How the NIO Formula E Team Employs AI for Slicker Pit Work
Imagine a patient is hooked up to a heart monitor. They’re alive, healthy and behaving completely normally but the image on the monitor indicates a significant heart failure is imminent. “That’s what it’s like if there’s an issue with one of the sensors on our cars,” explains Dave Chen, Head of Trackside Software for the NIO Formula E Team.
Chen adds: “That one sensor might be critical to the running of the car. If a sensor isn’t working we might just see a flat line. Or more confusingly, it might be giving inconsistent readings. We need to get to the bottom of what the problem is quickly.” Enter cyber protection solution provider, Acronis, which has used its experience to develop an Artificial Intelligence/Machine Learning (AI/ML) software tool to recognize the signs of failing sensors.
Large amounts of historical data from the sensors around the car, tagged as ‘good’ and ‘bad’, were fed to learning algorithms. This information included complex sensor faults such as signal dropouts, ‘noisy’ data, and incorrect calibration. The result was a sophisticated computer model that has learned what problematic data can look like. This AI/ML then flags sensor problems to NIO’s engineers.
Dave Chen says: “It lets us spot sensor trouble before wasting time analyzing incorrect information.” In-car sensors are an integral part of the 21st-century racing car and in Formula E they’re more important than in other forms of racing. That’s because Formula E regulations restrict each team to just 20 trackside staff. Of those, about 10 engineers (five per car) will attempt to improve lap times by poring over reams of data fed to them by the car’s sensors.
These sensors are used to measure all manner of temperatures, pressures, and speeds. Sensors on springs and dampers tell engineers how hard the suspension is working and give a clue as to whether different settings might translate into a quicker car.
Some sensors read infra-red energy to deduce how hot the brakes are getting. Others use the speed of deceleration to detect if a car has had an accident. They then open circuit breakers around the battery to isolate it and ensure trackside marshals aren’t at risk of an electric shock. There are even sensors in the drivers’ gloves to measure their heart rate and blood oxygen levels.
The cars Formula E teams use for testing can contain around 150 sensors. But to rein in spending, ensure some performance parity and therefore closer racing, the cars raced in Formula E are limited to between 50 and 60 sensors each.
Weighing between 15 and 30 grams each, these sensors are about half the size of an AA battery. They may be small, but each generates multiple channels of data. The sensors around the car then send this information through 2km of cabling to a data logger that’s about the size of a tablet computer but as thick as a small paperback book. When the race car stops in the pits the information is downloaded; each car generates about 200Mb of data in qualifying, 1Gb during a race.
One differentiator between Formula E and Formula One is that ‘live’ telemetry is extremely limited. Formula E engineers, therefore, can’t watch how a car is performing on-track and consider adjustments before they next lay hands on it. NIO’s Dave Chen says: “That means we have to process the data and crunch the numbers more quickly so we can make any necessary physical changes to the car. Sensor faults can lead us to waste valuable time.”
And as vital as they are, sensors do fail. “They’re subject to heat and vibration, particularly as Formula E races on street circuits which can have quite aggressive curbs compared to those on purpose-built tracks,” Chen reveals.
NIO is no stranger to employing Artificial Intelligence/Machine Learning to solve problems. In its road-car business, the company’s AI/ML scientists and software engineers work on problems such as decision making in uncertain situations, behavior prediction, and detection of rare events to make self-driving cars more capable and safer. And AI/ML is already used in NIO’s production cars: their ES8 and ES6 models feature NOMI, an in-car personal assistant that employs AI to ‘learn’ users’ habits.
For the NIO Formula E Team, AI/ML has become just as integral. Chen says: “We spend pretty much every day poring over the data we collect in an almost forensic way. These cars are very closely matched in terms of lap time. You have to work very hard to understand how your car is behaving to make it better. Missing a failed sensor and the dead ends that can send us down lose us opportunities to improve the car’s performance. At this level, we really can’t afford that."