Transforming Wind and Solar Efficiency Through Analytics

Adam Sroka (00:05)
What if the key to unlocking more clean energy wasn’t just building new turbines or solar farms, but getting more out of the ones we already have? Today I’m joined by Philip Bradstock, co-founder and head of analytics at Bitbloom. Philip has spent his career in renewables from turbine design and modelling through to analysing how these machines really perform in the field. At Bitbloom, he and the team are helping owners worldwide squeeze out more from their performance and the hidden percentages that are lost in the weeds.
monitoring more than five gigawatts of assets across the go. We get into data, AI, and the readiness of building a business in this space. Stick around. It’s a great conversation.
Hello, and welcome back to the Hypercube podcast. I’m delighted today to be joined by Philip Bradstock, co-founder and head of analytics at Bitbloom. Philip, for the audience that maybe are less familiar, can you introduce yourself and a little bit about who Bitbloom are and what they’re up to?
Philip Bradstock (01:22)
Yeah, so yeah, thanks a lot for the introduction and I’m very pleased to be here. So my name’s Philip Bradstock and I’ve been in the renewable energy space essentially my whole career starting in 2010 and been on a little bit of a journey through mostly wind energy leading to where I am today with Bitbloom. So I started off my…
My career with Garuda San, which is now called DMV or now part of the wider company DMV. And there I started off doing aerolastic modelling, so modelling turbines and doing turbine simulations for the purposes of turbine design. I was doing that for quite a while before then moving into some wake modelling. So more looking at flow and terrain over the site and looking at energy losses in turbines due to.
their wakes on one another in the process of validating those models. So those models are validated with operational data from turbines are started to get the possibility to really look at some real operational data from wind turbines. And once I’d been in the modelling space for quite a long time, this was a real insight to see actually how they really work in the real world, which is quite often not as you imagine it. So this kind of led me down one path.
another, particularly with one of my co-founders Staffan and who has lot of experience analysing this operational data to then found Bitbloom, which is essentially the purposes of analysing operational data from wind turbines and now a few solar plants to really optimise their efficiency and find out any issues that are going on. So we now sell software and services on top of those software to analyse the performance and condition of now around five gigawatts of assets worldwide and to make sure the owners of those are basically getting all the energy out of those that they should be in and that they’re in tip-top condition.
Adam Sroka (03:36)
Very cool. So my doctorate was in laser physics, computational laser physics, very much like numerical computations of real systems. And again, because a lot of the analytical models are really hard to apply to modern day complicated light laser cavity designs. So I totally get that transition from, okay, well, here’s the theory. Here’s what’s all the nice polished exercises for your masters or whatever into this is what a real system looks like and it’s almost impossible to apply those as they are, they are some significant tweaking. So walk me through a little bit, like what’s the difference between MSC energy forecasting, modelling to what SIFT is doing, what you’re doing at Bitbloom.
Philip Bradstock (04:24)
That’s interesting just to go back on the degree thing, because I also did a physics degree and was involved in laser science as well when I was doing my degree. And a lot of, you’re right, a lot of the numerical modelling that you do in physics is applicable to a lot of systems. And that’s really the skills that led me into doing the wind turbine modelling early in my career.
Adam Sroka (04:49)
So what’s the difference between like what you would do in a foundational kind of modelling, like the MSC energy forecasting type exercises to what SIFT is doing. just walk me through some of the delta in the more sophisticated modelling you’re doing. Just as a modelling nerd, I’m really interested.
Philip Bradstock (05:14)
Yeah, well, I think, you know, this always gets brought up whenever you’re dealing with data is data cleaning. And it’s often given as a bit of a one-liner, how have you your data cleaning? And there’s a lot of different aspects to that. Firstly, there’s, you know, the more obvious stuff, just cleaning out kind of erroneous values. But then there’s a lot of sort of harder to spot issues, like sensor miscalibrations or the bias in that data that might not be immediately obvious. So, in wind, you know, we have the problem of how we’re measuring the wind speed locally in those turbines and then correlating that against external references. And often those local measurements can have a bias and be misleading. so knowing all these kinds of peculiarities, I think is very important. But also there’s so much depth you can go to in the analysis of data. So whilst there’s the RMSE modelling, which will be covering those kind of most important basic tenets that I think most organisations are doing, there’s a lot of further analysis that can be done to really understand how these turbines are performing because there’s a lot of room for error or there’s quite a high uncertainty on how…
in particular with wind turbines are performing. So it’s really important to do further analysis to really understand that.
Adam Sroka (06:43)
What is it about energy that makes this appealing career choice a place to start a business for you?
Philip Bradstock (06:51)
Yeah, my interest in the energy transition in general runs pretty deep. So I first started to listen to talks on, I guess, the energy problem when I was a teenager at school. And it was clear then that something, you know, was going to come to a head in this space. I think back then quite a lot of the talk was about, you know, peak oil or running out of oil, which I think more sort of turned out to be being an issue, also climate change was already very much on the spectrum there and it’s even more in that sort of public concern now. And it was always clear something had to change there. And that’s kind of been my driving force throughout is the energy transition and making that happen. So kind of, when I left uni, I converted that kind of interest into actually sort of
working in the field in, in particular in wind energy, but really, I think it’s really interesting what’s happened in that 20 year space. It’s, been a fast moving space and the picture has definitely changed from then, you know, let’s put up a few wind turbines and solar plants. And most people will call it alternative energy to now where we’re really seeing high penetration and, and all these things that we were talking about 20 years that we need to do stuff like actually
store that electricity, have storage solutions for high penetration renewables and demand side flexibility and have EVs and these kind of things. They’re all happening now. So it’s really nice to see where we’ve got to and that transition to now be kind of in full speed.
Adam Sroka (08:39)
I agree. Do you know what? I’ve not heard anyone say alternative energy for ages, like years. that’s really like rushed back some memories for me. But I totally agree. think that for me personally, it was similar. It’s like if I can nudge the needle ever so slightly the right way, like if that’s what I spend my time, my labour on, then I’ll be happy. And I think a lot of people are in that boat. My experience is the people that want to work in energy specifically tend to be like
you attract better people for less of the kind of traditional compensation and like they’re less money hungry, they’re less kind of, they’re more just keen and in on the mission. And it’s lovely that you build a kind of community of people that are kind of spiritually aligned to what’s trying to be achieved, but also keen on solving many problems and so on and so forth. And as Mel to say, it can’t be quite lucrative energy. It can be very lucrative if you’re in the right parts of it, but it’s a common thing I think I see a lot across many organisations in energy that you tend to get really good, really clever, really, really keen people.
Philip Bradstock (09:47)
So yeah, I think with a lot of passion as well. And I think you get less kind of silent quitting in renewables as well, because the people are working together for the outcome, even though they may be frustrated by their current work. know, most people want the same kind of positive outcome from their work. And I think that comes through in how hard people work in this space.
Adam Sroka (10:14)
I co-founded myself and also something we see a lot of is that you’ve got solution, you’ve got expertise, you’ve got something that really works and then you bump into a new energy company or a new organisation that you know like you’re exactly the kind of people I built this for. I’d love to showcase it, work through it and the first sort of few people you speak to are like no we already do that, no we do that.
In that kind of scenario, what’s the Bitbloom elevator pitch? How do you walk through those kinds of objections of people?
Philip Bradstock (10:54)
Yeah, I think it’s a really good question because we do come up against that and there’s different levels to which you could do it. And I think in renewable energy that are these devices operating in a variable environment and in particular with wind, you have a huge uncertainty on how well these devices are performing. So if you know, if we compare to a gas turbine, for example, where you’re very much controlling the inflow of your input fuel and you’re able to measure that, I imagine in quite high precision, I imagine it must be easier to identify the efficiency of your gas plant than it is a wind turbine, because a wind turbine is operating not only in a variable environment, it’s a hugely complex
input environment of turbulent flow across the whole of the rotor and so many different environmental factors influencing that. And I think the result of that is that you can’t, with only your kind of basic power curve monitoring, which is where a lot of organisations are at the moment, at least internally, you don’t really know how your turbines are performing to within kind of 5%. So they could be up to 5%.
kind of below their optimum and no one’s really noticing. And also the warranties that they come with kind of have a five percent safety net because of the difficulties with kind of analysing their performance, that that’s the kind of safety net you need to have for those manufacturers to pass off those warranty curves. So that’s a lot of energy.
and a lot of value in those few percent. That was kind of OK for, you know, the early kind of pioneer days of wind. But we’re really now in a situation that is very mature industry now and owners really want to be getting those last few percent off their assets. But doing so is not trivial, given the complex environment that they’re operating in. there’s, yeah, it’s really important to be doing as good a data analysis as you can to understand and try and get back those few percent that may be arising from any number of different complex issues that can occur with these devices. I think on the other side of what we do, which is with condition monitoring and predictive maintenance, that’s really important as well because these devices are kind of large in quantity. They’re often in remote places and wind turbines are
difficult to climb and maintain. There’s an expense there just due to access, even more so if you go offshore. So there, there’s real value in having good insights to how your various components are operating and if there’s any signs of early failure, because otherwise that can lead to very expensive downtime. So it’s really a very different situation to traditional energy sources.
there’s a lot more that can be done beyond the sort of basic power curve monitoring in order to maximise that.
Adam Sroka (14:28)
And I think it’s always funny because that you get that all the time. We really do that. We do that kind of analysis. We do that forecasting. And then eventually you get under the hood and you see like some ropey spreadsheet that like is full of problematic circular references and all sorts of stuff. And like the data isn’t being treated right. And you’re like, OK, OK. And you have to really gently then sort of massage them into understanding the difference between.
what they’re doing and what industry best practices and things like that. It’s a really common pain we bump into. I guess to stay on the theme of running a business and being a co-founder, what kind of urged you onto that journey and how are you finding it?
Philip Bradstock (15:13)
I think in many ways it was accidental. And I guess I started off in a relatively large organisation and a kind of like market leading organisation in terms of the consultancy it was delivering. And from there sort of transitioned into working for a much sort of startup business. And I think the natural sort of curiosities and
and ideas that I had led me sort of one thing led to another there through a bit of freelancing into actually founding this business. I think it’s probably wanting to put my ideas into practice and not having someone above me deciding whether that’s a good idea or not. So being able to be innovative, I think. But yeah, it’s definitely not for the faint-hearted running your own business. You have a lot of ups and downs. There’s a lot of sort of tricky paths on sort of just admin and management side you perhaps don’t really foresee when you just have some good ideas on the technical side. So it’s definitely quite full on. And yeah, you get a lot of high ups and low downs.
Adam Sroka (16:38)
I totally agree. I think like a lot of techies don’t realise that the idea is almost worthless compared to the reality of running the business. there’s plenty, ideas are really cheap and really easy, but implementation and operations of those ideas is the really hard bit. And yeah, you’re totally right. As someone that did an MBA, like I find that you learn so much more being in the weeds and doing it than you can ever like do by studying or reading blogs or going to lectures and stuff. But
I thought if I go back now, it’s good fun. And I think I quite like it. The chaos.
Philip Bradstock (17:19)
Yeah, I think it’s hard to go back once you’ve been in it to go back to having a ⁓ hierarchy and a political structure and everything.
Adam Sroka (17:33)
So what’s the master plan? We are in the GB market mostly, I imagine. Are you looking for global domination? Are you exploring other markets?
Philip Bradstock (17:43)
Well, actually, assets that we monitor are pretty global, actually. So there’s definitely a UK leaning and after that a European leaning. But what do you also got? I mean, the type of customer we serve really well is the kind of medium-sized independent power producer, IPP for short, who may own typically around like one gigawatt of assets, for example. And quite often those may be based in one country, but they’ve got assets across the world or in many countries as part of a kind of diversification plan and also sometimes in multiple technologies as well, wind and solar and battery storage. So in some cases, for example, we’re serving British customers who
have assets in South America, for example. Beyond that, in terms of looking at companies in different markets that we can serve, think Latin America is a really interesting market. They’ve had a lot of renewables deployed there. They’ve got very good renewables resource in general, ⁓ but don’t have the same historical track record or experience of ⁓ operating them as Europe does just having been in this game for a bit longer. on the wind side, then a lot of manufacturers are kind of European based and their expertise, sort of competence centres are in Europe, which often leaves the owners a little bit stranded in terms of like local expertise in the market. And we can really help them bring in some expertise on their side, on the owner’s side.
Take a case to the OEM manufacturer and say, yeah, can you change this on the turbines because they’re not performing as they should be.
Adam Sroka (19:45)
You mentioned at the top of the episode that you’re starting to monitor some solar and you mentioned their bears. If you look to flexibility and monitoring bears, what other asset classes are you looking at? Do you have like an abstraction of the ultimate asset model? How’s that going?
Philip Bradstock (20:03)
Yeah, so on the data and analytics side, the tools that we’ve built are data agnostic. So they’re really only assuming sort of that it’s machine time series data. So then it’s a question of the specific targeted analysis we’re able to build in those and how the main expertise then influences those. at the moment, yeah, we’re monitoring about five gigawatts of assets at the moment.
sort of four and a half is wind and half a gigawatt in solar. So it’s an off-growing interest to us. We definitely have more historical domain expertise in wind, but a lot of our underlying tools we can apply to solar and yeah, we’re hoping to expand that area because there’s a lot of insights believe we can bring to solar. The problems are a little different because I think whereas with wind turbines you’ve got a complex mechanical system, rotating parts with control systems. You’ve got a lot of things to monitor in there and a lot of insights to bring. Whereas on solar, it’s a bit more of a kind of quantity problem and the hierarchy in the site. So you have less kind of individual power producing devices, if you like, but there’s a structure to that. So there’s panels and strings and inverters and just processing the data and finding out the strings that aren’t working properly, it’s much more of a data hierarchy problem rather than monitoring efficiency of complex pieces of rotating machinery.
Adam Sroka (21:52)
How do you bring all that together and surface those? So if you’ve got mixed portfolio sites, and if you do, how do you surface those insights back to the customer? And how do you communicate the output of your work generally?
Philip Bradstock (22:06)
Yeah, so we have a web platform which is running automated analysis. So it’s bringing in new data from the sites every night, running new analyses, and then disseminating the sort of results of those analyses through a web interface and through various charting and alerting. So typically the
kind of monitoring we’re doing, we’re not looking at kind of intraday control room kind of base stuff. We’re looking at kind of medium to longer term trends of things that you need to optimise. And yeah, so we’re delivering that through various visualisations of these analysis and they have alerts attached to them so that when the various KPIs that we’re calculating breach alert thresholds, then those are raised through the system. But we don’t stop there because I think in our experience, just producing alerts is quite far off what you really need to do in order to get action on these devices and make the appropriate adjustments. So there’s a task management system built in to the interface as well. So alerts, you can then attach to two tasks. And actually what we see in many cases is a singular issue. It’s being alerted for many different analysis. So
You are then sort of writing a task or an issue description, if you like, to link these various pieces of evidence we’re getting from different alerts and for an analyst to go in there and say, yeah, this is now pointing to this issue. This is a possible root cause. We suggest you go and inspect this, that, and this.
Adam Sroka (23:58)
Yeah, that makes sense. And it was going to be one of my follow on questions. Like, how do you, how do you go from, okay, well, that’s interesting to driving action. Because I find a lot of us in energy and tech generally are very guilty of producing lots of really clever analysis that, that end users actually can’t surface into like real recommendations. These clever bodies of work end up kind of falling by the wayside because people haven’t really taken that next step or done that last mile. But it sounds like their task management system makes a lot of sense and helping connect the dots for people can help bring that forward.
Philip Bradstock (24:39)
Yeah, absolutely getting action to actually happen on the site is often the hardest part of it. And I think in wind, you quite often have a difficult relationship, it’s fair to say, between the owner of the assets and the OEM, who’s normally got the service contract to service those turbines, and they’re well placed to service those turbines because
they, you know, they design them and they’ve got a competitive edge there, but they’re not being incentivised to maximise the efficiency of those turbines. That’s the owner who needs to do that. And when issues are spotted, getting action to happen, you really need to lay out clear evidence of what’s happening and why they need to go and fix something. And because, just you know, saying our model produced this alert and therefore you should go and check the turbine is not going to cut it, unfortunately. So it’s always been deep in our ethos, kind of the explainability of the analytics and kind of producing an evidence base to make sure action does actually get done.
Adam Sroka (26:01)
And what kind of role does automation play in that kind of workflow then, like going from analysis to tasks to.
Philip Bradstock (26:08)
Yes, so all the analytics and the alerting that is highly automated. So from a monitoring perspective and human time in the system, if everything’s running fine and no alerts are being triggered, then kind of very little time is needed. However, when that last step going from alerts to you know, writing an issue description and kind of producing the evidence base. We still have a human in the loop there, and I think that’s that’s the best thing to be done, because, as I said, often there’s there’s different alerts sort of pointing to the same issue and you can build up an evidence base. And really, job with the software is to make this job of producing these insights as efficient as possible. But
realistically speaking, we’re still in a world where we need those human in the loops, those domain experts to go and interpret these different analyses and say, yeah, this is probably what’s going on. Or in some cases, write something off as a false positive or low priority. Prioritisation is another very important aspect because we see a lot of different issues, quite a few you could probably leave.
they’re quite low value so they can kind of safely be left in the background for a while.
Adam Sroka (27:42)
of customers that are like just using the platform as part of their workflow totally self-sufficient versus a kind of more managed service model where you’re advising them and helping them make best use of it.
Philip Bradstock (27:55)
Yeah, so the goal of the organisation has always been to make the software as self-sufficient as possible so that we’re very kind of encouraging of customers who do want to use the software themselves and do their own monitoring. We typically find with our owner operator customers, a lot of them are operating quite lean. lot of them have growing portfolios.
or portfolios growing faster than their teams are. And therefore, the current trend we see is that most of the owner operators kind of want a managed service there for us to do the monitoring for them. And we are also the best at using our own software. So that can also make sense for them. We’re also supporting some sort of consultancies.
who do sort of asset management as a service and a few more of them doing a bit more of their own monitoring in house as well as some of our more like technically focused IPP companies as well have teams of analysts and using the software in house. So it’s a bit of a mix, but we see a bit more of a trend for wanting that full service at the moment.
Adam Sroka (29:21)
And what about AI, right? like large average models have come along and sort of a blow in a way, all technology use cases and the future of everything. It’s going to replace everyone. Where do you see that fitting in the stack? Are you using it already? Is there future product roadmap stuff that you think actually that could be interesting?
Philip Bradstock (29:41)
I’d perhaps like to break down AI a little bit because it’s a of a catch-all term these days. And I think in the last year or so, it’s almost become synonymous with LLMs. Whereas, of, us, you know, starting several years ago, they weren’t really in the picture and AI meant kind of machine learning of some sort or another. So on the analytics side, then we’re using machine learning.
particularly regression models, and that’s most commonly useful in normal behavior models. having a data-driven model of what you expect some particular signal to do and then looking for deviations in that. And that might be on the power production side, but also on temperatures of components for predictive maintenance. We’ve always actually taken a philosophy that all analyses should be
sort of classical statistics or physics based first. So we rather than seeing survey mail as a magic tool, we see what insights we can deliver with normal statistics, because if classical statistics is good enough to produce those insights, then that is the model you should be using, because for one, it’s far more explainable, but also sometimes there’s other
factors of that statistical analysis you can leverage and also much cheaper compute costs, et cetera. But the explainability is very important there. But there are some cases in ⁓ this space with very complex environments where statistics and basic physics models aren’t really going to cut it. And you can really deliver additional insights
with machine learning and go to sort of much higher accuracy and more insightful results. So that’s kind of on the machine learning side. Then on the LLM side, that comes in a very different place in our software. So I was talking before about the task management system and there now we’re using large language models to, one thing is to be able to sort of summarise what
often become quite complex evidence bases, but also to link that to disparate pieces of information. So quite a common thing on the operations side is having a lot of maintenance logs, which essentially become unstructured information and being able to link that to tasks and issues where we’re raising, then that’s really useful because we can use AI, LLMs, to say,
OK, what maintenance has been done on this turbine that could be related to that, to this issue I’m writing here. So and that can tell you kind of what maintenance has been done and linked to you the appropriate documents. So it’s really, really useful, therefore, combining this unstructured information with structured analytics we’re producing through conventional analytics if you like or not through an LLM.
Adam Sroka (33:06)
Yeah, now I totally get it. Like a painfully hearing like machine learning and stuff can be called traditional and our traditional AI, which is just makes me feel sick. But the some of the use cases we’ve yeah, similarly, we’ve deployed where things like chats are my wind turbine and look at the logs and chats and logs and stuff like that, actually get like a human understandable summary very, very quickly of, yeah, really complicated, like 50.
million log messages over the last week and stuff like that. It’s a pretty exciting space to be in and you can get really good results quickly. I think I’d love to see more innovative stuff pop up in the space, but that’s only going to happen over time, right? As like more people experiment and try weird and wonderful things with these models as they get more more powerful.
Philip Bradstock (33:58)
I think it’s what LLMs excel in is summarising data, really. It’s not as, you know, some people have a tendency to think it’s sort of they’re truly intelligent or rationalising or anything like that, but they’re really good at summarising data. And what we have in the operations world is a vast amount of unstructured data being created through reports. And this is where they can really specialise to link these pieces of
disparate information.
Adam Sroka (34:29)
So if you had a magic wand and you could make every asset owner operator fix one thing about their data or the way they’re working, what would it be?
Philip Bradstock (34:40)
think to do better analysis with the data that you already have. I think there’s a trend sometimes to these organisations to want to get more and more data or deploy more sensors so that they can have better data. But ⁓ even though this kind of data can be tricky sometimes, there’s a lot of analysis that can be done with the data already being produced. There is already quite a lot of data being produced. ⁓
And even with relatively low frequency 10 minute or one minute data that we’re working with, there’s a lot of insights that can be gained on that. implementing the appropriate automated pipelines and analysis, whether it’s with us or going on your own, there’s a lot more that can be done beyond basic power curve monitoring and energy loss calculations, availability calculations, lot that can be done and more should be done. Okay.
Adam Sroka (35:45)
Well, it’s been a really interesting discussion. think there’s a lot of parallels and yeah, the joys and pains of running an energy analytics business I can totally relate to.
What I would love to know just before we wrap up is how are you finding the quest for talent and like bringing new people on the journey and teaching people about this stuff? Are you finding it easy to like hire, train, deploy people or are you finding that it’s actually difficult to find the niche skills, the energy expertise? Like what’s your experience on that from?
Philip Bradstock (36:28)
So I think domain expertise is quite hard to come by, it says. So I think that’s just the nature of how quickly the renewable energy industry has grown. That if you want someone with 10 years experience in renewables, you’re looking for someone who is in the industry when it’s a lot smaller than it is today. So there’s a lot of competition for those sorts of people.
On the other side, on the sort of data analytics side ⁓ and on the software side, there’s a huge number of people interested in tech and data analytics these days. And there’s a huge number of people very interested and passionate about the space of renewable energy that we work in. So even though they don’t have expertise, then, ⁓ you know, picking bright people from that
kind of ilk, if you like, I think can be a very valuable strategy. There’s a lot of people very willing to learn and able to apply their skills to a new space. yeah, differs a bit depending on kind of how conservative you are about wanting the main expertise or whether you’re taking people who are willing to learn.
Adam Sroka (37:52)
Well, it’s been a pleasure talking to you, Philip. I guess one thing I always like to ask before I let anyone leave is if people in the audience want to find out more about you, about Bitbloom, or if there’s anything you’d like to plug or promote, where should we direct?
Philip Bradstock (38:09)
So I personally am quite active on LinkedIn. Usually try and post a few sort of interesting posts a bit more beyond the just normal plug it. And then also the Bitbloom website, you can find out a lot more about us. We also have a YouTube channel where we’ve got some tutorials on some of our software if you want to get a bit.
deeper into understanding about how our software works and how to use it, then you can go there. And apart from that, we’re at a lot of the renewable energy conferences in Europe. Sometimes I have a stand there. So if you happen to be at one of them, you can drop by our stand.
Adam Sroka (38:59)
Listen, although it’s been an absolute pleasure speaking to you today, be sure to check out the YouTube channel and hopefully bump into you at the next energy event.
Philip Bradstock (39:08)
Yeah, yeah, likewise. been a really great conversation and yeah, good to speak to you. Thanks for inviting me.
Adam Sroka (39:15)
As always, it’s been a pleasure and thank you for listening to the Hypercube ##podcast. If you want to find out more about what we’re up to or meet other people in the space, join the Beyond Energy community where data and energy professionals from across the globe collaborate on what’s coming next across the energy.