Features Hub

Q&A: Director of Digital Engineering Transformation, Mercedes-AMG Petronas Motorsport

Fri 14 Feb 2020

In this exclusive Q&A, Geoff Willis, Director of Digital Engineering Transformation for Mercedes-AMG Petronas Motorsport, explains how Mercedes is using data science and visualisation to stay ahead of the pack

How large a team does Mercedes-AMG Petronas Motorsport dedicate to data insights and analytics?

We want to enable a very large proportion of our organisation to benefit from the data that we collect. That means we are putting a lot of effort into making the mechanisms of interacting with data and getting the data insights, much more automated. And we’re generating a set of analytical tools that will, at one level, present in terms of business information type, but also for specialist areas, tools that also enable them to interact with their modelling and simulation tools. So, in terms of a team specifically working on data, it’s a small percentage of the team. But in terms of the number of people who are directly driven by data, then many hundreds of the team are really focused on data.

In which areas of the sport and the business do these areas provide the most gains?

Predominantly we use data for performance development, but we also use it for reliability. The current generation of Formula One cars are phenomenally reliable and that is thanks to a very rigorous approach to understanding historical data, comparing current data with historical data and underlying models and predicting failures. There is a substantial tie up there, and I think in terms of performance development, we use data to validate the models that we base our simulation on.

Plus, we now drive our performance development through predominantly simulation-driven directions. The data has a significant impact – although quite a lot of the car we can model with great confidence, there are certain parts of the model that are very much more dependent on experimental data, particularly in the area of tyre technology, for example.

Additionally, we now take more and more data from our competitors – from our cars, but also from our competitors’ cars during the race. When I say we take data, we are not tapping into their data, I should be clear, we are provided with a lot of data. The organisers give us GPS data on all the cars and we can see what others are doing – as we build these, this also helps to drive our models. Data drives more and more understanding of what the competition is doing. This allows us to be more dynamic with our race strategy tools. It has raised the complexity of the gaming that’s being done by other teams within the race event.

Geoff Willis, Director of Digital Engineering Transformation at Mercedes-AMG

Register For Big Data & AI World Now.

Tangible, meaningful and insightful data & AI is within reach

Join 763 expert speakers and 526 technology suppliers at London’s ExCeL, March 11-12.

What is one of the most off the wall ideas that anyone in your team has had that turned out to be inspired?

I was thinking about this earlier. It’s an interesting question to ask. We encourage in a number of areas a certain amount of, what we call ‘set aside time’. Here we encourage people to spend some structured time not doing their day job. Thinking about something that’s interesting, something that is sort of relevant and thinking how that can be brought in. Sometimes it’s not directly relevant, but it triggers conversations.

You can imagine perhaps looking at the analysis of the way birds’ flapping wings work, or how they control feathers and think: Well, I can see what they’re doing, maybe I can turn that into something on a car. Lots of interesting conversations come from this. You can see people thinking about applications of technology, thinking: I have got a machine for measuring something, but maybe I can turn it around and measure something else with it. It’s difficult to try and think of a really off the wall idea. It’s a good question.

How do you visualise and interpret so much data (500GB a weekend)?

You’ve got two ways of looking at the data. You can either look at all of the data and clearly 500 gigabytes of data is going to need some sophisticated ways of looking at it. To do that, we need the latest technologies, developed by amongst other people, our partners. The other way you could look at it, is to consider which of that data is interesting, as it wasn’t what you predicted. If you look at that data and 99.99% of it is exactly what was expected, it’s not actually very interesting. It ticks the box that your simulation is correct, but what you’re interested in is where data is reflecting what you didn’t expect.

It’s very much the way that the aircraft industries work with plant models. When you’re testing an aircraft, you measure the data and the moment the data doesn’t agree with the so called plant model simulation, you stop flying and ask why it doesn’t match. Is this a problem with the plane or a problem with the model?

We’re trying to find what is ‘interesting’ in the data – is it actually telling us something new? From that, we can see the unexpected behaviour, a sign saying that our understanding is not quite correct. You can then go back, refine that model, use that data to validate a new model. But of course, with an awful lot of your simulated optimisation data, you’ve just got to trawl through it to get the ability to correlate the optimum path of deploying the car and the optimum path of developing the car.

Plus, when you’ve got a set up for a race car in a race weekend, the environment is changing all the time. You’ve got to land on the head of a pin on Sunday and that’s why you collect all the data – it’s often a very noisy background and things are changing. With simulation tools, we can see a pattern, a trend and where we’re going to end up. You’ve got different ways to visualise and interpret different data, but certainly the team are finding that with tools like TIBCO Spotfire to slice and dice a lot of data, they can really drill down to the bits that they’re interested in.

What are some of the challenges of managing that amount of data?

You’ve not only got the data you measure on the car, but also all the contextual data around it – environmental data, proximity of competitors, the environment, the weather, the effects of the track. Then you’ve got the internal contextual things: what specification was the car in, what did you do, how was the driver driving it, what engine roach did you use, what were you trying to achieve?

We’re finding that one of the challenges in order to be able to use that data and harness new data science techniques, is the work involved in curating, preparing and cleaning that data. We’ve learned, like many other organisations have, that data dumped into a data lake needs work on it. It’s that next level of abstraction – how you turn data to information. Learning how to structure data is the learning path we are on at the moment.

What sorts of data sources are the most powerful for the team?

Simulation and tools for computational fluid dynamics are absolutely an integral part of design now. I don’t think you’ll find anybody in the pit lane that would not agree that the combination of digital simulation and physical model testing is really key to the rates of development.

I’d also say the optimisation data is so powerful. The automated trade-offs between all the design criteria: the chassis, the suspension, the aerodynamic power unit characteristics, the optimisation of the tyres. That is a complete package, it is something we didn’t have 10 years ago and we only had weekly six or seven years ago. And that’s really optimal control – optimisation is a very important tool.

To what degree do your data insights depend on cloud technologies?

We largely use private cloud at the moment – there are some specific technical reasons for that. There are however some areas where we are now using cloud resources and particularly in the areas of data science. We will, I think, continue to have a balance between private cloud and public cloud.

Data is such an integral and valuable commodity to team that there’s still that nervousness about letting it out. But I think that as we become more comfortable with that, we’ll be further exploring cloud technology.

How integral is machine learning to your data efforts?

Right at the moment, it’s not integral, but it’s a growing area. We have some areas of applications which are already productionisable for us, and some areas where we’ve we got to do more work. At the moment, it looks like the nature of our data and the architecture of our data, is probably not quite right and we’ll need to explore a different machine learning technique. Bit by bit, as we’re working with a number of technical partners including Daimler, we’re exploring this area. So I’d say we’re still in transition.

What is the coolest thing you are working on right now?

I never cease to be amazed by all the fantastic bits on the race car. Sometimes you look at some of the bits on the car and think: how did we end up with that? They are driven by the combination of the human insight from the data and the optimisation techniques. Yes, you just look at it and think that’s impressive. You want to tell everybody about it, but you can’t tell everybody about it. And you see that, both in the physical objects and also in the virtual objects. In terms of some of the things I’ve seen through optimisation and simulation, I think –  gosh if we had that insight 10 years ago, we would have been absolutely unstoppable!

It’s not the coolest thing I’m working on, but I think one of the things I like about Formula One is it’s ‘can do’ attitude. I think that’s very cool. You have things that are verging on disaster, but you don’t run around headless, it’s okay, we just get on with it. We see that both trackside and in the factory. There is a rapid deployment of people who look over their shoulder, realise somebody else is in a lot of trouble, so they join in, they know how to fit in, how to interact and act as a support.

One particular example was a sensor or an actuator that was buried deep inside the engine – not an engine part, but buried deep inside. The team had to change it at very, very short notice. You almost couldn’t see the car for the number of mechanics working on it. They were rushing through their legs, under their arms, behind their heads and none of them got in the way of the others – they all knew what they were doing and the thing just happens.

That’s a very graphic, physical example, but sometimes you see exactly the same thing going on in the organisation where we’ve had some bits fail. The whole group of people got involved, the right people broke down the problem, tested, modelled and analysed. All coming back together in a couple of days, they knew the cause, the reason, the mitigation and the solution. That’s applying your resources – we have a huge amount of resource and if you line them up in the right direction, it’s a really impressive machine to watch going into action.

That’s probably the really cool thing I like about Formula One.


data science mercedes
Send us a correction Send us a news tip