Research
AI Therapeutics
In the field of AI healthcare, current advancements heavily rely on vast empirical data, which poses challenges when dealing with application problems where obtaining sufficient data is difficult. As a breakthrough solution to these problems, we propose the utilization of well-established mathematical and physics models to develop physics-aware/physics-informed neural networks for the medical domain. Our research laboratory is at the forefront of integrating these innovative approaches, bridging the gap between traditional scientific principles and cutting-edge AI technologies. By incorporating physics-based knowledge into neural networks, we aim to enhance the interpretability, generalizability, and robustness of AI models in healthcare applications.
AI-mediated Treatment Planning & Monitoring System
We aim to develop AI-guided systems that informs doctors about crucial infromation enabling personalized and optimized treatment procedures. We present a cutting-edge approach where patient-specific numerical analysis, generated based on individual medical imaging (i.e., MR, CT, etc), serves as the foundation for creating sufficient training data. By training various artificial intelligence models (CNN, GAN, VAE, etc) on this data, we enable real-time generation of pressure fields within the human body and provide optimal treatment plans. This revolutionary method empowers us to offer personalized and effective AI-guided therapy, revolutionizing the way we approach patient care.
Super-Resolution Neural Network for Simulation Result
To obtain precise simulation results, high-dimensional methods or high-resolution simulations in space and time are necessary. However, these high-dimensional simulation techniques often come with high computational costs, making them impractical for clinical settings. To overcome these limitations, we adopt a super-resolution (SR) technique, which enhances the spatial resolution of the simulation space from low-resolution (LR) to high-resolution (HR) without incurring additional computational expenses.
Computational Mechanics
Necessity of efficient and better numerical technique is constantly in demand along with development of technology and advent of more complex real-world systems. The modern computational methods are well-established, but there are still remained region to be optimized in the numerical procedures. We focus on developing breakthrough numercial technique by appropreiately utilizing artificial intelligence technology in the middle of numerical procedure.
Continuum Mechanics Based Beam Element
Beam finite element is highly efficient structural element which can significantly reduce the computational cost, but it also has strict assumption and limited modeling capability. The primary research goal is to enhance the beam modeling capabilities by enriching additional displacement modes into the continuum mechanics-based beam formulation. This advancement aims to address complex problems that could not be effectively solved using conventional beam elements.
Shell Finite Element
The objective of this work is to improve performance of 3-node triangular shell element based on the concept of the MITC3 shell element. In these works, the locking problems are successfully alleviated by applying numerical techniques such as new tying scheme, Hellinger-Reissner functional, and the partition of unity method.
Composite Material Model
A nonlinear 3D finite element formulation for functionally graded beams is developed. This work is the world first investigation into FG beam finite elements in 3D with consideration for the warping effect. An efficient method to represent the intricate warping effect for elastoplastic torsional analysis of composite beams is also developed. A major challenge in this work is how to account for the evolution of warping functions efficiently as materials yield at various locations at different rates. The proposed method shows an excellent performance in various numerical examples despite its simplicity.
AI-powered Finite Element Analysis
Finite element procedure is well-established, but there are still areas that need to be optimized. Applying optimization methods during the analysis procedure can certainly improve performance, but on the other hand, it becomes difficult to practically utilize them due to the computational costs involved in the optimization process. As a breakthrough technique, we use pre-trained AI models that can minimize these computational costs and improve the performance of finite elements.
