Research interests

Machine Learning/Artificial Intelligence-Assisted Design Optimization

Design technologies have constantly been evolving from year to year. Engineering design has always been a wonder; it constantly changes the landscape of humanity and civilization. In the age of data, the roles of machine learning (ML) and artificial intelligence (AI) are highly instrumental. Let us now venture into the era of data-driven and data-centric engineering design. Through the aid of statistics, ML, and AI, my research focuses on leveraging such technologies to improve the process of design optimization and exploration. In particular, I have been focusing on developing Gaussian Process and Neural Networks-based methods to handle small/medium and big data, respectively.

In addition, I also have an interest in how machine and human knowledge and intuition can collaborate to efficiently design engineering systems, which is why I am now pursuing research on interpretable and explainable machine learning. The important question is, "what can an ML/AI system tell me more than just prediction? Can a machine tell me important design insight and guidelines?". My main objective is to design technologies that can be applied to design and optimize mechanical/aerospace systems intelligently. However, I also have great interest and experience in applying ML/AI methods for other domains of engineering, including biomedical, marine, and civil engineering.

Let's see one example in the figure below. Supersonic commercial past seems to be a thing of the past, but it will not be so in the future. Various universities, companies, and research institutions worldwide have constantly been researching novel shapes and technologies to realize commercial supersonic flight again. Collaborating with Tohoku University, which was in charge of the problem definitions and computer simulations, we applied Gaussian Process Regression and Bayesian Optimization to find optimal supersonic wing platforms that simultaneously minimize drag, noise, and root bending moment. Further post-processing using unsupervised neural networks was then performed to extract important design insight. The knowledge obtained from this research will be useful to researchers and designers in supersonic aircraft design. In our case, ML/AI were the principal enabling technologies; they greatly helped us evolve the supersonic wing designs.

Figures reference: Jim, Timothy MS, Ghifari A. Faza, Pramudita S. Palar, and Koji Shimoyama. "A multiobjective surrogate-assisted optimisation and exploration of low-boom supersonic transport planforms." Aerospace Science and Technology 128 (2022): 107747.

Probabilistic System Analysis

Every single thing in this world is uncertain, and so is an engineering system. Therefore, quantifying the impact of uncertainties on engineering systems is profound to reduce variability and increase safety. Take an aircraft as an example. There are multiple sources of uncertainty, from manufacturing errors to uncertain flight conditions, not to mention uncertainties in the computer model itself. My research focuses on the uncertainty quantification of engineering systems by leveraging advancements in statistics and machine learning. Further, I am interested in how interpretable machine learning can aid humans in investigating the internal mechanism of how uncertainties affect any system. Finally, sensitivity analysis is important to understand the contribution of multiple input parameters to the system. Such an endeavor greatly helps us in variable screening, that is, to understand to rank the sources of uncertainty according to their contribution to the variability.

The figure below shows an example of how to propagate the uncertainties in the case of a transonic airfoil with uncertain geometry. The ultimate goal is to understand the impact of geometrical errors on the drag force. To that end, we utilize an uncertainty quantification called polynomial chaos expansion combined with sparse grids and computational fluid dynamics to quantify the output uncertainty. The process also yields the so-called Sobol indices, which informs us which variables are the most important contributor to the variable. Such a procedure can greatly aid us in reducing variability by identifying the most impactful variables to design better and safer systems in the future. We care about your safety! That's why.

Figure reference: Palar, Pramudita Satria, Lavi Rizki Zuhal, Koji Shimoyama, and Takeshi Tsuchiya. "Global sensitivity analysis via multi-fidelity polynomial chaos expansion." Reliability Engineering & System Safety 170 (2018): 175-190.

Optimization in Fluid Dynamics

How to reduce the drag force? How to increase the thrust of a robotic fish? How to control the flow around aerodynamic/hydrodynamic bodies to improve efficiency? Such questions can be answered with the help of numerical optimization techniques. I am greatly interested in how we can optimize fluid systems or aerodynamic/hydrodynamic bodies to improve performance, which eventually will benefit the development of engineering products or systems. Besides, I am constantly asking questions such as: "Why is such a design optimal?", "Why are fishes so efficient in generating thrust?", "Why can a hummingbird easily switch flying modes with ease?" Answering the "Why" is as important as the optimization process itself. Therefore, optimization is just the first step; the next step is to inquire into and shred the veil of the mystery of nature. Through the grandeur process of evolution, nature has generated highly efficient swimmers and flyers. Therefore, I am highly interested in biomimetics, that is, how we can apply biological concepts to design highly performing artificial systems that mimic natural animals.

The figure below illustrates one of my recent works with my colleagues that investigates the optimal flapping fish fin for maximum thrust production. Using experimental apparatuses and ML-based techniques, we let our instruments seek the optimal combination of stiffness and flapping frequency to maximize thrust. After finding the optimal combination, we thoroughly analyzed the relationship between the kinematics, stiffness, and thrust production to answer why such a combination is optimal.