New physics-informed AI Proves Its Significance by Accurately Solving Problems Even Unexpected

- KRISS·POSTECH joint research team develops physics-informed AI acoustic simulation technology -

- Developed technology is useful for real-time detection of sound, noise, and vibration as well as optimization of environmental systems, and thereby applicable to smart factories and digital twins -

The joint research team between Korea Research Institute of Standards and Science (KRISS, President Hyun-Min Park) and Pohang University of Science and Technology (POSTECH, President Moo Hwan Kim) became the first in Korea to develop an AI technology that directly learns the acoustic physics.

▲ KRISS-POSTECH joint research team

(From left: Seungchul Lee, associate professor of the Department of Mechanical Engineering at POSTECH; Hyung Jin Lee, KRISS senior researcher; Soo Young Lee, POSTECH MS/PhD integrated program researcher)

The new AI acoustic simulation technology can play a pivotal role in predicting real-time changes of sound, noise, and vibration inside a certain system and resolving involved problems. By using the technology, one can conduct the sound/vibration monitoring of various mechanical products, such as household appliances and vehicles, as well as large-scale structures, such as buildings and bridges, and optimize their performances by reflecting decisions derived from the AI simulations. 

The developed AI simulator can be applied to a digital twin technology, which shows great promise in the industrial field. A Digital twin is a virtual replica of an object in reality, existing for simulations of various environmental conditions. Proper moments for repair and maintenance of an actual equipment or system can be efficiently determined by monitoring the performance of a digital twin in the virtual world. A smart factory, which is operated solely by an AI technology covering data collection and factory control, clearly highlight the need of a digital twin.

▲ The joint research team is reviewing experimental results obtained from the AI acoustic simulation technology.

There are two technologies that can be utilized for acoustic simulations of a digital twin at present; one is the conventional AI technology and the other is the engineering simulation technology based on the finite element method. The former works fast and accurately for trained scenarios, but it is significantly weak for unexpected problems. The latter, on the other hand, works accurately, but it is not appropriate for real-time applications due to slow performance. 

The proposed AI acoustic technology overcomes the aforementioned current issues in simulation of a digital twin. The proposed technology shows superiorities over the conventional AI technology in the aspects of a computational accuracy and a handling capability for unexpected problems. Furthermore, the proposed technology achieves an ultra-fast computational performance, which is 450 times superior to the engineering simulation technology.. With the outstanding performances in terms of accuracy and speed even for unexpected problems, the proposed technology is expected to accelerate the commercialization of a digital twin. 

The core mechanism of the proposed AI technology is a physics-informed deep-learning algorithm, which trains the AI neural network to directly learn the physics. The proposed technology derives an accurate analysis even in unexpected circumstances because it can cope with the problems based on the knowledge of physical theories.

▲ KRISS-POSTECH joint research team

Hyung Jin Lee, a senior researcher of KRISS, said, “We can learn a new language through experience, but with an additional training of grammatical rules we can learn it more accurately and efficiently. Our proposed physics-informed mechanism corresponds to the training of grammatical rules. Our proposed technology will create a new paradigm in the AI deep–learning regime specifically for acoustic problems.”

Professor Seungchul Lee of POSTECH said, “Synergy was achieved between KRISS and POSTECH, which specialize in acoustics and AI, respectively. We are now continuing the research to accomplish the actual application of our technology into a digital twin.” 

The study received grants from KRISS and the Institute of Civil Military Technology Cooperation funded by the Defense Acquisition Program Administration and Ministry of Trade, Industry and Energy. The results were published online in Engineering with Computers (IF: 7.963, JCR Top 2.63%), one of the leading journals in mechanical engineering, on April 9.