Hear mention of the Lotus F1 racing team and you might immediately think of speed and innovation. After all, the team designs a new car every two weeks during racing season for drivers who speed up to 200 miles an hour around race tracks.
What fans may not know is that behind the scenes, the Lotus F1 information technology team is about as innovative as it gets. It employs the absolute cutting edge in cloud, big data, Internet of things, analytics, and continuous development.
The setup will blow your mind. Using custom and off-the-shelf products, the IT team builds its own private cloud running 50 virtual servers on site at each race. This lets the group collect and analyze as much as 30 MB of data per lap from as many as 250 sensors on the race car.
"What we're doing is taking real-time data and trying to extract as much knowledge as we can," Michael Taylor, IT director for the Lotus F1 team, told CITEworld. "The key is how we use this information to make informed business decisions, to optimize strategy, and improve the performance of the car. We're using real-time data and analytics to make decisions that directly and visibly impact the outcome."
Formula 1 racing teams are allowed very little time to test their cars on tracks. A team like Lotus F1 will do lots of simulation in its own facilities, including in a wind tunnel that uses 60 percent scale models, but it's allowed only four two-day on-track testing periods during the season. Otherwise, new parts can only be tested out during a race weekend.
That means it's critical to collect as much data as possible when the car is on the track. The Lotus F1 team attaches as many as 250 sensors to its cars, logging "everything you can think of," Taylor said. For instance, sensors track tire degradation and temperature, airflow and aerodynamics, the travel on the throttle, brake and oil temperature, and minute movements inside the gear box. Some sensors may send data once a second while others may transmit 1,000 times a second, he said.
Some of the sensors are off-the-shelf hardware while others are developed in-house. On a machine like a Formula 1 race car, even the tiniest component can impact performance, so the team has to study whether the sensors themselves might impact the speed of the car, Taylor said.
Since the teams have such limited time for practice and test, collecting data in real-time is crucial. When pressed, many businesses that say they collect and analyze data in "real-time" will admit that they do a data dump once or twice a day. Lotus F1, however, transmits data in true real-time over wireless networks operated by the tracks in the 1500 MHz frequency range.
But not everything can be sent over the wireless networks because of bandwidth limitations. "If there were no limitations on bandwidth, we'd send more," Taylor said. At the same time, onboard storage is limited, and if it fills up, it will start overwriting stored data, so a careful balance must be struck. The team collects and retains all data from the cars for later analysis and so it doesn't want to lose any of it.
During race weekends and testing periods, engineers with different specialties view different dashboards. "A performance engineer is looking at optimum lap time and key stats," Taylor said. Meanwhile, engine and hydraulics engineers have different dashboards displaying graphs relevant to their expertise.
Much of the data is useful during the few days that the team is allowed to test the cars and during practice sessions on race weekends. "During practice sessions we're learning about the car and trying to optimize the setup," Taylor said. "We're assessing key systems and taking measurements and making the driver feel more comfortable."
The team is also correlating the data against modeling and wind tunnel data that it collected earlier back at headquarters. That helps ensure that what the team learned at headquarters is playing out on the track.
Even though the Lotus F1 team is collecting as much as 30 MB of data per lap to learn about performance issues as they happen, it's quite limited in how it can respond during a race. The team isn't allowed to do much tweaking of the cars beyond changing tires and asking the driver to use different techniques.
However, some performance, environmental, and competitive data helps guide strategy during race days. Before a race starts, the team has run simulations that paint a picture of when the car is likely to need to stop to change tires and where it's likely to finish at the end of the race. But once the race starts, that picture is likely to alter. "If someone crashes on the first lap, that changes everything. We're constantly crunching data as the race unfolds and reoptimizing strategy," Taylor said.
For instance, race teams are limited in how much fuel they can use. If a sensor tells engineers that the car is burning too much fuel, the team can advise the driver to "lift and coast," or ease off the throttle when braking to conserve fuel.
In addition, the engineers are collecting other data beyond what's being generated from the car. All the teams publish some data, like car speed, to the public, which means each team is able to gather that information about competitors. They also have data about factors like track temperature.
If the temperature of the track is a few degrees hotter than expected, the tires are likely to wear thin in fewer laps than originally expected. That temperature is also impacting competitors. So the Lotus team might change when it calls the driver in to replace tires, taking into account how competitors might behave, so the driver won't come back into the race behind certain competitors.
So how does the Lotus Formula 1 team manage this real-time collection and analysis of critical data?
For each race, the team sets up a private network on the site of the track, consisting of five physical servers running 50 virtual servers and capable of storing 40 TB of data. This local cloud is absolutely vital to keeping the car running, Taylor said. "You can't stop a race and say, 'hang on while we reboot,'" he said.
Much of the software Lotus F1 uses to collect and crunch the real-time data is custom-made. The developer team "focuses on apps that provide the team some sort of competitive advantage. These are the kinds of things you can't buy and you don't want competitors to get their hands on," he said.
The IT team includes 20 Avanade developers, testers, and QA engineers who specialize in Microsoft's .NET technology and work full time for Lotus F1. Rather than directly employ a team of developers, Lotus F1 decided it made sense to use a team from Avanade that can draw on past experience and also tap into the resources and experiences of the greater Avanade community, he said.
That team builds the software that's used to collect and analyze the data that Lotus F1 is ingesting trackside. The developers are continually iterating those apps, just like the engineers who build the cars. "It's about bringing those tiny improvements all the time. The basis of the sport is to bring those things out quickly and try and maximize use of them as quickly as we can. It's much the same in our IT organization," he said.
Traditionally, developers might spend two to three months developing a new functionality for an app, he said. "We don't have that time. We need it in two weeks for the next event," he said.
Back at headquarters, which Lotus F1 refers to as "the factory," the IT team also uses a host of cutting edge technologies. It does advanced 3D modelling and uses a high performance supercomputer an Intel-based HP machine -- to study fluid dynamics around surface shape and structure of the cars. Its wind tunnel uses a 60 percent scale model to test new parts.
The factory also has a simulator where drivers can test out the latest build of the car. "If you're producing physical parts, its a waste of time and money," Taylor said. "So we look at ways to do modeling to get parts at the point where we say, 'ok, it works,' and then make it two days later."
The Lotus F1 IT team does use some off-the-shelf products and services. It's now using Office 365 and runs its public Web site, www.lotusf1team.com, on Azure. It also uses SharePoint, Lync, and Dynamics AX. The team at headquarters also uses Power BI, Microsoft's set of services that support collaboration and sharing of data visualizations.
"If it's essential to the organization but doesn't really offer any clear differentiator, we look to push that into the public cloud," Taylor said.
Looking to the future
In the future, the Lotus F1 IT team would like to be able to better take advantage of cloud technologies. For now, engineers trackside have heavyweight laptops because the software does around half of the processing work on the laptop and half on the server side. If the team could shift more of that work to the cloud, engineers could consume the data on mobile devices or tablets, he said.
It's not quite possible yet though. "Latency is a challenge," Taylor said. Part of the problem is that the tracks are sometimes in relatively remote spots and their networks aren't always as fast as the team might like. "It will come," he said.
In addition, the available wireless networks aren't as capable as the team would like either. The team has done some trials using WiMax wireless networks to transmit data from the cars, but so far they haven't been able to handle hand offs between base stations with the cars travelling as fast as they do.
The Lotus F1 team would also like to make better use of all the historical data that it's been storing over the past ten years or so. "There's huge value we can unlock from historical data," Taylor said. "We could probably glean a significant amount of intelligence if we can correlate between what happens on the track and what happens in design and manufacturing."
His team has done some work with Microsoft's Azure team to explore the potential here. He didn't think that Lotus F1 could take advantage of Microsoft's recently introduced Azure Machine Learning service because of the uniqueness of the organization's data. For now, because the value of such a machine learning project isn't totally clear, he hasn't committed to moving resources away from other work to further investigate the potential for machine learning.