US researchers are developing an engine capable of learning in a similar way to the human brain, to increase fuel efficiency and reduce harmful gas emissions by 50 per cent.
The researchers aim to apply neural network technology, based on biological systems, to standard vehicle engines. The software is designed to learn the dynamics of the engine ‘on the fly’ – including the best mix of fuel and air – and adapt the engine’s operation to improve efficiency, said Dr Jagannathan Sarangapani, associate professor of electrical engineering at the University of Missouri-Rolla.
‘If the engine were not going in the right direction, the controller would guide it to a better regime. To do that we have to have a controller that injects the fuel at the appropriate time, in the right amounts.’ Calculations and adjustments will be made in real time, every engine cycle, he said.
The researchers aim to operate an engine at cooler temperatures, by using more exhaust gas recirculation (EGR) than has been possible before. EGR reduces the combustion temperature, which in turn reduces nitrous oxide emissions. Current engines become unstable with too much EGR, but a neural network could control this by making small adjustments to the fuel-air ratio between cycles.
Dr Jim Drallmeier, associate professor of mechanical and aerospace engineering at UMR, said calculating the precise fuel demand would also make an engine more efficient, and obviously lead to less carbon dioxide emission.
‘When we talk about operating it lean we mean very lean – much leaner than a standard production engine could currently operate in a stable mode,’ he said.
However, the researchers have only tested their control system in simulations. Rapid calculations and adjustments in an engine are difficult to implement.
‘In a fraction of a millisecond a control system would need to measure an engine parameter (for example, heat release), determine what this is, predict where it’s going on the next cycle and push the engine in the right direction,’ Drallmeier said. ‘This is not something I would anticipate on a 2008 vehicle.’
Sarangapani added: ‘It will take us another few months to tell if it’s possible to do this in real time. And even if there is a breakthrough, engine manufacturers would have to duplicate the results.’
The researchers are investigating neural network technology fast enough to calculate and adjust engine dynamics in real time, and will test an engine in the laboratory this year. But if the test engine works, it is not just car engines that will benefit, said Drallmeier.
‘It’s not so much whether it will work in a Chevy or a Ford, but if it will work on spark-ignition engines across the board, from lawnmowers to four-wheelers.’
The control mechanisms could also be applied to hybrid engines, the researchers said.