America's Cup: Oracle Data Mining supports crew and BMW ORACLE Racing
BMW ORACLE Racing won the 33rd America’s Cup yacht race in February 2010, beating the Swiss team, Alinghi, decisively in the first two races of the best-of-three contest.
BMW ORACLE Racing’s victory in the America’s Cup challenge was a lesson in sailing skill, as one of the world’s most experienced crews reached speeds as fast as 30 knots. But if you listen to the crew in their postrace interviews, you’ll notice that what they talk about is technology.
'The story of this race is in the technology,' says Ian Burns, design coordinator for BMW ORACLE Racing.
Learning by Data
'One of the problems we faced at the outset was that we needed really high accuracy in our data because we didn’t have two boats,' says Burns. 'Generally, most teams have two boats, and they sail them side by side. Change one thing on one boat, and it’s fairly easy to see the effect of a change with your own eyes.'
With only one boat, BMW ORACLE Racing’s performance analysis had to be done numerically by comparing data sets. To get the information needed, says Burns, the team had to increase the amount of data collected by nearly 40 times what they had done in the past.
The USA holds 250 sensors to collect raw data: pressure sensors on the wing; angle sensors on the adjustable trailing edge of the wing sail to monitor the effectiveness of each adjustment, allowing the crew to ascertain the amount of lift it’s generating; and fiber-optic strain sensors on the mast and wing to allow maximum thrust without overbending them.
But collecting data was only the beginning. BMW ORACLE Racing also had to manage that data, analyze it, and present useful results. The team turned to Oracle Data Mining in Oracle Database 11g.
Peter Stengard, a principal software engineer for Oracle Data Mining and an amateur sailor, became the liaison between the database technology team and BMW ORACLE Racing. 'Ian Burns contacted us and explained that they were interested in better understanding the performance-driving parameters of their new boat,' says Stengard. 'They were measuring an incredible number of parameters across the trimaran, collected 10 times per second, so there were vast amounts of data available for analysis. An hour of sailing generates 90 million data points.'
After each day of sailing the boat, Burns and his team would meet to review and share raw data with crewmembers or boat-building vendors using a Web application built with Oracle Application Express. 'Someone in the meeting would say, 'Wouldn’t it be great if we could look at some new combination of numbers?’ and we could quickly build an Oracle Application Express application and share the information during the same meeting,' says Burns. Later, the data would be streamed to Oracle’s Austin Data Center, where Stengard and his team would go to work on deeper analysis.
Because BMW ORACLE Racing was already collecting its data in an Oracle database, Stengard and his team didn’t have to do any extract, transform, and load (ETL) processes or data conversion. 'We could just start tackling the analytics problem right away,' says Stengard. 'We used Oracle Data Mining, which is in Oracle Database. It gives us many advanced data mining algorithms to work with, so we have freedom in how we approach any specific task.'
Using the algorithms in Oracle Data Mining, Stengard could help Burns and his team learn new things about how their boat was working in its environment. 'We would look, for example, at mast rotations—which rotation works best for certain wind conditions,' says Stengard. 'There were often complex relationships within the data that could be used to model the effect on the target—in this case something called velocity made good, or VMG. Finding these relationships is what the racing team was interested in.'
Stengard and his team could also look at data over time and with an attribute selection algorithm to determine which sensors provided the most-useful information for their analysis. 'We could identify sensors that didn’t seem to be providing the predictive power they were looking for so they could change the sensor location or add sensors to another part of the boat,' Stengard says.
Burns agrees that without the data mining, they couldn’t have made the boat run as fast. 'The design of the boat was important, but once you’ve got it designed, the whole race is down to how the guys can use it,' he says. 'With Oracle database technology, we could compare our performance from the first day of sailing to the very last day of sailing, with incremental improvements the whole way through. With data mining we could check data against the things we saw, and we could find things that weren’t otherwise easily observable and findable.'
Flying by Data
The greatest challenge of this America’s Cup, according to Burns, was managing the wing sail, which had been built on an unprecedented scale. 'It is truly a massive piece of architecture,' Burns says. 'It’s 20 stories high; it barely fits under the Golden Gate Bridge. It’s an amazing thing to see.'
The wing sail is made of an aeronautical fabric stretched over a carbon fiber frame, giving it the three-dimensional shape of a regular airplane wing. Like an airplane wing, it has a fixed leading edge and an adjustable trailing edge, which allows the crew to change the shape of the sail during the course of a race.
'The crew of the USA was the best group of sailors in the world, but they were used to working with sails,' says Burns, 'Then we put them under a wing. Our chief designer, Mike Drummond, told them an airline pilot doesn’t look out the window when he’s flying the plane; he looks at his instruments, and you guys have to do the same thing.'
A second ship, known as the performance tender, accompanied the USA on the water. The tender served in part as a floating datacenter and was connected to the USA by wireless LAN.
'The USA generates almost 4,000 variables 10 times a second,' says Burns. 'Sometimes the analysis requires a very complicated combination of 10, 20, or 30 variables fitted through a time-based algorithm to give us predictions on what will happen in the next few seconds, or minutes, or even hours in terms of weather analysis.'
Like the deeper analysis that Stengard does back at the Austin Data Center, this real-time data management and near-real-time data analysis was done in Oracle Database 11g. 'We could download the data to servers on the tender ship, do some quick analysis, and feed it right back to the USA,' says Burns.
'We started to do better when the guys began using the instruments,' Burns says. 'Then we started to make small adjustments against the predictions and started to get improvements, and every day we were making gains.'
Those gains were incremental and data driven, and they accumulated over years—until the USA could sail at three times the wind speed. Ian Burns is still amazed by the spectacle.
'It’s an awesome thing to watch,' he says. 'Even with all we have learned, I don’t think we have met the performance limits of that beautiful wing.'
Read more about Oracle Data Mining
Hear a podcast interview with Ian Burns
Download Oracle Database 11g Release 2
Story republished from: www.oracle.com/technology/oramag/oracle/10-may/o30racing.html
by Jeff Erickson Share 11:41 PM Sat 24 Apr 2010 GMT