2nd US-Japan Workshop on Data-Driven Fluid Dynamics, Lecture notes of Fluid Mechanics

Download 2nd US-Japan Workshop on Data-Driven Fluid Dynamics and more Lecture notes Fluid Mechanics in PDF only on Docsity!

2nd US-Japan Workshop on

Data-Driven Fluid Dynamics

September 05- 07 , 202 2 Kobe Meriken Park Oriental Hotel, Kobe, Japan Sponsored by

Objectives:

Over the past two decades, the fluid dynamics community has enjoyed the advancement in

computational, experimental, and theoretical techniques to analyze a variety of fluid flows.

Developments in computational and experimental hardware, numerical algorithms, and unsteady

measurement techniques have enabled not only detailed analysis of flow physics but also initiated

cross-talks amongst the various disciplines of fluid mechanics. With these powerful toolsets now

available, the fluid dynamics community has started to examine complex flows with high levels of

unsteadiness, nonlinearity, and multi-scale dynamics. However, there still exist limitations on how

modern analysis techniques can be applied to specific fluid dynamics problems. Theoretical and

computational approaches are often limited to relatively simple flows at low Reynolds numbers,

while practical applications require extension to more complex unsteady and turbulent flows.

Extending the current state of the art in flow analysis to higher Reynolds number flows requires

tackling high-dimensional physics and the associated big-data from numerical simulations or

experimental measurements. Some of the recent innovations in data science may hold the key to

address these issues. The aim of this workshop is to gather fluid dynamics and data science experts

from their respective areas and discuss their ongoing progress and challenges in emerging analysis

techniques, including data science, computational & theoretical fluid dynamics, and advanced

experimental diagnostic methods, that can be shared with others to facilitate breakthroughs as a

community. This event will stimulate discussions and collaborations between members of the

research communities to identify key areas that can make the largest impact and to offer a vehicle to

further strengthen research collaborations across the Pacific. Following the success of the first “US-

Japan Workshop on Bridging Data Science and Fluid Mechanics,” we are holding the second

workshop now entitled the “US-Japan Workshop on Data-Driven Fluid Dynamics.”

Target Areas:

Data-inspired techniques for fluid dynamics, including but not limited to data-driven analysis,

modeling, estimation, and control of fluid flows.

Attendees

Yoshiaki Abe (Tohoku University)

Kei Ambo (Honda R&D)

Dave Amels (Intelligent Light)

Byungjin An (Ebara)

Hessam Babaee (University of Pittsburgh)

Ryan Carr (AOARD)

Jiro Doke (Mathworks)

Kozo Fujii (Tokyo University of Science)

Koji Fukagata (Keio University)

Kai Fukami (UCLA)

Susumu Goto (Osaka University)

Hiroshi Gotoda (Tokyo University of Science)

Michael Graham (University of Wisconsin-Madison)

Tomoyuki Hosaka (Hitachi)

Traian Iliescu (Virginia Tech)

Michio Inoue (Mathworks)

Anya Jones (University of Maryland)

Yoshinobu Kawahara (Kyushu Univ/RIKEN)

Soshi Kawai (Tohoku University)

Yoimi Kojima (JAXA)

Petros Koumoutsakos (Harvard University) - keynote speaker

Steve Legenski (Intelligent Light)

Rajat Mittal (Johns Hopkins University)

Takemasa Miyoshi (RIKEN) - keynote speaker

Takayuki Nagata (Tohoku University)

Kumi Nakai (AIST)

Masamichi Nakamura (Morgenrot)

Yuki Nakamura (Honda R&D)

Akira Namatame (AOARD)

Taku Nonomura (Tohoku University)

Fortunato Nucera (Honda R&D)

Shigeru Obayashi (Tohoku University)

Kie Okabayashi (Osaka University)

Yuya Omichi (JAXA)

Yasuo Sasaki (Tohoku University)

Koma Sato (Hitachi)

Koji Shimoyama (Tohoku University)

Kunihiko Taira (UCLA)

Tomoaki Tatsukawa (Tokyo University of Science)

Atsushi Toyoda (Intelligent Light)

Aiko Yakeno (Tohoku University)

Chi-An Yeh (North Carolina State University)

SCHEDULE DAY 1 (September 5)

REGISTRATION

8: 00 - 8 : 45 REGISTRATION AND COFFEE

OPENING REMARKS AND LOGISTICS

8 : 45 - 9 : 00 Kunihiko Taira (UCLA)

KEYNOTE TALK (Chair: Kunihiko Taira)

9: 00 - 10:00 Petros Koumoutsakos (Harvard University)

Alloys of AI and Computational Science for Modeling Prediction and Control in

Fluid Mechanics

10:0 0 - 10:2 0 COFFEE BREAK

SESSION I (Chair: Kunihiko Taira)

10 : 20 - 10 : 45 Taku Nonomura (Tohoku University)

Advanced Fluid Measurement based on Active Use of Modal Decomposition

10: 45 - 11:1 0 Michael David Graham (University of Wisconsin-Madison)

Data-Driven Manifold Dynamics for Modeling and Control

11:10-11:35 Fortunato Nucera (Honda R&D)

GAN-Based Automobile Flow Field Prediction from Simple Geometric Data

11:35-12:00 Chi-An Yeh (North Carolina State University)

Data-Enhanced Resolvent Analysis of Turbulent Flows

12:00-1:20 LUNCH

DAY 2 (September 6 )

KEYNOTE TALK (Chair: Kozo Fujii)

9:00-10:00 Takemasa Miyoshi (RIKEN)

Big Data Assimilation Revolutionizing Numerical Weather Prediction Using Fugaku

10:00-10:20 COFFEE BREAK

SESSION IV (Chair: Kozo Fujii)

10:20-10:45 Byungjin An (Ebara Corporation)

Data Science Application in Ebara

10:45-11:10 Aiko Yakeno (Tohoku University)

Challenges for Delaying Transition to Reduce Airplane Drag

11:10-11:35 Steve M. Legensky, David A. Amels, Earl P. N. Duque & Atsushi Toyoda

(Intelligent Light)

Augmenting DMD and Frequency Analysis with Visualization for Engineering

Applications

DISCUSSION

11 : 35 - 12 : 00 Moderator: Koji Fukagata

12: 00 - 1:30 LUNCH

TOUR OF RIKEN CENTER FOR COMPUTATIONAL SCIENCE

Depart the hotel at 1:30. Detailed schedule will be shared at the workshop. Bus available from

hotel to Riken. Attendees will be split up in two groups (tours will end around 3:00 or 3:

depending on your assigned group). Please use public transportation for the return trip. Attendees

are free to explore the city of Kobe after this tour.

https://www.r-ccs.riken.jp/en/

DAY 3 (September 7)

SESSION V (Chair: Michael Graham)

9:00-9:25 Susumu Goto (Osaka University)

Application of Reservoir Computing to Turbulence

9:25- 9 :5 0 Kie Okabayashi (Osaka University)

Preliminary Study on Learning and Test Modes of Data-Driven Cavitation Model

9: 50 - 10 : 15 Soshi Kawai (Tohoku University)

Unsupervised Machine-Learning for Super-Resolution and SGS Modeling of Very

Coarse-Grid LES

10: 15 - 10: 35 COFFEE BREAK

SESSION VI (Chair: Taku Nonomura)

10: 35 - 11 : 00 Hiroshi Gotoda (Tokyo University of Science)

Complex-Systems Analysis of Thermoacoustic Combustion Instability in a Swirl-

Stabilized Combustor

11: 00 - 11: 25 Hessam Babaee (University of Pittsburgh)

Reduced-Order Modeling Using Time-Dependent Bases with Applications to

Turbulent Combustion

11:25-11:50 Yoshiaki Abe and Shigeru Obayashi (Tohoku University)

Digital Transformation of Aircraft Design with Carbon Fiber Reinforced Plastics

DISCUSSION

11 : 50 - 12: 10 Moderator: Kunihiko Taira

CLOSING REMARKS

12 : 10 - 12: 20 Kozo Fujii (Tokyo University of Science)

12:20-1:30 LUNCH & AJOURN

REGULAR TALKS (20 minutes + 5 minutes Q&A) Yoshiaki Abe and Shigeru Obayashi (Tohoku University) Digital Transformation of Aircraft Design with Carbon Fiber Reinforced Plastics Digital design of aircraft has been one of the most challenging topics in engineering fields, which requires highly multidisciplinary simulation tools and has yet to be fully realized in industries. This study has developed a multiscale design framework of aircraft wings with carbon fiber reinforced plastics (CFRPs), wherein a multiobjective design exploration of wing geometries were performed using a genetic algorithm. The underpinning tool starts with evaluating mechanical properties of a unidirectional laminate via a microscale analysis of carbon fiber and matrix resin, which are subsequently applied to macroscale analyses of a steady- state fluid-structure-coupled simulation with structural sizing. The framework is able to estimate trade-off solutions of CFRP wing structures based on only the material properties of carbon fiber and matrix resin. The optimization was demonstrated for the aircraft-grade fibers of T700S, T800S, and T1100G, wherein the trade- off solutions achieve the higher performance, i.e., weight and drag reduction, by increasing the stiffness of fibers. The optimized solutions were also analyzed by the proper orthogonal decomposition (POD) to clarify the effects of planform on aerodynamic loads and structural sizing results. The POD successfully identified geometrical modes that correspond to three characteristic planforms in optimized solutions. Furthermore, some preliminary results on data-driven approach for unsteady fluid-structure interaction problems are presented. Byungjin An (Ebara Corporation) Data Science Application in Ebara Technology development to effectively control unsteady flows is an important issue related to our products in terms of implementing new functions and value-added improvement. In order to address the issue, it is necessary to investigate the characteristics of flow and understand them correctly, and theoretical analysis, experiments, and numerical analysis approaches have been carried out conventionally. In recent years, attempts have been actively carried out to apply data science in addition to the conventional methods. We present examples of new approaches to applying data science and discuss the effects of data science on our product and future application development. Hessam Babaee (University of Pittsburgh) Reduced-Order Modeling using Time-Dependent Bases with Applications to Turbulent Combustion Many important problems in fluid mechanics are described by high-dimensional partial differential equations (PDEs). The computational cost of solving these problems using classical discretization techniques increases exponentially with respect to the number of dimensions –– a fundamental challenge that is dubbed the curse of dimensionality. On the other hand, many of these high-dimensional problems have a much lower intrinsic dimensionality, that if discovered, can mitigate the curse of dimensionality. This calls for techniques that extract and exploit correlated structures directly from the PDE. This approach is in direct contrast to classical discretization techniques that disregard multidimensional correlations and result in inefficient solutions for high-dimensional problems. While there are numerous data-driven dimension reduction techniques that can extract these correlated structures by solving the full-dimensional PDE, these techniques are only feasible for lower-dimensional PDEs (e.g., 2D/3D). This same workflow is impracticable for many high-dimensional

PDEs as computing the solution of the full-dimensional PDE is the very problem we cannot afford to solve. To this end, we present a reduced-order modeling framework, in which the correlated structures are extracted directly from the PDE –– bypassing the need to generate data. These structures are exploited by building on- the-fly reduced-order models (ROM). The correlated structures are represented by a set of time-dependent orthonormal bases and their evolution is prescribed by the physics of the problem. We present several demonstration cases including reduced-order modeling of reactive species transport equation in turbulent combustion as well as sensitivity analysis and uncertainty quantification in fluid dynamics problems. Jiro Doke (MathWorks) MATLAB Ecosystem: Teaching Tool in the Field of Data Science Data science is a field that has become important not just for engineering and science students, but also for students from a broader range of disciplines. In addition, computational thinking, which is one of the foundation skills needed in data science, has been a critical component in education for many years. The MATLAB ecosystem provides tools to help incorporate computational thinking into the classroom. In this session, I will introduce (relatively) recent MATLAB features that can be used in the classroom to enhance the learning experience for the students and to foster computational thinking skills. Live scripts allow instructors to prepare an executable lecture note, that can be further extended by individual students. Interactive controls in live scripts allow students to explore concepts. More sophisticated interactivity can be accomplished through apps. During the COVID pandemic, many instructors have made use of apps to create virtual labs to replace or supplement physical labs. Throughout the session, I will perform in-product demonstrations as well as introduce ready-made educational content provided by MathWorks and the user community. Koji Fukagata (Keio University) Reduced Order Modeling and Estimation of Flow Fields Using Convolutional Neural Networks – Towards Machine Learning Assisted Flow Control Application of machine learning is currently one of the hottest topics in the fluid mechanics field. While machine learning seems to have a great possibility, its limitations should also be clarified. In our group, we have started a research project to construct a nonlinear feature extraction method by applying machine learning technology to “turbulence big data,” extracting the nonlinear modes essential to the regeneration mechanism of turbulence and deriving the time evolution equation of those nonlinear modes. In this presentation, we will introduce some examples on learning and regeneration of temporal evolution of cross- sectional velocity field in a turbulent channel flow using convolutional neural network (CNN). We will also introduce the application of CNN for super-resolution analysis and extraction of low-dimensional nonlinear modes for flow around a bluff body accompanying vortex shedding. We also introduce our attempts to interpret the nonlinear modes extracted by CNN autoencoder and to use them for an advanced design of flow control, as well as an attempt for uncertainty quantification and applications to experimental data. This work was supported by KAKENHI KIBAN (A) (FY2018-2020, No. 18H03758) "Construction of feature extraction method for turbulence big data by machine learning" and JSPS KAKENHI KIBAN (S) (FY2021- 2025, No. 21H05007) "Creation and implementation of an innovative flow control paradigm utilizing machine learning" by Japan Society for the Promotion of Science. Kai Fukami (UCLA) Reconstructing and Modeling Unsteady Flows with Physics-Inspired Machine Learning Recent advances in numerical and experimental technologies facilitate access to massive spatio-temporal high- resolution fluid flow data. Machine learning has been recently recognized as a powerful tool to compress such rich

et al., Phys. Fluids 33, 064108 (2021)]. In this workshop, we present the nonlinear dynamics of thermoacoustic combustion oscillations in a swirl-stabilized combustor from the viewpoints of symbolic dynamics and complex networks. Michael David Graham (University of Wisconsin-Madison) Data-Driven Manifold Dynamics for Modeling and Control The success of many machine learning applications is often attributed to the "manifold hypothesis", the idea that many nominally very high-dimensional data sets actually reside on or near a manifold of much lower dimension within the ambient space. In applications such as fluid mechanics that are governed by dissipative PDEs we expect this hypothesis to be strictly valid, as dissipation smooths out small scales leading the long- time dynamics to lie on a finite-dimensional invariant manifold sometimes called the inertial manifold. We describe a data-driven reduced order modeling method that (1) estimates manifold dimension and determines a coordinate representation of the manifold using an autoencoder, and (2) learns an ODE describing the dynamics in these coordinates, using the so-called neural ODE framework. With the ODE representation, data can be widely spaced and no time derivatives of data are required. We apply this framework to chaotic bursting dynamics of Kolmogorov flow and transitional turbulence in plane Couette flow, finding dramatic dimension reduction while still yielding good predictions of short-time trajectories and long-time statistics. This method enables highly efficient and effective design of reinforcement learning control algorithms, as we illustrate with turbulent Couette flow. An important extension of this approach emerges from the recognition that for a general manifold, no single intrinsic global Cartesian coordinate representation can be found. In the language of topology an "atlas" of overlapping local coordinate representations, or "charts", must be used. We use this framework to represent nonlinear dynamics of dissipative PDEs on manifolds of intrinsic dimension. Traian Iliescu (Virginia Tech) ROM Closures and Stabilizations for Turbulent Flows In this talk, I will survey reduced order model (ROM) closures and stabilizations for under-resolved turbulent flows. Over the past decade, several closure and stabilization strategies have been developed to tackle the ROM inaccuracy in the convection-dominated, under-resolved regime, i.e., when the number of degrees of freedom is too small to capture the complex underlying dynamics. I will present regularized ROMs, which are stabilizations that employ spatial filtering to alleviate the spurious numerical oscillations generally produced by standard ROMs in the convection-dominated, under-resolved regime. I will also survey three classes of ROM closures, i.e., correction terms that increase the ROM accuracy: (i) functional closures, which are based on physical insight; (ii) structural closures, which are developed by using mathematical arguments; and (iii) data- driven closures, which leverage available data. Throughout my talk, I will highlight the impact made by data on classical numerical methods over the past decade. I will also emphasize the role played by physical constraints in data-driven modeling of ROM closures and stabilizations. Michio Inoue (MathWorks) MATLAB Ecosystem: Data Science with Open Source and Beyond Why MATLAB for data science? One answer is its the ecosystem. With the recent updates with the enhanced capabilities, MATLAB streamlines the workflow that includes data-centric preprocessing, model tuning, model compression, model integration, and automatic code generation, even with models developed outside of MATLAB, such as TensorFlow and PyTorch. In particular, easy-to-use apps in data preparation come in handy

to support the preparation and labeling of data, including image, video, and audio formats. In this session, I will describe the above features and will also highlight how to leverage the public cloud platforms like Microsoft Azure and Amazon Web Services (AWS) to speed up your data analytics, training, simulation, and deployment workflows. I will demonstrate how you can take full advantage of the ecosystem to do data science. Anya Jones (University of Maryland) Towards New Methods of Lift Regularization in Discrete Gusts Understanding and mitigating a wing’s response to gusts and unsteady winds has been a topic of much research over the past decade, but it has only recently been demonstrated in the laboratory that given real-time lift measurements, a closed loop controller can be used to regularize lift transients using pitch actuation. However, modern optimal and robust control techniques are limited by the quality of the model for the system dynamics. Furthermore, in practical applications it is often the case that force measurements are not available, and controllers must instead act on inertial or flow measurements. These challenges present an opportunity to take advantage of recent advances in machine learning and data science to make progress in the discipline of controlling unsteady aerodynamics systems. The UMD gust-enabled towing tank is currently undergoing renovation to make it possible to explore non-traditional control methods for gust alleviation, including reinforcement learning. The model control system will be able to use arbitrary sensor input to explore a set of actions, while exploiting knowledge accumulated through many iterations. Initially, the set of actions will consist of all permissible angle of attack histories and the measurements used will be real-time force signals. As the wing is repeatedly towed through a discrete, large-amplitude transverse gust, the learning agent should discover that no pitch input is required before and after the gust encounter, and that pitching down into the gust and then up out of the gust can result in a much lower lift transient. Future experiments will replace force measurements with input from various types of sensors, prompting the learning agent to discover novel strategies for detecting and mitigating gusts. Importantly, this upgraded experimental facility will make it possible to test different types of algorithms and approaches, facilitating collaboration with other groups with machine learning expertise. Soshi Kawai (Tohoku University) Unsupervised Machine-Learning for Super-Resolution and SGS Modeling of Very Coarse-Grid LES The scale-resolving methods, such as LES, are now essentially replacing the RANS in the academic turbulence research and have increased attention to the industry. For example, thanks to a high-fidelity numerical scheme and wall modeling we developed, we have achieved the LES of full aircraft configurations using up to 50Bn grid points on the supercomputer Fugaku. This study tries to develop the very coarse-grid LES modeling, one of the challenging topics in LES, using machine-learning-based super-resolution to reduce the required grid points for the LES drastically and further accelerate the use of LES in the industry. The key issue here is that the very coarse-grid LES flowfield is not equivalent to the filtered DNS flowfield. In this study, we develop an unsupervised machine-learning (CycleGAN)-based method for the super-resolution reconstruction of the coarse LES flowfield to the DNS quality flowfield to evaluate the SGS stress components for the very coarse- grid LES.

Fortunato Nucera (Honda R&D) GAN-Based Automobile Flow Field Prediction from Simple Geometric Data The exciting improvement in hardware performances (specifically GPU and TPU) and the rising demand for fast prediction and generative systems has triggered, in the past decade, a renovated interest in the field of deep neural networks and has led to the development of Generative Adversarial Networks (GAN) in 2014. GANs have therefore been deployed on tasks related to super resolution, Image-to-Image translation, and even hyper-realistic deepfakes. At the same time, due in part to the advent of EV vehicles, the automotive industry has become more competitive and fast design and engineering are now essential. It is in this framework that several attempts have been made to switch from aerodynamic simulation through CFD to prediction via neural networks. In literature, the typical flow field prediction system relies on some geometric preparation (commonly, voxelization) and an approximation model - typically very deep Convolutional Neural Networks (CNNs) - with a final Mean-Squared-Error (MSE) loss on the target data. We argue that this fully data-driven approach is insufficient, as it purely relies on distributional summary statistics, without taking into account the nature of the matter of interest (aerodynamics). In our work, we contend the necessity of the inclusion of the residuals of the Navier-Stokes equations within the loss function in order for the prediction model to pass a CFD expert inspection (the analogue of a visual Turing test for classification). In addition, we employ a traditional Generator/Discriminator GAN architecture where each of the components is a deep CNN. We demonstrate the generalizability of the results through training on a set of morphed spheres of the size of a soccer ball with the associated Large Eddy Simulation (LES) results and prediction on an automobile shape, and draw conclusions regarding the performance of the prediction system and its possible future improvements. Kie Okabayashi (Osaka University) Preliminary Study on Learning and Test Modes of Data-driven Cavitation Model The goal of this study is to develop a "data-driven cavitation model" based on a machine learning model that is trained by the measurement data, rather than a conventional mathematical model, as a breakthrough in the development of cavitation models. We also aim to improve the model by the data assimilation aspect of using measured data. As a preliminary study, this report attempts to develop a framework for a data-driven cavitation model using CFD data calculated by a homogeneous fluid model. The goal in this case is to obtain a data- driven cavitation model that reproduces the homogeneous fluid model. The target is cavitating turbulent flow around a Clark-Y11.7% hydrofoil with an angle of attack of 0°. The machine learning model is U-Net, which is a kind of convolutional neural network used for object detection. The training dataset consists of the velocity, pressure, and liquid volume fraction fields of the current step as input data and the liquid volume fraction field of the next step as the training data. The training results show that U-Net is generally able to predict the volume fraction of the liquid of the next step. However, the error from the CFD result is relatively large at the trailing edge of the hydrofoil, where unsteady behaviors such as detached vortices and cloud cavitation occur. Therefore, considering the implementation of the data-driven model into CFD, it is necessary to change the loss function for emphasizing these unsteady phenomena, or to use a machine learning model with higher prediction accuracy. Acknowledgment: This study is financially supported by JSPS KAKENHI No. JP22K03925. This study is partly achieved by the research proposal-based use of large-scale computer systems of Cyber Media Center, Osaka University.

Kunihiko Taira (UCLA) Toward Extreme Aerodynamics with Data-Driven Approaches Data-driven methods are indispensable tools for complex physical problems that are challenging for traditional theoretical, experimental, and computational techniques. This is especially true for complex dynamics without an established model and those that provide an extremely large amount of data over a vastly different set of parameters. The aerodynamics of flying vehicles in extreme levels of disturbances falls into such a category of physical problems due to the strong nonlinearity and enormously large combinations of disturbance parameters to consider. As external disturbances in the atmosphere, in urban canyons, or over mountainous terrains hit flying bodies, the interacting dynamics between the disturbance and the wake can exert transient forces that can render standard flight control approaches ineffective. The present study aims to extract the key dynamics from such aerodynamic scenarios from a large number of numerical simulations using machine learning-based techniques. One of the challenging aspects of extreme aerodynamic problems is the complete lack of theoretical foundation and a unified perspective. In this talk, we will discuss our ongoing efforts to understand extreme aerodynamics and develop novel data-driven approaches to analyze, model, and control the violent behaviors of these turbulent flows. (This work is supported by the Vannevar Bush Faculty Fellowship and the Air Force Office of Scientific Research.) Tomoaki Tatsukawa, Kozo Fujii (Tokyo University of Science) Preliminary Study of Extraction of Speed-Control Strategy in En-Route Air Traffic using Multi- Objective Optimization and Machine Learning Although air traffic demand is temporarily decreased due to COVID-19, it is expected to grow in the following decades, causing the overcapacity of large-scale airports. In this study, we attempts to extract the optimal speed control strategies in response to the target airport and airspace for air traffic controllers. As a case study, this study focuses on the arrival traffics at Tokyo International Airport (RJTT). Combining the rule-based simulator and the multi-objective optimization, the cruise speed of 432 target flights corresponding to the design variables are controlled at 150 NM from RJTT for the purpose of minimizing flight duration of both the cruise flights and the pop-up flights. As the data exploration technique, the decision trees for two main traffic flows are constructed to classify the speed control target. The target variable is the binary value converted from the design variables of non-dominated solutions whereas the explanatory variables are the air traffic information obtained at the moment of the speed control. As a result, the constructed decision trees output the significant features such as the separation and the congestion and the corresponding threshold values varying in accordance with each route cluster. This result suggests that our approach has the potential to help air traffic controllers to automatically select the speed control target regardless of airport and airspace. Aiko Yakeno (Tohoku University) Challenges for Delaying Transition to Reduce Airplane Drag Reducing drag of an aircraft continues to be one of the most important research issues in the field of fluid engineering. Drag reduction technology by "laminarization" is one of the essential technologies for next- generation passenger aircraft, and our laboratory is promoting joint research with multiple companies such as MHI and JAXA. The transition mechanism at the leading edge of the swept main wing of the passenger aircraft has not been clarified so far. In our study, we succeeded in large-scale flow computation at the practical cruising state, and showed that generation of the transitional coherent-modes is explained by the energy transient amplification in a short target time (Yakeno and Obayashi, Physics of Fluids (2021)). We also made innovative proposals that overturn the conventional concept of "surface roughness - > increases drag",