어텐션 기반의 설명가능한 신경망을 사용한 충격파 예측 = SHOCK PREDICTION USING ATTENTION BASED EXPLAINABLE DEEP NEURAL NETWORK

다국어 초록 (Multilingual Abstract)

As artificial intelligence (AI) technology gains popularity, interest in using AI as a new method is also growing in the field of aerodynamics. However, AI models based on deep neural networks have the drawback of being difficult to understand in terms of their internal mechanisms. Consequently, there has been an increase in attempts to interpret AI models using explainable artificial intelligence (XAI) technique. This study focuses on ensuring the reliability of AI models when applied to aerodynamic problems. To achieve this, attention-based XAI was utilized to predict shock waves on airfoils. Also, to enhance the fidelity and stability of the attention mechanism, multi-head attention method using attention in parallel was employed. Additionally, a new metrices based on entropy were proposed to evaluate the fidelity and stability of XAI. This demonstrated that the application of multi-head attention improved the performance of XAI.

참고문헌 (Reference)

1 Ribeiro, M.T., "Why Should I Trust You?: Explaining the Predictions of Any Classifier" 2016

2 Kulfan, B. M., "Universal parametric geometry representation method" 45 (45): 142-158, 2008

3 Schlegel, U., "Towards A Rigorous Evaluation Of XAI Methods On Time Series" 4197-4201, 2019

4 Markidis, S., "The Old and the New: Can Physics-Informed Deep-Learning Replace Traditional Linear Solvers?" 4 : 2021

5 Krishna, S, "The Disagreement Problem in Explainable Machine Learning: A Practitioner's Perspective"

6 Sutskever, I., "Sequence to sequence learning with neural networks" 2 : 2014

7 Petsiuk, V., "RISE: Randomized Input Sampling for Explanation of Black-box Models"

8 Jameson, A, "Origins and further development of the Jameson–Schmidt–Turkel scheme" 55 (55): 1487-1510, 2017

9 Alvarez-Melis, D., "On the robustness of interpretability methods" 2018

10 Graves, A., "Neural turing machines"

11 Bahdanau, D., "Neural Machine Translation by Jointly Learning to Align and Translate"

12 Li, J., "Low-Reynolds-number airfoil design optimization using deep-learning-based tailored airfoil modes" 121 : 107309-, 2022

13 Hochreiter, S., "Long short-term memory" 9 (9): 1735-1780, 1997

14 Park, S. H., "Implementation of kw turbulence models in an implicit multigrid method" 42 (42): 1348-1357, 2004

15 Aupoix, B., "Extensions of the Spalart-Allmaras Turbulence Model to Account for Wall Roughness" 24 (24): 454-462, 2003

16 Yoo, S., "Explainable artificial intelligence for manufacturing cost estimation and machining feature visualization" 183 : 115430-, 2021

17 Hong, Y., "Enhanced High-Order Scheme for High-Resolution Rotorcraft Flowfield Analysis" 60 (60): 144-159, 2022

18 Luong, T., "Effective Approaches to Attention-based Neural Machine Translation" 2015

19 Kim, B., "Deep Fluids : A Generative Network for Parameterized Fluid Simulations" 38 (38): 59-70, 2019

20 Yang, S., "Data-driven physics-informed neural networks : A digital twin perspective" 428 : 117075-, 2024

21 Vaswani, A, "Attention is all you need" 2017

22 Cook, P.H., "Aerofoil rae 2822-pressure distributions, and boundary layer and wake measurements. experimental data base for computer program assessment" AGARD 47-, 1979

23 Lundberg, S.M., "A unified approach to interpreting model predictions" 2017

24 Pulliam, T. H., "A diagonal form of an implicit approximate-factorization algorithm" 39 (39): 347-363, 1981