Natalie:Good afternoon everybody and thank you all for joining us today. I see a lot of familiar faces on the webinar attendee list, so I wanted to welcome you all back and welcome those who are new to our sessions. Today we are going to be talking about how we use our predictive modeling digital clone for bearing supplier decisions. There is lots of decisions out there and different bearing suppliers, and we want to make that decision easier and we’re going to take you through our steps for that.For those of you who don’t know me, my name is Natalie Hils, and hello to everybody who does know me. I am the Marketing Program Manager here and I run all of the webinar series. Here is my contact information on the screen, so please feel free to reach me if you need anything at all. I have plenty of resources that I’d be happy to pass over to you.
Before we get started, I just want to take you through our webinar series. We do host one webinar per month, three per quarter, and we always go in this order. We start with a deep dive about our technology, then the next month we take it a step further and we demonstrate to audience how that technology is applied in different industries. Today we will be talking about how we apply it to bearing decision making. Lastly we would like to tie it all together with the industrial internet. So how are you able to use our prognostic models and tie it to the industrial internet to make executive decisions, supply chain decisions and budgeting decisions in the future?
All of these webinars are hosted on our website. If you do miss any of them, they are under the video recording tab on the right hand side of our website, and the Power Points are available as well as the recordings for you to follow along. If there is a certain topic that you would like me to send over, I would be happy to. Please reach out to me at any time. Just before we get started, a few housekeeping things. I want to make sure that everybody is able to input some questions if you do have them. There is a go-to webinar control panel on the right hand side of your screen. Please, we encourage a two way conversation. We’ll slot out some time at the end of the presentation for that. Before we get started I just want to do a warm welcome to Dr. Elon Terrell. He is our Computational Tribologist here, and really the brains behind it all. Elon, please take over whenever you’re ready.
Elon: Thank you Natalie and of course I have to defer to the rest of our technical staff of course, in additional to all of our other staff members at Sentient who collectively make the brain trust of Sentient and make us go. Our webinar today that we want to focus in on, is involving decision making for bearing suppliers. There is a big challenge that bearing OEMs have to make, or at least device OEMs, in regards to bearing selection. Essentially, bearing selection is something which the supplier . . . Device OEMs have to take into account a number of different things. Not only cost, but also the design goals of the particular device.
The selection criteria you can see here, essentially there are not only dimensional constraints, of course we have to keep the bearing within the dimensions of the particular device that we’re looking at. Also we are talking about external dimensions. We’re looking at the inner bore, the outer bore, and the width of the bearing. But then also there’s other considerations that must be made. Tolerance of the bearing, the rigidity of the bearing, and also the capacity of the bearing to handle the load that is being specified. So those different considerations must be taken into account, and of course they have to be weighed with the consideration of cost in order to keep the overhead cost low on the particular device.
Now, in regards to bearing selection, what we would normally see, and those who work with bearing procurement would generally be able to attest, is that bearing models, they tend to have very standard designators that are universal to most major manufacturers. This is just an example here of a cylindrical roller bearing which you can kind of see, the designator corresponds to the type of bearing. So a cylindrical roller bearing is a letter N, for instance. Whether or not there’s lips on the outer ring, that’s another designator. Whether there’s a certain width that we’re talking about, that is another designator, and the diameter and the bore size, those are other designators.
So essentially we are talking about designators for the shape and size of the bearing, and if we’re talking about that, it’s generally universal to all major manufacturers. The question would be, “Well, what is the difference between different suppliers of the same model of bearing?” There’s a number of other things which we’ll touch upon.
This raises that question. Essentially since the bearing model is standardized across different manufacturers, how does a device OEM go about choosing a particular supplier for a particular application? And how do we also take into account the operating conditions under which the device is being subjected, in regards to choosing not only the model of the bearing but also the supplier of the bearing? Now, let’s back up a little bit and talk about the standard classic approach towards life ratings. Let me just refer to the classic approach to life ratings, which is to define a static load rating which is called C, which is essentially defined as the load in which a bearing can carry for a million revolutions, with a 10% probability of fatigue failure.
If we wanted to scale this up with different loads and look at how the bearing life would be estimated under bearing loads, then we would take that static load rating and we would divide it by that load, and then raise that divisor by a certain power or P, which is three per ball bearing, so ten over three for roller bearings. And we’d end up getting the bearing life which is called the L10 life, 10% probability of fatigue failure of that life, in terms of millions of revolutions. Now this is the classic way to estimate life, and pretty much if one were to take a look at any bearing in a particular catalog of a certain supplier, then they would see that static load rating and they would be able to define what that life is.
There are different things that can be used to adjust this life rating based upon the materials that are seen, and then also the upgrading conditions that the bearing is subjected to. Just to look at this classically, this classic load rating can be adjusted for certain metallurgical differences, say for vacuum-degassed steels, then we would have an adjustment there. For lubrication conditions we would also make a certain adjustment. In that way we would also be able to tune the life rating based upon these different factors here. Essentially these are multiplication factors. The L10 life would be multiplied by A-sub-M or A-sub-L, or any other type of these adjustments in order to get modified life rating.
In looking at this approach, one can see that it’s pretty much empirical in nature, and to that end it does have quite a few limitations. So not only it being empirical, it’s something that can vary from one application to another. There’s different things that can vary that can cause the life of the bearing to change, but then also, the life rating is limited by the fact that it does not account for other types of operating situations such as the bearing temperature, misalignment. And then, one also can find out that it does not account for internal features, such as changes in microgeometry, like the crowning of the rollers, changes in internal clearances, and then also whether there’s any surface treatment that’s put into place in the rollers and the races such as surface finish and surface coatings. So different things like that are not taken into account in the classic approach.
Going a little bit further, one would find that the classic approach also does not take into account other factors such as the material quality. The grain size distribution inside the material, the presence of defects and also residual stress distribution. So this is lack of accounting for say, [inaudible 00:12:14] or surface hardening or something along those lines.
All right, so this leads me to Sentient’s approach for bearing supplier selection. Essentially what we will do is we will account for a number of different things. Not only we would do global loads analysis, but we will also do a material characterization, lubrication analysis and microstructure base failure. Now, in taking a look at all this, let’s maybe delve into a certain case study that we’ve looked at recently. This is involving the Clipper Liberty 2.5 megawatt wind turbine. A very high powered wind turbine.
Okay, so going into the different steps that we would take for bearing procurement, we would first off perform a system level loads analysis. This is taking into the account the varying loads that would be seen in different components. What you see on the left side is a gear box, a diagram of a gear box for a Clipper Liberty. Of course on the left side, this is the input coming in the blades, the rotor, and is divided up into four output shafts and we have represented that in a powerful diagram.
Mathematically we have performed multi body analysis in order to get an understanding of the loads that are seen on each of the individual components, the gears and the bearings. What we would see is eventually that the torques that are seen on the different bearings. We wanted to investigate in a particular field of operations, say we have this gear box in the field. If we swapped out bearings on a certain shaft, this is the intermediate shaft, what would be the expected life of each bearing, the exact same bearing with different suppliers? How would their life be different in the field of operation?
These two bearings are seen here, what you’ll see here are bearings from two different suppliers. We have Supplier A and Supplier B. They have the exact same external form factors, so they can be swapped out with one another. And you can see that they have very similar, actually the exact same exterior dimensions. The exact same bore, the same outer diameter, the same width. You can see here that also, if one were to take a look at the static load rating, this is taken directly from the catalog, one would see that the static load rating is very similar between those two bearings.
The load rating C is, for Supplier A, 2040 kilonewtons. For Supplier B is 2110, which is slightly different but not too different. One would be able to think at first glance that the expected life would be very similar between these two bearings. Now, from our life analysis, we can kind of confirm that actually, we just take into account a particular load of 1500 megapascals. We would see that the L10 life is just about three E to the nine cycles for both bearings. This is solely taken from the catalog, just using the information from the catalog.
Now, going into Sentient’s approach, we would perform, first off, internal geometry measurements. Looking at the 3D profiles of the rollers of each one of the bearings, we would actually see that the rollers for Supplier B were slightly larger in diameter that that of Supplier A. And then also, the inner and outer races, they had a higher crowning profile than that of Supplier A. So, there’s more of a roundedness in the races. Geometry differences actually can play a pretty big role in the context just as they’re seen, which would in turn, play a role in differences and the expected life of both of those bearings.
Taking into account the internal geometry, we performed a component level analysis. What you’ll see here is essentially the same simulation but with different components highlighted, shown just for clarity. The left image is the axle bearing assembly, and the middle is essentially that same assembly but with the outer races removed, and the right most image is that same assembly but with the rollers removed as well so that we can see the stress distribution on the inner races. What you can see here is of course, there is quite a bit of contact in the outer lip of both of those races, or inner of them, I should say.
We can take a close up of both of those races, both the inner race and outer race, and one can see that there is some edge contact in both inner and outer races, which should be taken into the account in our approach. And it is taken into account in our approach but for the classic approach, that particular item is not accounted for, actually. Going a step further in regards to the internal geometry characterization, we performed optical profilometry of the rollers and races using our in-house 3D non-contact optical profilometer which you can see on the left side.
This is the image that we like to show in which we imaged a penny, and if we were to take a look actually the sensor we see Abraham Lincoln. We imaged, stayed in 3D, and we can see the profile here. We added a nice top hat and beard just for effect, actually. But you can see that the power of this device in which we can image surfaces in three dimensions, and I get the nice profile. We did the exact same thing for the surfaces on which we were able to characterize the profile of the surface, not only the roughness, but also other parameters, in particular the skewness and the kurtosis. Just as an example of the skewness of the surface, we can see actually three identical surfaces, at least surfaces with the same roughness in terms of that being identical, but they do have different skewness. The three rows of images on the bottom right, we see the surface that has a positively skewed roughness, has a lot of asperity peaks that are above the mean line. We would have a lot of high contact areas, those would be high stress points due to those highest asperities above the surface.
On the middle of those bottom three images we can see an evenly skewed surface. One which has a skewness of zero, in which the asperities are distributed above and below the mean line evenly. And one can see on the bottom right image, that’s a negatively skewed surface in which most of the surface is, most of the asperity peaks are actually negative asperity peaks below the mean line. That’s actually most beneficial for handling pockets of lubricant.
So even thought there are surfaces that may have the same roughness, there are other characteristics of the surface such as the skewness that dictate how it behaves inside of the contact, sliding contact interface. So we take that into account in our model. For our particular bearings, we would actually see that they would have different roughness characteristics internally. For the Supplier A, we actually have a negatively skewed surface and for Supplier B, that’s on the outer race. And for Supplier B, we actually have a slightly negatively skewed surface on the outer race and a positively skewed surface on the inner race. So we take that into account, and there’s also differences in the kurtosis value, and essentially that’s a measure of the jaggedness of the surface, and we can see the differences there.
Now, we would take this into account in our lubrication analysis, and I’ll touch on that in another couple of slides, wherein we would create an equivalent surface based upon the roughness of the surfaces that we measure. That would account for the roughness that we measured. So by making statistically identical surfaces and putting that into our lubrication code, we would be able to predict the expected friction and traction that would be seen in between those rollers and the races.
Now, a level beyond that of the friction and lubrication is that of material characterization, and what you can see here are micrographs of the metals of Supplier A and Supplier B, those particular bearings, we would cut them up and we would measure . . . We would actually image those, the subsurface metals of both of the suppliers. One can see from these images actually that there’s a difference in the material quality between that of Supplier A and that of Supplier B. And that Supplier A actually has a higher amount of retained Austenite, whereas Supplier B is mostly a pristine granular structure. There is differences also in the grain sizes of that of Supplier A and that of Supplier B.
Both of those differences, what caused differences in the expected life? In regards to the material quality, we can expect the presence of defects, essentially retained Austenite, to serve as nucleation points for cracks. And so because whereabout that we would predominantly see the [inaudible 00:23:16] load in these bearings is essentially crack base rolling contact fatigue, then the presence of defects and inclusions is actually quite important.
Putting our surface roughness measurements into play, we would perform a mixed of astrohydrodynamic lubrication simulation of the bearings, considering the contact between the rollers and the races. And we would be able to extract out the contact pressures and the expected amount of traction between those two surfaces. What you see here is a depiction of our model in which we would have our contact pressures as the rollers are traveling across the races, and the bottom left most images, you can see the asperity contacts with respect to the time that would be seen as a result of the contacts in asperities.
The elastohydrodynamic lubrication model, I’ve touched on in a previous webinar, but just to say a couple of things about it and that essentially it is an interplay between four different effects. Essentially we have the pressurization of the lubricant, which happens as a result of the high contact pressures. Because of those pressures, the surfaces of the rollers and of the races, they actually deform elastically. Because the lubricant is being pressurized, there is actually which is called piezoviscosity, which is pressure based increase in the viscosity of the lubricant, so the lubricant viscosity increases exponentially.
It almost behaves, almost like a semisolid, because the pressures are so high. And also, of course, we are interplaying with the asperity contacts. So all those different . . . These different physical phenomenon are occurring simultaneously, and in addition there is also friction based heating of the contact interface which is taken into account as well.
Taking into that traction finding, we are able to consider that as being an input into out material microstructure model, wherein we would take our grain measurements and our residual stress measurements, as well as the measurements of defects, and we would create a statistically identical material microstructure as that which we have measured.
Accounting for the traction that’s traveling over the surface, which is the contact surface which you see on top, we would simulate the initiation and propagation of fatigue cracks. We’d generally start below the surface and they propagate above the surface as you can see on the left most . . . the left bottom image. And as a result they would create a surface pit which you can see on the bottom right image. And at that point in which the surface pit nucleates up to the surface and the material is falled away, then we would consider the bearing to have experienced a mechanical failure and we would consider . . . we would count the number of cycles in which the rollers would have to travel past the races in order to get that failure and we would get the life prediction in that regard.
This is the approach that Sentient takes, accounting for not only the material mark obstruction, but also lubrication and internal geometry, amongst a number of other factors in regards to predicting the expected life of the bearing. Now, this is a result of the simulations that we performed, wherein we would have Supplier A and Supplier B. What you see on this image is essentially, is the number of cycles to failure versus the applied load for that particular bearing. So at that particular load, then what we would see at that particular load we would have a number of cycles to failure at that particular load.
One would see actually from this image is that there is a significant difference in the expected life of the bearings from that of Supplier A and that of Supplier B. Essentially, Supplier B under the same load, Supplier B bearings would actually have a longer life than that of Supplier A. Actually that of almost an order of magnitude improvement. If one were to take a look at a pressure of 1500 megapascals, then we can see that . . . If one were to take a look at the pressure of about 1500 megapascals, one would see that the L10 life is two E to the ninth cycles for Supplier A and 1.75 E to the 10 cycles for Supplier B, so almost all order of magnitude improvement.
One would say, “Okay, these are some interesting simulations, but do they actually correspond to what is going on in real life?” Oops, to answer that, we actually do have some predictions of the expected benefit that we will see, that we’ve seen in the field, and they actually do . . . The findings that we’ve found in the field do correspond to that which we have seen in our predictions.
So just to summarize, bearing models designators are generally standardized and across different manufacturers, and so, in order to make a really good consideration for bearing supplier selection, not only the external form factor should be considered but also variations in the internal geometry and the metalurgy lubrication in operating conditions need to be taken into account. The classic approach does not necessarily take these items into account, at least not in the physics space manner. Those are normally empirical in nature, so there’s a lot of variations, a lot of other factors that must be taken into account to really see how a bearing will operate in the fielded scenario.
Sentient has taken these items into consideration, and has found that even with bearings that have the exact same external form factor, very similar life ratings from the catalog, there’s a noticeable discrepancy between suppliers, and this is generally due . . . This we found is due to differences in internal form factors, differences in internal geometry, metalurgy, and then also there’s some material quality that must be considered for changing the expected life. I guess with that I’ll wrap it up and if anybody has any questions, then certainly feel free to ask.
Natalie: Thank you Elon, and thank you for taking us through that. Please feel free to reach out to us. Our beautiful picture of a winter run that Elon and I were able to climb together last December is displayed here with all of our contact information as well. So please feel free to reach out to us. If you have any questions, I will be sending the recording and the Power Point slides over within the next week, so please look for those from me. I just wanted to invite you to our next webinar. We’ll talking about how easy it is to spend $5 million when doing different operation maintenance programs for your gear boxes, and what Sentient’s approaches and how we’re able to lower that cost a little bit there.
We also look at seeing everybody at the WIA Windpower show in Orlando, Florida, so hope to see a lot of you there. I see a lot of familiar wind faces and names on the attendee list today. Safe travels and we’ll see everybody there.
Thank you all and we’ll see you at next presentation. Thanks Elon.
Elon: Thank you.