AI on its way to replacing humans in hypersonic weapon design: Chinese study
- Veteran hypersonic weapons adviser leads research on an artificial intelligence system that trains itself to better analyse wind tunnel experiments
- In China, as hypersonic research advances to Mach 8 and above, the amount of experimental data to be processed and analysed has risen significantly, says team
The machine could identify most of the shock waves occurring in wind tunnel tests, even though it was not instructed on what to look for, the researchers said.
Without human intervention, the AI machine built a knowledge base of its own to aid the development of new engines for hypersonic missiles or planes that could travel longer distances at much faster speeds, according to its creator.
The research team, led by professor Le Jialing with China Aerodynamics Research and Development Centre in Mianyang, Sichuan, published its findings on March 16 in the Journal of Propulsion Technology, a peer-reviewed publication run by China’s aerospace defence industry.
22:01
Are aliens among us? Michio Kaku on UFOs and China-US space race | Talking Post with Yonden Lhatoo
Le has advised the Chinese military on hypersonic weapon technology for more than three decades, according to openly available information.
Le and his colleagues said the human brain could no longer keep up with the rapid pace of hypersonic technology development.
As more countries join the race for hypersonic flight at five times the speed of sound or faster, experiments to simulate the extreme flight conditions are being carried out in many wind tunnels around the world.
What are hypersonic weapons, and why is there a race to develop them?
Each experiment produces tens of thousands images per second. These photos must be manually scrutinised by experienced researchers, often pixel by pixel, for clues to solve theoretical or engineering problems.
In China, as the focus of hypersonic research advanced to Mach 8 and above in recent years, the amount of experimental data to be processed and analysed rose significantly, the researchers involved in these projects said.
While AI was good at handling data, there was a catch in its application to hypersonic research, Le’s team said.
Most AI systems need to be coached by humans. In a typical training session, for instance, the researchers must carefully outline a shock wave, label it with information and then correct the AI if it makes a mistake in practice.
But the aerodynamics at hypersonic speed were so complex, and the experienced scientists so busy, it was “difficult to provide a large amount of training for deep learning models”, Le said in the paper.
The AI system they developed required no training.
The machine would label what it believed was a shock wave by evaluating the location, brightness and colour of each pixel.
The technique, known as unsupervised segmentation, is based on a mathematical theory on graphics that can establish a relationship between seemingly unrelated objects.
Ukraine war is backdrop in US push for hypersonic weapons
The AI would use these initial results as training material to improve its performance on shock wave recognition continuously until it could detect the shock wave patterns.
In a wind tunnel experiment, the shock waves identified by the AI were an 85 per cent match to those marked by human experts, according to the researchers.
Its overall accuracy was nearly four times that of traditional computer software.
An image of a hypersonic wind tunnel test contains a large amount of turbulence, and it can take human experts “an enormous amount of time and energy to label the critical shock wave structures pixel by pixel”, said Le and colleagues in the paper.
The AI, based on a low-cost, three-year-old graphic card, took about 9 seconds to process an image.
02:06
Chinese hypersonic weapons test ‘has all of our attention’, US General Mark Milley says
This knowledge would allow the AI to predict the occurrence of shock waves and come up with hypersonic weapons designs to better control air flow, the researchers said.
China plans to extend hypersonic flight technology from military to civilian sectors.
An official goal by 2035 is to build a hypersonic plane that can carry passengers to anywhere in the world in an hour or two.
The plan was considered too challenging by some aviation experts because no commercial airlines operating today can fly faster than the speed of sound.
Some Chinese scientists believe AI can help overcome some of these challenges.
One technology developed by PLA researchers last year, for instance, would allow a hypersonic aircraft to write new software by itself after launch for more precise flight control.
This knowledge would allow the AI to predict the occurrence of shock waves and come up with new hypersonic weapons designs with improved ability to control the air flow, said the researchers.
China plans to expand the application of hypersonic flight technology from military to civilian sectors.
An official goal is to build a hypersonic plane that can carry passengers to anywhere in the world in an hour or two by 2035.
The plan was considered too challenging by some aviation experts because no commercial airlines in operation today can fly faster than the speed of sound.
Some Chinese scientists believed that AI could help overcome some of these challenges.
One technology developed by PLA researchers last year, for instance, would allow a hypersonic aircraft to write new software by itself after launch for more precise flight control.
