How a data center improves car engineering

Applying continuous integration to cars

Much like modern applications, a racing car is a complex system of thousands of components. McLaren's development process involves monitoring hundreds of sensors in each vehicle which are used to improve operational decisions during events. This data is then fed into machine learning algorithms to improve the models used to similar vehicle performance.

The resulting designs are never really ready. McLaren tests new prototypes and fast tracks new designs in time for the next race. Engineers go from simulation, to validation, to test in short iteration cycles. McGrath said, "There is no doubt that data is a source of competitive advantage."

The design facility for new cars is a data-driven environment. New configurations are tested by skilled drivers in simulators that mimic where the new model is expected to be driven. The resulting cars are customized for individual drivers on individual tracks. The tight coupling between simulation runs and driver performance can be fed back into new designs in 20-minute iteration cycles. McGrath said they can achieve more in one day using this environment than in a week of on-track testing.

Simulations make data actionable

Live data is streamed back from races to the super computers running in McLarens headquarters in the UK. This provides actionable intelligence, which can be used for making critical decisions during the race. This allows a collection of applications in McLaren's data center to analyze thousands of possible strategies throughout the race to achieve peak times. For example, service workers need to change the tires throughout the race. If they change them too frequently, the car loses time. But if they wait until a tire blows, they face a longer delay. Predictive intelligence helps identify the optimal time for changing tires and doing other sorts of maintenance during a race.

McLaren is working with other industries to leverage the same principles like high-speed rail transit. McGrath said they start with a focus on transforming the experience of the customers and then transforming the business. This allows the railway to go from schedule-based maintenance to condition-based maintenance. They are also looking at transit capacity to address challenges beyond that. The rail schedule in the UK has traditionally been written two years in advance. They are now working with McLaren to implement dynamic scheduling based on actual metrics.

Similar principles are also being used to improve air traffic control at Heathrow Airport. The traditional strategy has been to measure information about the air traffic, and then react to anomalies to prevent crashes. Better data models will allow the airport to begin thinking about optimizing for key performance indicators (KPIs) like getting passengers in and out and smoothly connecting customers between flights.

As part of this process, McLaren is developing a second screen experience for air traffic controllers that can look into the future and suggest change to schedules and predict what effects these will create. This is not seen as a threat to operators because they are included as part of the loop. This approach promises to drive the development of applications that promote cognitive enhancement for all types of decision makers.

Mainstream businesses could also benefit from applying similar predictive applications. McLaren built a simulation for Goldman Sachs to model their data center usage so they could sell off the extra capacity they were not predicted to use.

Even the physical data centers could be improved through better simulation applications. Data centers now consume 2% of the world electricity, about half of which is used for cooling. McLaren is working with a data center builder to create applications to test out new designs. For example, the latest simulations show that doughnut-shaped buildings would require significantly less cooling than square buildings.

McGrath said that organizations should apply design thinking around their business objectives to build applications that deliver superior business performance. This is not just about faster pipes, or more efficient algorithms. It involves a process of iteratively testing out simulations of the impact of variations on business objectives.

Next Steps:

The role Agile plays in software development

Hiring software developer requires artificial intelligence

Try outsourcing when all else fails

What methods have you found to successfully improve the software development process?