The research paper from Data Chemical Engineering Laboratory (headed by Associate Professor KANEKO Hiromasa) was featured in the cover illustration of the international academic journal Industrial & Engineering Chemistry Research, Volume 64, Issue 7 (2025).
Computational Fluid Dynamics (CFD) simulation is used to design devices and processes. In CFD simulation, parameters such as device shapes and operating conditions (x) are varied, and the results (y) are assessed. This cycle is repeated to design an x that yields a desirable y. The machine learning mathematical model, y = f(x), can be leveraged to design x more efficiently, in other words, to optimize x with fewer CFD simulations. However, conducting a comprehensive search across all conditions in the multi-dimensional space of x is challenging, frequently resulting in overlooked information or local solutions. This research developed a “direct inverse analysis method for the models” that predicts x directly from the target value of y during CFD simulation, achieving a comprehensive search of simulation results. This enabled the optimization of x in ways that were not possible with conventional methods. With the proposed method, the rational design of x was achieved based on the CFD simulation results, demonstrating a high level of consistency with actual outcomes.
