Water Sports: Hyper Hydration (2023)

Water Sports: Hyper Hydration is a foray into machine learning involving optimization algorithms. It places a tubular geometry, representing a water bottle at the center of the inquiry. Firstly, single objective optimization for the 3D profile that empties itself the fastest and then mulit-objective optimization by balancing emptying speed and surface area. This project was developed as part of the Computational Explorations seminar at ITECH master's program in collaboration with Chris Kang, Markus Renner, and Cornerlius Carl. The code and studies for this project can be found on GitHub.

The project's initial phase involved implementing various algorithms encompassing both single and multi-objective optimization techniques to efficiently optimize the shape of a water bottle. The project aimed to develop a water bottle model with fixed top and bottom sizes while allowing for a variable profile along the entire circumference. The primary objective was identifying optimal parameters for efficient water release, minimizing the time required for water to exit.

The single-objective optimization focused on maximizing water expulsion speed from the bottle. The multi-objective optimization strategy introduced an additional factor - the size of the exit cap. The aim was to strike a balance between water release speed and minimizing the cap size, a challenge due to the conflicting nature of these two objectives.

The workflow utilized a Python library to generate NURBS surfaces exported as STL files. These were processed by a Blender script to combine static bodies representing the cap and water storage. An alternative Python library was later used to address seam issues. Post mesh generation, FluidX3d, a modified CFD simulation, used STL files as input. Adjustments were made to the simulation's source code using C++ for dynamic data exchange. A Python script managed the process, leveraging algorithms to trigger simulation runs.

The bottle's design utilized 5 circles with 6 control points each, totaling 30 parameters. Control points allowed flexible shape creation along a line between the central point and a radius of 1. Static bodies were consistently incorporated at both ends of the bottle, serving distinct purposes. The first housed water storage, while the final acted as the cap for controlled water release.

Initial observations from single-objective optimization were inconclusive regarding curvature patterns. Throughout experimentation, the rbfopt algorithm outperformed others in speed and robustness, demonstrating advantages in optimization efficiency. Optimized single-objective designs aimed to enhance drainage efficiency through vortex creation, achieving an 11% improvement compared to a simple cylinder. Further analysis was needed to understand these improvements.

Single-Objective Convergence & Robustness

In the context of multi-objective optimization, it was observed that both NSGA2 and NSGA3 yielded similar outputs. However, NSGA2 outperformed other algorithms in terms of convergence and objective achievement. Unfortunately, the extensive simulation time of approximately 12 hours per run posed a challenge in evaluating algorithm robustness.

Multi-Objective Convergence & Robustness

In contrast to single-objective optimization results, the outcomes of multi-objective optimization displayed distinct differences in optimized objectives and overall bottle shape. Unlike single-objective optimization, which focused on specific improvements, multi-objective optimization considered multiple objectives simultaneously.

Clustering

The optimization process explores a broader range of design possibilities and objectives, leading to a greater diversity in the optimized bottle shapes and objectives achieved. The multi-objective optimization aims to strike a balance between conflicting objectives, such as maximizing drainage efficiency while minimizing simulation steps or achieving an optimal shape while considering stability or other factors.

Results (in simulation steps)

Unoptimized: 17,396	RBFOpt: 15,849	CMAE-S: 15,863	Random: 16,004

The diverse objectives and inherent trade-offs in multi-objective optimization generated a broader spectrum of optimized bottle shapes and objectives. This diversity allowed for a comprehensive exploration of the design space, enabling decision-makers to select solutions aligned with their specific needs and priorities