the pedigree as a background from where we come, Department of Energy, Department of Defense, some of the largest operators in the world. Next slide.

Just a high level overview again, we are providing a fundamental capability, a much more holistic approach to the problem. Multi-body dynamics tribology which we’ll talk about today, material science and other real world variability. We can do that at multiple levels. At a DigitalClone Component, we do it the component level. It takes several components to put them together. You have a system to predict what components are most likely to fail, when and why and do some trade-off studies on that more a bit on the engineering side, and we take those models, again, connect those to the live data over the industrial internet. And we can do individual asset site and fleet type prognostics across that. You have to manage your fleet and increase return on investment.

Now, just an example of where this is used today, whether the wind market is within your central remote operations centers, depending on what you call that. So you’d be able to have this up and help dispatch and understand and plan, kind of a gimmicky slide and picture but kind of gets the sense that this would fit well within that type of environment and is today over 5100 wind turbines under DigitalClone Live today using this technology.

These are, again, they’ll be up and running into the command centers and the remote diagnostic centers to help plan and dispatch and improve and do trade-off studies, about improve your fleet operation. Just to put couple slides of some of our partners, we’ll just put them up there. There are quite a bit within different industries, the first being companies’ we test so that we can have kind of the world’s most tested products.

These would be cases where we are testing out new designs, changes in operation, changes in material suppliers, that sort of thing to improve the quality of those designs themselves. On the next slide then we are also showing a little bit of what we call the DigitalClone Live Services that we are talking about which is more of the asset management side.

The assets that are already out in the field but you need to better operate those machines and improve the return on investment on those machines themselves. I want to get into the presentation now with Dr. Elon Terrell. Elon, I’ll pass it over to you to introduce yourself.

Dr. Terrell: All right, thank you. My name is Dr. Elon Terrell. I am a computational tribologist here at Sentient, a relatively new member to the team. I’ve been with Sentient for about nine months now. Before that I was a faculty member at the Columbia University. I still serve as an adjunct faculty member there and my background is in tribology. I have my PhD in mechanical engineering from Carnegie Mellon University. It’s a pleasure to be speaking with you here today.

Let’s see if I can get this next slide going here. Let’s talk about why we model. Why does Sentient prefer to take the modeling approach? Essentially, because physical testing is expensive. It’s time consuming. It takes a long time but if we want to look at trade-offs between different components, the traditional way which would be to make that component, test it in the field, it would take quite a bit of time to do so. It would be a lot easier if we were to make a physics based model of that component and be able to test it in the field of operation, all in virtual space. We have what we call our computational test laboratory where we do computational tests of components in simulated fielded operations.

This is just a nice quote that we like to quote here at Sentient. This is Nikola Tesla who was commenting on Edison’s methods. Of course, both of them were very effective researchers and engineers and scientists. But Tesla, of course, he had his own, I guess, criticism on Edison’s methods in regards to being inefficient. I can’t say that I was around back then, so I can’t really comment on that, but Tesla, he said that Edison’s methods were inefficient because he relied too much on physical tests. And if he did a little bit more calculations, a little more theory, it would have saved him 90% of his labor. It’s a very interesting perspective from Nikola Tesla.

What do we model? We model just about anything. We want to know, first off, whether the model is able to capture the physics. We start from the ground up in terms of fundamental physics. We build up the model from governing equations. We put in the appropriate assumptions, of course, make sure they are valid and we put those into the model. I think at this point, let’s see, we’re going to have a poll question, correct?

Natalie: Yeah, that’s correct, Elon. Thank you. Before we get started, I just wanted to ask the first poll question and see if anybody knows what tribology is. I’m going to go ahead and launch that and give you guys a few minutes to answer.

Dr. Terrell: If anybody snuck a peek at the following slide, then you cheated.

Natalie: All right. It looks like we have a lot of people answered at this time. I’m just going to go ahead and share that result. It looks like just about everybody does know what it is. I’m hoping that Elon will take you through and teach you some new things about what tribology is. So Elon, I’m going to pass it over to you now.

Dr. Terrell: Sounds great. Thank you, Natalie. Okay, great. As many people in the audience know, tribology is the study of the science and engineering of interacting surfaces in relative motion. It’s the study of essentially three items which is friction, lubrication, and surface wear.

Essentially, we dumb everything down in terms of looking at rotating machinery. We can look at it in terms of the sliding components. In any machine that has moving components will have sliding surfaces somewhere, it will be oftentimes many places. Since the focus of this talk is on wind turbines, what I have here on the right image is an outline view of the nacelle which is the main body of the wind turbine, the part that contains the drive train.

The wind turbine in and of itself, we’ve highlighted actually quite a few tribological interfaces, places where we would have sliding surface motions. And that would be the main bearing, the pitch bearing, the gearbox, the generator has its own bearings, the yaw bearing. These are all areas that have sliding surfaces, rolling contacting sliding surfaces, and as a result they would all experience potentially friction and wear. Friction would be loss of efficiency just because we are expending energy to overcome the roles of sliding of the surfaces, and they may experience wear as well which is the degradation, the wear down of the surfaces which might require them to be replaced after a while. So that’s the study of tribology.

Now just giving a little bit of background as to why we want to model. Traditionally, we would be looking at some kind of gear or bearing. In terms of one of the particular wear most that we would see, we could see the buildup of what we call aspolation or pitting crack here as the roller is going around the races. Traveling around a divot may form in the material. These are essentially progressive images from one to 12 that show the progression of that aspolation as the roller continues to travel around that race. It starts maybe small, then it gives a period of propagation of cracks, and then there’s a period of rapidly accelerating growth.

Oftentimes the traditional way to look at this is to investigate postmortem, after the part has already gone through the accelerated growth stage and we start to see a really damaged component. From there, of course, we would have to do some major operations to repair it or just replace it altogether.

Now we would actually want to start from the very beginning. We would want to start from maybe even before the formation of that small pit. Even before that number one image that we would see, we can predict how that pit would start to form, how it would propagate, and keep that accelerated damage from happening because once that starts happening, actually it leads to a lot of other items. It leads to vibrations. It leads to metal chips that would get into the lubricant and those chips would get into other components and that would cause debris denting and those types of things.

We really want to stop it early on. It’s almost like catching cancer earlier on. We want to really isolate it and keep that from happening, keep it from growing. Okay, so kind of go to the next slide here. Now we’re going into more fundamentals on tribology. What we have here is what we call the Stribeck Curve which outlines the different lubrication regimes that would occur based upon the lubricant film thickness.

A lubricant would have its own film which would be within those two sliding surfaces to help separate the surfaces and mitigate friction and wear. Now depending on how thick that lubricant film is, we can have actually different lubrication regimes as a lubricant film is relatively thin. And it’s actually normalized in the X axis with respect to the roughness of the surface. If that lubricant film is thin compared to the roughness of the surface wearing what’s called the boundary lubrication regime and we have contacting asperities and the wear tends to be relatively high, the friction is very high and so the efficiency is relatively low.

As a lubricant film is made thicker, that could be through maybe a more viscous lubricant, maybe through higher rates of rotation, something along those lines, then that lubricant film will start to increase in thickness and we will have a separation of the surfaces. That will keep the asperities from sliding together and we will have reduced the amount of friction and ideally reduced the amount of wear. We would actually want to get to the sweet spot where we have that lubricant film normalized thickness of point one which is called elastohydrodynamic lubrication where the lubricant is just about fully separating the surfaces. We’d have maybe a minimal amount of asperity contact and minimal amount of wear.

All right. The focus on the study of elastohydrodynamic lubrication is quite interesting because it has to do with three interacting phenomena. Essentially, they all interplay together. If you kind of look at the left image there, we see we have a piece of viscosity which is essentially the increase of the density of the lubricant. Actually the density and the viscosity of the lubricant as the lubricant is pressurized. We are bringing the lubricant into a sliding or rolling domain and as it is going in there it becomes pressurized, the viscosity increases, and because of those high pressures, the surfaces of the solids deform elastically and also the lubricant, of course, is pressurized. So it’s three main items that are occurring all together.

It’s a very interesting study so it requires a couple equations for modeling all of them. I’ll go through a little bit of them just to give a little bit of background into them and how we go about our models. That’s the fundamentals of elastohydrodynamic lubrication. Elasto is, of course, the elastic deformation of the surfaces, the hydro deals with the lubricant flow.

Now I’m going to get into a little bit of equations now. There’s going to be maybe about four or five slides of equations, just kind of bear with me. I’m going to talk about a little bit of lubrication fundamentals, and then after that we’ll transition into something a little bit more interesting. So bear with me here.

Those of you who are familiar with tribological modeling and maybe those who are familiar with mechanical engineering, I think you might have seen these equations but I’ll just kind of outline them a little bit. These are what we call the Navier-Stokes Equations. Essentially, these are the governing equations for fluid mechanics. They govern how fluid flows based upon how it’s pressurized and based upon how it’s sheared. Also we can put kind of gravity in there, and we can see how the lubricant is able to flow as a function of gravity.

These are three equations which are all coupled together. These are the equations that we generally use to kind of start off with when we deal with most types of fluid flows. In terms of lubrication, we can actually make quite a few assumptions to take those three equations which you see at the top and reduce them down into the bottom three equations that you see on the bottom right of the slide.

The lubrication assumptions that we typically make, you can see that we assume that gravity and inertial forces are negative. We assume that the lubricant flows laminar as opposed to turbulence. We are not dealing with high speed, vortices, eddies, those types of things because the lubricant is generally pretty viscous. We assume that there’s no slip at the boundaries, that the film thickness of lubricant is relatively thin and that we are dealing with a Newtonian lubricant which means that the stress of the lubricant is proportional to its strain. So bear with me if you’re kind of glossing over, bear with me here.

From those bottom three equations which I showed before, we can do a little bit of mathematical exercises and we can get to what’s called the Reynolds Equation. The Reynolds Equation is an equation which gives the relationship between the pressurization of our lubricant and essentially the conditions in which it’s based. So H is the lubricant film thickness. That’s the thickness across the space of the lubricant from one surface to another. P is the pressurization of the lubricant, how much pressure that we see. You can see that we are really solving for P. We really have to solve for P in order to get an understanding of how the lubricant is becoming pressurized. I’m not going to go into all these different terms there but just to discuss how they are grouped together.

The first two terms are what’s called Poiseuille terms which give the pressurization and the cross flow and the normal flow direction. This second group of you see here, they are the Couette flows which give the contribution of shearing lubricant flow. The third term here is what’s called the normal squeeze and the translational squeeze terms. These are really interesting to think about. It can be related much easier to real life, I think. Take a flat piece of paper and you drop it horizontal onto a surface, you’ll see that that paper takes a little while to settle. That’s because there’s forces coming back on that piece of paper which are squeeze forces. So we take that into account on lubrication. Then also we deal with local expansion of the lubricant. Let’s just say it’s heating up with or something. In real science we have to deal with that. The density change is because of heat.

Now, the Couette terms, we can actually separate into their own set of terms which are wedge terms, stretch terms, and physical wedge terms. And I think time won’t permit me to go through all these details here, but we can separate them out. What we normally deal with in terms of lubrication is the physical wedge terminal. The best analogy that I can kind of think of is almost as if you’re driving in a car and you hold your hand out the window, but your hand is angled a little bit with respect to the flow and you hold your hand up. Your hand’s going to go up, and that’s because you’re wedging fluid underneath your hand.

The same type of thing occurs with lubrication although it’s a little bit of a different type of phenomenon but it’s the same thing. A very similar type of effect wherein we’re squeezing lubricant into a shearing domain, and as a result the lubricant is pressurized and that’s what causes it to separate the two surfaces from one another.

We’re still with equations here, so bear with me once again. We can actually take this Reynolds Equation that’s been normalized so put into non-dimensional form and we couple that together with an equation which governs the lubricant film thickness. This is based upon solid mechanics. It’s a little bit outside of the scope of this talk here so I’m not going to go into detail with it, and we’ll couple those two equations also with the keys of viscosity which is the viscosity changing with pressure. Like I said before, we have an exponential relation between lubricant pressure and its viscosity.

We are working our way through the equations here, once again bear with me. We take the Reynolds Equation and we can actually take it, and we can put it into a form that makes it easy for us to solve numerically on a computer. What we’ll do is we’ll actually take the domain that you see here which is the sliding surface domain and if you look at it from the top, then we will be able to discretize that in terms of grids, putting it in terms of grid spacing. And then we can convert that Reynolds Equation into what we call a finite difference form which allows us to solve over the grids in order to solve what the pressure distribution is.

All right. Now, just talking a little bit about elastohydrodynamic lubrication, how we go a little bit beyond just EHL by itself. EHL, of course, is elastohydrodynamic lubrication. Ideally, we would have a lubricant film that would be between two surfaces that are ideally smooth. But, of course, we would know and, of course, hopefully I think everybody appreciates the fact that there’s no such thing as an ideally smooth surface.

All surfaces, at least, especially those surfaces that we are dealing with in engineering, are rough. For that reason we have to actually deal with what’s called mixed elastohydrodynamic lubrication or mixed EHL for short and which we are dealing with not only the deflection of the surfaces, the increase of viscosity of the fluid because pressurization and also the pressurization of the lubricant itself. But we also have to deal with the contacting of asperities. The asperities and the lubricant would help to support the applied low that we are putting onto our gears or onto our bearings.

For that reason, I think this might be the last equation slide we have here so we’ll be able to get moving on some other stuff in a minute. We would use what’s called the Greenwood Tripp model or essentially asperity contact model which is statistical in nature where we would assume that the asperities have some kind of statistical distribution and based upon how the two surfaces which are both statistical surfaces are able to get close to one another, we can govern what is the expected contact pressure, and what is the expected low balance that we would see.

I’m sorry, I think this might be the last set of equations here. As a result, we would have forces due to friction. This would be fluid, the lubricant shearing, and also solid which is the contacting asperities. We would deal with friction force. We sum them both up together. That essentially gives us a direct idea as to how much friction that we would see which is directly related, of course, to the efficiency of our components. We would balance out the loads as well by balancing out the pressures of the asperity and also the lubricant.

Now, let’s transition. I have a little bit of an activity here that I would like to have everybody do, that I like to do for my tribology classes. I teach tribology at the university. If everybody can put their hands together, so press your hands together, your palms together, and slide your hands back and forth, rub your hands back and forth. I’m going to give you about 10 seconds to do that. Put a lot of force there when you’re sliding back and forth. You do that what you’ll start to feel is some friction that builds up into your hand, I’m sure, right? That’s the expected thing.

What we would have is an amount of friction that’s built up in our hands. The same thing happens in our materials, in our gears and bearings as well especially if they are not lubricated very well. For that reason we actually have to introduce a thermal effect into our EHL model. That would be dealing with not only the pressurization, the lubricant peels of viscosity, and the deformed surfaces but also the temperature rise due to frictional heating. The scope of that is a little bit beyond this talk as well, but we do deal with that.

Let’s transition into another aspect of this talk here. Where do we get our inputs for our model? That’s the question here. What we want to do is we want to build up a physics based model which uses as much of the ground truth of the components as we can. What we would do is we would take the surfaces of the materials and we would measure them. We actually have an optical profilometer that we measure, and I’ll show it on another slide. But we measure the roughness of the surfaces.

One thing that we have established before is that all surfaces have roughness. It’s not straightforward to characterize roughness. We actually have to characterize it in a few specialized ways. What I have here on this particular image set here is actually surfaces that actually have the same roughness value, they actually have the same average roughness, that they look very different from one another in nature because they have different shapes to the roughness. We actually have to characterize the shape.

For that particular purpose, we’ll use what’s called skewness. This is actually three different surfaces, three different roughness profiles that have been measured. One which has a positive skew, R skew which is greater than zero. One of these which has a zero skew and another one which has a negative skew. You can kind of see the difference between the two here. The negative skew surface tends to have most of the surface is above the mean line which is this dotted line here and what we have here are kind of pockets of divots here.

Actually, this surface has been shown to be better for retaining lubricant, for instance. We would characterize the surface not to go into all these different equations here but through four different parameters, and there is a fifth one here on another slide which is another one. But the mean on the surface, the root mean squared roughness and also the skewness which I showed before and what’s called the Kurtosis which gives an idea as to the sharpness of the surface.

In addition, we would also characterize the surface using with what’s called auto correlation function. This gives an idea as to how the shape of the surface changes as we go further and further away from a particular point. How homogeneous, so to speak, is the surface.

Now we would characterize all this using optical profilometry. This is the non-contact profilometer that we have in our laboratory. This is a penny here that we could image. We could image it down to a very small level. This is a small close up of Abraham Lincoln inside of our penny here. We can do a 3D scan of that particular aspect, and we see Honest Abe coming out in full glory, and we even added like a nice hat and beard to him just to, of course, make light of a few things. You can kind of see what we can do here. This is non-contact scans of the surface.

Now, we would actually scan our bearings and/or gears in a few different locations to get a good idea as to how the surface looks and how it changes as we go across the material here. These are some surface scans here shown. This was for the inner race and the outer race for our bearings here. What we would do is we would take that raw data and we would generate all the statistics that we would need to generate; the roughness, the skewness, Kurtosis, so on and so forth. And we would make our own surface. It would be a random surface which is actually statistically identical to the surface which we measured.

Here’s some profiles here of surfaces that we measured versus surfaces that we generated. The top row shows all surfaces that we measured. This is for a gear tooth, the addendum, and the pitch, different locations along the surface of the gear. The bottom set of surfaces are statistically generated surfaces that we generated, the random surfaces, and they are actually statistically identical to that which we have measured.

We would take the surface profiles along with information on the lubricant with the mixed EHL formulation, and we would create a mixed EHL model of the drive train components. What you see here on the bottom is pressure profiles of the interface. This is with the asperities. Asperities are roughness elements, for those of you who don’t know, for a progressively smoother surface. For a relatively rough surface, we have these high spikes in pressure and, of course, as our surface becomes smoother and smoother, we have relatively less amount of pressure spikes. We can predict that based upon the surface profiles which we measure. We would get these profiles perhaps from the manufacturer, from section components, and we would put them into our profilometer.

Here are the result of some of our models here. This is time dependent operating conditions. This is for a gear tooth. What you see at the top, these top four images are essentially as the gear tooth is moving into and out of mesh what we would see as the load, the curvature, and the side to roll ratio, the different microscopic parameters that we need in order to do more of our elastohydrodynamic lubrication modeling. Essentially what you see here, these are the two surfaces. This is the top surface, the bottom surface of our gears which we generated using the reports that I described before generating our random surface but it’s statistically identical to the surface that we measured. What you see here are the film thicknesses as a result.

Just to reiterate this point, this is our deterministic EHL model for two surfaces. This is the ground finish surface and the super finish surface. We can see with time, the asperity contacts and also the result in contact pressures. As a result, what we’ll have here is the resulting coefficient of friction. Once again, this is tied directly into the efficiency of our gear chain.

As the gear teeth are moving into zero and out of mesh, so this is a long particular mesh cycle, this is for ground gear tooth. We actually have a coefficient of friction variation. You can see the numbers there, and we can compare that, say, if a supplier is looking to compare super finish versus regular finish. This is actually ground finish, and this is super finish surfaces, so very smooth surfaces. We would actually have an order of magnitude decrease in our coefficient of friction which we can predict.

We can also predict fretting wear. This is vibration based wear. We can see at the top image on the left these are contact pressures and also contact tractions over time, over vibration cycles for fretting. It’s very important in the electrical contact industry, by the way. We would also have a wear profile as a result of fretting as well. From this we can come up with what we call a fretting map which gives an idea as to the amount of failure and the type of failure that we see based upon the forces, the traction force and the normal force. We would generate that. It correlates well with common findings of fretting maps rather than found from experiment which correlate displacement amplitude and normal force with the type of fretting wear that we would see.

Now, I’m going a little bit away from tribology because we need to see how we would put this into the big scheme of things. We would take our tribological models, and we would put them together with our microstructure model. These are some section gears that we’ve cut up and we’ve imaged. We’ve also looked at the microstructure using X-ray diffraction and SEM. We’ll put that together and from that we will create a representative volume element so it’s our own generated microstructure.

This is within the material which has the same type of microstructure as that which we measured although we randomly generated it. But it’s statistically the same as that which we measured. We would take the parameters that we have measured from our tribology or predicted from our tribological simulation. We would couple that together into our microstructure model. This is what we would see here.

At the top here what you would see is, say, a roller which is going past a surface and has different asperities which are in contact. The surface is down here, and you can see the stresses in the surface. Using that stress distribution with that contact traction, we would be able to find out what we need to find out which is what is happening in terms of wear and fatigue.

This is a result of simulation showing four different cases that we simulated using traction and our microstructure model showing the initiation and propagation of fatigue cracks in the material and also essentially this would create what we call a spall or a chunk of removed surface from the material.

What’s interesting about this is that there’s a couple of different things. First off, these patterns of cracks are very similar to that which we see in the fields or in the laboratory as well under similar types of conditions. This is all created using computational modeling. That’s the first thing. The second thing is that they simulate different types of failure. We do not necessarily need to prescribe a particular failure mode. The model already gives us the failure mode, whether that be fretting, whether that be thinning fatigue, micro pitting, spallation [SP]. It will tell us based upon the traction conditions and the microstructure conditions.

From there what we would do is we would determine the time of mechanical failure. How long does it take in order for the cracks to propagate and for us to see if there’s fatigue spall, for instance? We would calculate this and we would put in terms of something which is very understandable in the industry which is, say, like a life distribution. Putting everything together we would take the input loads from, say, a wind turbine. We would use multi-body dynamics in order to understand what’s going on a particular bearing or gear. That takes the loads from the turbine. and it’s able to isolate the individual components. and we see that there’s a certain torque on each bearing that we can find out, on each gear that we can find out. We would do a component level analysis from there.

That component level analysis would involve the use of finite element analysis, FEA, in order to predict the stress distribution on the surface. From there we can determine what we call hot spots. What are the areas of high stress on the surface? Those surfaces, those areas will be isolated for further analysis. So go forward here.

Now let me just bring forth an example just to summarize this conversation here. Now we were approached in the past by a component OEM. They were weighing out three different bearings for analysis. They wanted to see if they swapped out one bearing versus another bearing and it’s in the same location in the same operating conditions in that particular device. They wanted to see how each bearing would be able to perform in a fielded operation. We took each bearing and we measured the geometry of the bearing. Each one of them, I believe, were tapered roller bearings and we measured them. They had different internal geometries but external geometry were the same of each one of them. They were designed for swap out.

We measured the internal geometry, outer race, inner race of the rollers, and from there we were able to calculate the contact pressure distribution amongst the rollers. This is essentially what I alluded to before, and we were able to determine the hot spots. What are the rollers that see the highest pressures? Those will be used, as I mentioned, for further analysis because those will be the ones that will be most susceptible to damage and early wear.

We also did characterization of the surfaces of each bearing. What you see here are three rows of images showing bearing A, bearing B, bearing C. And you can see that the microstructures of each bearing were considerably different. One thing that I do have to note is that typical ISO Standards, not to disparage them, but just to point out the difference in our approach versus classic approaches. There were not taken into account differences in microstructure nor were they taken into account difference in lubricant, by the way, but we would take that into account.

Looking at the different microstructures and, of course, each one of these bearings, although they are designed for the same role they have different material qualities. In particular, there’s an inclusion here, there’s retained austenite in bearing C and so that, of course, goes into our model to look at microstructure analysis. We had taken each one of those bearings, the microstructure of them, and created our own simulated microstructure. Like I said, that it was the same as the microstructure we measured from the bearings. We would take them and we would perform our simulations, and we would be able to determine what is the failure mode and when does it occur.

This is essentially bearing C where it shows the formation of a spall which is a divot in the surface. We can see the pit size, and we can simulate the crack propagation and so on and so forth. As a result, what we have here is a live comparison between our two bearings, between bearing A and bearing C. This is all in computational space. We can see how the changes in microstructure and internal geometry affects the resulted expected life of the bearing. This is performed over multiple simulations. We actually have, I believe, maybe 30 or maybe 40 simulations which we’ll run on a particular component. For each one, we will generate a new random geometry based upon the geometry that we’ve measured.

Each one of them will have different representative surface base upon the surface that we measured. We were able to get a statistical distribution as to the expected life of the bearing versus the number of cycles. I believe that might be, in terms of the talk on tribology and just to kind of summarize here …

Natalie: Not to cut you off but before you go through that summary, can I just ask the audience a quick and final poll question?

Dr. Terrell: Mm-hmm.

Natalie: After that presentation, I think this is pretty relevant. We want to launch this and ask how important is it to have this type of analysis done? Very important for you, important, not important, or you need some more information on this. I’ll give you guys a few seconds. I know we are getting to the 2:00 timeline here, and we want to be mindful of everybody’s time. So thank you.

All right. At this time I’m going to go ahead and close the poll. It looks like everybody had a chance to answer. The majority of you either think it’s very important or important, and there’s a few there that need some more information. As always, we’d be happy to provide you with that more information maybe giving you a clear picture on that question. I’m going to pass it now over to Elon. You can take it through the summary slides, and then we just have time for a few questions. Unfortunately, we are reaching that 2:00 timeline.

Dr. Terrell: Okay. Just to quickly summarize, we first discussed why modeling has been official, why it can be potentially helpful. We discussed our approach to EHL and mixed EHL for gears and bearings, and then also how that relates to looking at the big picture as a whole of modeling rolling contact fatigue. I believe that at this point we can open up the floor for questions from our audience.

Natalie: Absolutely. Again, Elon, thank you and sorry to cut you off there before, but I do want to make some time. We have a lot of good questions that are coming in. And again, if I don’t get through all of them today, I would be more than happy to follow up with you before the end of the week. And I’ll reach out to Elon and make sure those questions are answered.

Here is our information on the slide here, and Elon and I actually got to go up tower. It was just about a month ago now so I thought I’d share this picture with you guys. It was quite an experience.

For the first question, Elon, I’m getting several questions from our audience about what type of applications our software works on. If you could maybe just take them through a high level of the different applications and different platforms that we’ve applied the software to, that would be great.

Dr. Terrell: Okay. We’ve had quite a bit of experience both within the wind sector and outside of the wind sector. I think Stephen Steen, he highlighted the various relationships we’ve been able to establish with both operators and OEMs of various components. Outside of winds, I’ll just kind of mention just a few. We’ve had active relationships with Boeing, Sikorsky, Bell, and there’s quite a few others. We have a list that goes on and on.

Within each one of those components or relationships, we will be mostly focusing on gear and bearing modeling although now we are branching out into some other items as well. We can simulate quite a few different failure modes, both within gears and bearings citing components in general. Yes, thank you.

Natalie: Wonderful, Elon. Thank you. The next question coming from our audience, again, I’m getting a lot so I will follow up following this session. But this question is where does the data come from that you use in these models?

Dr. Terrell: That’s a very good question. Thank you. We would establish relationship as I mentioned with either the OEM or the operator. Oftentimes the road that we’ll take will be our obtaining a section piece of the component in question from either the owner or the operator. They’ll cut up something or maybe we’ll cut it up, and we will take those sections and we will do surface characterization, microstructure characterization, all the different things that we’ll need to do. And in that regards we will get information in regards to the surface finish and the green statistics and those types of things.

We’ll also get information about the lubricant that’s typically used. The geometry is perhaps either from drawings or from the OEM or the operator in rehearsal. Put everything together and we can create our model from there, from that.

Natalie: Great, Elon. Thank you. The next one is what type of test validations have been done to verify that these models are accurate? If you could just take them through, I know we’ve done several validations and I think it might be great to hear those.

Dr. Terrell: I was really hoping that someone would ask this. Actually, let me fast forward to a separate slide. Now I’ll briefly touch on this. This is a comparison of our simulations with NASA field data of a spur gear. Now the NASA data is shown, I believe, in red; our data is shown in blue. The NASA data was generated over the course of 30 years, 30 years of gear test, whereas our results which we, if I might say, I believe it matches pretty closely with the experimental data took about two days of simulations in order to run.

You can see that the slope of the [inaudible 00:51:03] curve as well as the [inaudible 00:51:08] life of our data matches with the experimental data pretty well at a fraction of the time and at a fraction of the cost. Not only that but we can also predict the expected scatter in the data as you can see. As you can see, the NASA data has its own type of scatter because of the variability amongst the microstructure and the variability amongst the lubrication conditions even of the same type of bearing, even of the same model of bearing. We can do the same type of analysis in order to get that same scatter. Hopefully, that answers that particular question. Let me go back here.

Natalie: Elon, thank you. And then I see that person who entered that question also asked if this was shared in a document at all for review. I do have many case studies on some of those validations. They are available on our website. I’d also be happy to send that over to you in a direct link. I’ll get that over to you following this webinar.

But that just about wraps up our time. Again, I did have some more questions on there. Unfortunately I don’t want to keep people over. I know many of you have hard stops. I will get those questions answered and to you by the end of this week. Again, thank you for coming and hopefully we see you next month. Our next session, as I said earlier, is on Life Extension Solutions, and it’s titled Repairing Replacement Components.

That is on February 18th and it’s again from 1:00 to 2:00 p.m. If you are not able to make it, we’ll send out a link and the recording as well. We are hoping that you guys can join us and thank you all for coming today. Thanks, Elon, for the presentation. It was wonderful.

View Presentation Slides

Presenter

Presenter Elon Terrell, PhD
Computational Tribologist

Dr. Terrell is responsible for the development of useful life prediction and optimized system control for DigitalClone System. He holds a Ph.D. in Mechanical Engineering from Carnegie Mellon University, with emphasis on numerical prediction of material wear in unsteady, nanofluid-lubricated interfaces. He also serves as an adjunct faculty member in the Mechanical Engineering Department of Columbia University.