Many wind turbine operators that own GE 1.5MW turbines are considering to either uprate (using e.g. GE PowerUp) or derate their GE turbine fleet. Unfortunately, the impact on reliability, especially on the gearbox, through uprating or derating is unknown using standard industry techniques. This results in uncertainties that do not allow the operators to estimate the financial impact of uprating or derating correctly.

Recording Transcription:

Natalie Hils: Thank you all for joining us today for our webinar. It looks like we have a lot of people from all over the world that we’re really excited to talk about this topic. It’s really been a popular one in the industry today. For those of you who don’t know me, my name is Natalie Hils. I’m the marketing manager here in Sentient and I’m in charge of running all the webinars series. Please feel free to reach out to me at any point in time. I’d be happy to put you in contact with those on our team as well as give you resources. They’re all on our websites. Send you, guys, links to anything that is pertaining to you. We have a variety of case studies, white papers and webinar recordings and presentations that I’d be happy to pass over. Just to run through the series to those who are new, we do host three per quarter, one per month. We always also focus on these three topics here. We start of quarter with the prognostic technology approach in which we take a deep dive discussion into our technology and how it actually works.
The second month, we dangle into our life extension application much like you will see in today’s webinars session. When we’re talking about energy and wind, I mean other series we have done–focused on automotive as well as wind and others. We, again, have all those recordings and if you like to see any of them we have missed, please reach out to me or they are all on our websites under the video library tab. You can download the presentations as well as the recordings. We then finally wrap it up with the discussion on the industrial internet. As always, we really enjoy when our audience asks questions throughout the session so please feel free. On the right hand side of your screen, there’s a webinar control panel and input any questions there. At the end of this session, I’ll open it up to Dr. Adrijan Ribaric for a question and answer period. At this point in time, I did want pass it over to Adrijan, if you want to go and head us to the presentation. Thanks, Adrian.Adrijan Ribaric: Yeah. Thank you very much, Natalie. My name is Adrijan Ribaric. I’m the Head of System Modeling and the connection to the industrial internet here at Sentient. I’m assisting in the development of several extension products but also specifically, I manage the [Inaudible 00:02:31] Sentient DigitalClone system product and the connection to the industrial internet. My background is in the numerical modeling, simulation of the dynamic system and especially emphasis on [Inaudible 00:02:46] and finance elements analysis. I would like to go and talk a little bit about Sentient itself. First and foremost, Sentient is a prognostic technology company. We apply that technology in three fundamental areas. In the first area, we determine the life of the system and components in the design stage or already in the field and [Inaudible 00:03:26] and some of them I will introduce a little bit especially for the wind industry.Then, based on these models, we built theories of commercial services to filter baseline reliability for field and product. These services are built on a skillful platform called DigitalClone. The objective is to maximize the return of investments for your field and assets. In the third area, we do this on top of the industrial internet. Nowadays, lots of industrial equipments are connected to the internet where large amount of usage data, center data are all sort in one or the other form. Then, we interface with the data and connect them to our prognostic model so that based on the usage of these equipments, we can determine their remaining useful life in real time. Specifically, it means that life in predictions are bound in individually to specific aspect. Tension applies to prognostic technology in different industrial sectors. Here is the summary of [Inaudible 00:05:03]. I would say we are getting good traction in the market for prognostic and as you see, company name on this list and I just assumed that you are already familiar with Sentient.If you do not know Sentient well and would like to know more about its history and background, then please visit our webpage. Now, I would like to come to the main part of this presentation. Here, we’d like to talk about up and derating of the GE 1.5 megawatt turbine and the effect of it. What you see here in the diagram is a simplest exponential calculation. You would have the [Inaudible 00:05:52] that even if this turbine is not producing any power or you have cost like lend prices insurance and this cost also increase due to variable cause added to it due to [Inaudible 00:06:11] filter or minor and major repair as well as overhaul. In this simple calculation, we spread out the cost over several years and it’s higher the rating of the turbine so it’s higher the cost. Down here on the [Inaudible 00:06:30] you see the rating of the turbine in megawatt. Now, at the same time, you have income from the energy that’s going to be thawed. In reality, that’s not a straight line.What you do know is using the cost estimation–the income estimation, we want to find the best rating settings to find the maximum profit. This is usually done in a more complicated way but here just a simple example. Now, up and down ratings can be done in different ways. We say, for example, change the control logic which is shown in the upper figure or we can add [Inaudible 00:07:36] generator or we can even change the plate size, makes half larger plate of smaller plate. There are several other options as well. Now, what is the current baseline reliability? That’s very important question. What can happen is actually that based on reliability that you assume for the whole [Inaudible 00:08:12] might be different for different components, or for different assets. It depends on the location where the turbine resides but it also depends on the bearing of whom are the bearings to supplies from or who built the DS Water Delubrication and so on.

If you can’t answer this question, then all the subsequent questions, how does derating affect the reliability or how does derating affect the process or return of investment? Then, these questions cannot be answered. That causes high uncertainties in cost calculation and profit estimation for the future. At this point, I would like to set a proposed question. The question is how come confident are you in your baseline reliability analysis for your upgrading reliability study? Are you very confident, moderately confident, low confident or you don’t really have an accurate approach at this turn in time? I can already tell you, usually people that approach us between moderately confident and low confident and everyone has very good approach already but use Sentient technology to supplement the calculation to get more accuracy in this prediction. It turns out that most of them between moderate confident and low confident–and this is something that we expected and that’s why we are looking more and more into the wind industry to apply all our prognostic technology.

In the section of prognostic technology is a micro construction model that performs the damaged accumulation and all of it. Based on periodically reoccurring [Inaudible 00:10:47] it simulates the development of defect between [Inaudible 00:10:51] boundaries that you can see here. Then, the [Inaudible 00:10:58] initiate into boundaries separation. If you can see it here, then finally propagates into small cracks and these small cracks leads later on into long crack. Now, in this propagation model, nucleation is [Inaudible 00:11:16] is developing naturally in this model. Also, everything is done computationally. We don’t perform any physical task. It turns out that this process from nucleation to small crack is highly vulnerable in time. Even small changes in low, in all quality, temperature and so on can send a fact that process significantly. Furthermore, components such as such as here end about 90% of the [Inaudible 00:11:52] in the initiation of micro cracks in this area. Only about 10% of the [00:12:00] result from very small cracks into long cracks until complete operation of failure.

If you want to know more about the specific how this micro structural model works, then please visit our webpage. There are several white papers that explain it in detail and also validation and case study. Now, the technology was initially develop for this and bearings in rotorcraft. However, not much later, we transition it to the wind turbine and expanded it to the complete drive system of wind turbine. The platform for this technology has been called DigitalClone. Based on the wind profile that individual asset is experiencing but also other property for example like bearing supplier, e-manufacturer and so on. DigitalClones can then determine how long in asset would last. In other words, it would tell you what is the baseline reliability of each individual asset. Also, it’s an upgrading and downgrading [Inaudible 00:13:21] then there’s a [Inaudible 00:13:23] and tells what the consequences on the life of each turbine will be. It could be that if we’re operating, the lights become shorter or if you down rate, the light becomes longer. This varies significantly between the assets.

DigitalClone is actually the first commercial application that can provide that level of information undoubtedly and asset to help in light to owner operator. Sentient’s providing this life extension solution right now would uplift a 2.5 megawatt wind turbine. After review with GE, Sentient will also provide the same service to GE 1.5 megawatt wind turbine by the end of this year. Into 2015, Sentient will build a predictive model or other wind turbine. How do we do that actually? Well, wind turbines are collecting bunch of Sentient data or wind speed, wind direction is [Inaudible 00:14:31] also included wind direction on different kinds, identity and so on. All that data is saved in the data system. For Sentient one–Sentient gets access [Inaudible 00:14:46] to that data, we use that to build the wind profile. We use two seconds of OPC data or we use ten minutes average data from data system. With that, we build the wind profile by using the wind [Inaudible 00:15:08] calculating the turbulence and tendency obtaining the wind shear based on the wind on different heights.

Then, we apply that wind profile onto a turbine. Now, in this example, I show you the simulation that we use two seconds data on the wind turbine. We do the same thing for ten minutes. We do a full dynamic analysis where the system has the controller inside to control the pitch angle based on the controller settings within the turbine. What we then do is the wind is acting on the turbine and then we dynamically, we sample then the faucets on the hap for the high frequency. Once we sample these faucets and load, we create a load histogram for the complete turbine’s life that acts into this hap. Here’s an example of a histogram where it shows the [Inaudible 00:16:26] and each part here represents the time that the turbine spends in this specific bin area for this complete life relative to total life and percentage. We expanded here for the GE 1.5 megawatt. This is just an example of one specific turbine in a specific location and we did this for standard turbine. Then, also we performed some ‘what if’ scenarios.

What if we change the controller and upgrade it to 1.56 megawatts or in case vortex generator would be attached to the blade? Then, what you see is that the distribution in this histogram is shifting for all these cases. The consequences automatically receive subsequent life estimation is that we can immediately we can determine what kind of effect of these changes and upgrading would have on the turbine itself. Now, here I show you the example of 1.5 megawatt turbine. We take the load histogram in all six degrees of freedom on the hap so three translation and three moment around the hap itself. Then this load, then what we do is that for each low bin, we calculate what life for each of individual component. The loads will be then transmitted throughout the whole system model into each individual component model. Using our micro structure prediction model, we can then create a viable distribution that tells us the life of that component.

Of course, the manufacturer or the bearing supplier plays a big role here as well. Based on the supplier that you have, you get a specific viable distribution. Changing the supplier would change the viable distribution as well. We do this–we have the capability to do that for the 1.5 megawatt turbine but at the same time we can do that for the Clipper, C96 or C99. Now, how do we really get the life of each component itself? What we establish here is a multi-scale approach where we have a system level analysis where we apply the complete load on the hap and translate them into each individual component. Then, once we have the component load, we go into a component level approach where we use a high fidelity of the element model in order to determine the [Inaudible 00:19:45] of pressure distribution on each component or each contacting services. Using that, we go then into Sentient’s micro structure level model level, apply this contact pressure onto the model and the model itself then determines what the [Inaudible 00:20:10] and how micro cracks are initiating and propagating.

At the same time, we can do all kind of different variation. We can, for example say, what is the lubrication within the components itself? This changes properties in the component level. Also, what is the geometry or who [Inaudible 00:20:39] true bearing manufacturers, then we know what micro geometric changes are between the bearing has the effect on the contact pressure. We can run through all these different scenarios at the same time. Once we know for example what is the key manufacturer, we know what material properties are, then we know what’s a [Inaudible 00:21:08] looks like, what is the orientation, what is the average size and so on. Of course, in each of these levels, there are uncertainties. We have uncertainties in the system [Inaudible 00:21:23] should we tolerate it but also in the component levels what it’s [Inaudible 00:21:29] of the lubrication and then the micro-structural level as well. What we obtain, these uncertainties are propagated to all these models. At the end, we get a life distribution which we present in form of [Inaudible 00:21:49].

At this point, I would like to start another cold question. Have you ever looked at reliability and analysis in this form? If you have done that, I can really appreciate that but then you also appreciate what kind of effort it really takes to do this very high fidelity analysis on each computer or on each component. The advantage is calculations are very cheap and as we progress in time, we have more and more access to high performance computers. Right now, Sentient has access to University of [Inaudible 00:22:39]. Right now, we use 8,000 processes to do all these calculations. What you see here is that the majority of people has thought about that or would like at least to use this kind of fidelity of modeling and simulation in order to supplement their cost estimation for each individual asset. Now, what Sentient is doing, it provides–it performs all these calculations for your field and asset in real time and then provides the results through the DigitalClone life platform.

What you see here is–please try to [Inaudible 00:23:43] where we categorize based on the estimated life of each individual asset. You can immediately see what’s the remaining usual life for a specific type, but then at the same time, you can zoom it into specific asset itself. To obtain more information what that asset is now doing, you can even zoom in furthermore. Here, the system level and look at individual components and based on which components are failing, you can then optimize your maintenance schedule. As a conclusion, what I would like to say here is that many want operators are considering operating of the GE 1.5 turbine to generate more energy from the platform. Also, the DigitalClone ex-sheet calculator remaining usually like the GE 1.5 or other turbines like the Clipper, C96 or C99 so that you can understand the impact of upgrading or downgrading eventually. Important take away here from this short webinar is that there are optimum [Inaudible 00:25:14] cycles that can be traded for the GE 1.5 so that you can get high overall return of investment as I showed you at the beginning.

Simulation scenarios also include [Inaudible 00:25:30] and so on. I want to mention here one more time, right now, we provide the service called the Clipper 2.5 megawatt turbine. We will begin the service for the GE 1.5 prognostic service by the end of the year after review this GE. Then, I think we should come to some questions. Natalie, would you be so kind to see if the audience has some questions?

Natalie Hils: Thank you, Adrijan. I’m sorry there was a mute for a second. Thanks for taking us through that and I do have a lot of questions that we’ve got over the past half an hour here in our presentation. If anybody wants to input any more questions, please feel free to do so at this point in time. Go to webinar control tool panel at the right hand side of your screen. It just says ‘questions’ so I’ll give you guys all a few minutes to input those there. All right. In the meantime, I’m going to start reading the questions that I do have but again feel free to input them throughout the question and answer period. We will run through those. For the first question, Adrian, it is, “Can I see the impact of different main bearing suppliers I have in my older fleet that I’m looking to operate?”

Adrijan Ribaric: Yes. In one of the slides, I showed system representation for gearbox. We determined the life of each component. We determined the roles on each component but then also we account for different geometric properties of each bearing itself. These differences are due to the different manufacturer. In that way, if you know which bearing you have in your specific asset then we can tell you specifically to that bearing how long the life will be. Also, not only this but we can also say how long it’s like will be how each individual asset. Let’s say bearing type A might behave differently in asset A than in asset B and so this need to be taken look at individually at each asset and each component level.

Natalie Hils: Okay, Adrian. Thank you. The next question I have is, “I have a few control changes that I’m considering. Can you tell me if that will impact my high speed bearing life?”

Adrijan Ribaric: Yes, we can answer this question. In one of the slides, I showed that the effect of the [00:28:55] does a different upgrading configuration. These loads are translated through the whole system to each individual component. Different loads have different effects on different components itself. If you enter the specific one, we can tell what the difference of operating has on the loads of this individual component and how that bearing life will change due to the operating.

Natalie Hils:
Great. Thank you, Adrijan. The next one is, “Can DigitalClone be used to calculate the life of gearboxes in their current power?”

Adrijan Ribaric: In the current power settings, yes. What we do is we obtain the historical scale of data what this specific turbine has experienced. Then, we use that one and do a historical analysis of what kind of load is that access experience in the time frame from initiation until today. Then, we can say what damage has accumulated on each individual component and the overall system and what’s the remaining useful life for this system would be.

Natalie Hils: Great. Adrijan thank you very much. The next question I have here from the audience is, “I understand you’re doing this and when but what other industries are you able to do this for, if you are?”

Adrijan Ribaric: Sentient actually developed that technology for the aerospace industry. Well, we were funded for many years by the Department of Defense to do bearing and life prediction for helicopters. They wanted to go from planned preventive maintenance schedule into condition-based maintenance schedule. They were not quite sure how does different operation conditions of the tube of the helicopter would affect individual bearing. They funded that to build a technology in order then to optimize the condition-based maintenance schedule and system. We have done that from the aerospace or Rota craft, we went then for the two offload vehicles and mining. At the same time, the wind was basically dragging us into to use that technology in the wind industry as well. It turns out that the reliability challenges in the wind industry are unseen in other industries. Right now, only DigitalClone can fully answer these questions at a current date in an accurate way.

Natalie Hils: Thank you, Adrijan. The next question is, “What is your experience using actual gearbox condition data in real time to add to the life usage indicator?”

Adrijan Ribaric: We looked–and I think this is similar to the previous question which is how do we know for a certain turbine for a certain gearbox? What is the already accumulated damage and the damage that is or life in the–or what is the life that is left on that gearbox? The most important part of piece here in this calculation is the historical data. If more historical data is available? If more accurate will be the calculation? We reduced with accuracy. I mean, we reduced the uncertainty that exist in the output. If you have the historical data, we would take them, we will press the data and feed them into our system model and component model and as well as micro structural level model and can say what already the damage is at that specific component or what damage has that already been done from that component from the past. We do this for each aspect of gearbox that is already [Inaudible 00:33:56]. I think–I hoped that I answered the question to that.

Natalie Hils: If you didn’t answer that how the user wanted, please feel free to reach out to me with any of these questions, guys. I can always follow up and put you in touch with Adrian as well. I do have another question. “You guys talked a lot about Clipper and GE unit. I was wondering if you can apply this technology to other wind turbine suppliers as well.”

Adrijan Ribaric: The technology itself is not limited to one or the other turbine. What we do is we actually build a gearbox library for individual drive frame. We apply that technology to the aerospace automotive and now wind and heavy equipments, tanks and so on? We used this library of components to build all kind of individual drive train. Right now, we have done this in the Clipper and the GE. We are in the process of building it also for Vesta, Damons and Mitsubishi turbines that we want to introduce in 2015. If you have a concern about the turbine that is not rightly covered by us which is not a GE or a Clipper, then we will most likely be able to offer that service in the near time future. The rank of how we build this asset depends on the cosmic defects. Let’s say, more people are interested in Damons or Mitsubishi, then the next models we will introduce will be a Damon turbine.

Natalie Hils: Thank you, Adrian. We do have time for one more question. I still want to keep everyone too long here and be mindful of everyone’s’ time. I’m going to go read this last question for you. “How much historical data is needed to utilize the model for new [Inaudible 00:36:19] development? For example, do you need three years or can you use wind citing data and design modes to your [Inaudible 00:36:26]?”

Adrijan Ribaric: Healthy history is at least three years because we’ve witnessed that within 1 year, a lot of unusual occurrences can happen that are not really common in every year. Let’s say, we just obtain data for 1 year for 1 specific month and that month turned out to be very bad. We can say what the damage is created for that month but what we would use only that data is to project to the future what the remaining life would be. If we have, let’s say, only one year or only one month of data, then we would estimate your remaining useful life date on that month. In other words, the shorter the time history as less certain we are about our prediction.

Again, one month is not enough data. One year is something that we can work with but I would prefer at least three years and if you’ll have more, even better. On the other hand, let’s say you have turbines that has just recently in your gearbox and of course it has only one year of data dater right now. We just don’t use only one month of this dater data. We use actually everything that the turbine has experienced in the past with the old gearbox. Actually, as a matter of fact, what we can do is if we take the whole history, we can predict the life when the old gearbox ex-sheet failed and then tell what the new life of the gearbox will be.

Natalie Hils: Great. Thank you, Adrijan. Again, I just want to be mindful on your time so I’m going to cut the question and answer period. Here, please follow up with me. If you have further questions I will be happy to put you in touch with Adrian and the rest of our technical team as well. I do have further questions that I received throughout there so I will make sure to follow up with those individuals directly. Again, I just want to remind you of our webinar series. We do host other recordings online in the power point presentations just in case you had to drop off early from any of our sessions or there were some topics that you might have been interested in that we did earlier in the year. Please feel free to reach out to me. Again, I would be happy to provide you with a link to those directly. Thank you all again for joining us and have a wonderful day.


Presenter Adrijan Ribaric, PhD.
Head of Industrial Internet Services

Dr. Ribaric is responsible for the research and development of DigitalClone System. His expertise includes multi-body dynamics numerical models for the dynamic behavior of bearing and gear systems. He holds a Ph.D. in Mechanical Engineering from the University of Arizona.