Transforming Ship Structures: Neural Optimization for Reduced Vibration Using Active Input Power

Recent advancements in computational design are about to revolutionize how ships are constructed to combat the forces of propulsion machinery vibrations. A groundbreaking research paper by Shengyu Yan, Muhammad Muztahidul Hakim Zareer, and Jasmin Jelovica introduces a novel approach that harnesses neural networks for topology optimization, specifically aimed at minimizing active input power (AIP) in critical ship structures.

The Challenge of Vibrations in Ship Design

Ship structural vibrations pose significant challenges, leading to noise, fatigue, and even damage to equipment onboard. These vibrations stem from propulsion and auxiliary machinery that transmit energy through the ship's foundation and deck panels. Traditionally, mitigating these vibrations has relied on techniques like stiffened panels and damping treatments, often dependent on the designer's judgment. However, these conventional methods rarely target specific dynamic frequencies, which can lead to suboptimal designs.

Innovative Neural-Reparameterized Topology Optimization

The authors propose an innovative method called Neural-Reparameterized Topology Optimization (KATO), which employs a convolutional Kolmogorov-Arnold network (cKAN) to tackle forced-vibration design problems. By focusing on minimizing AIP rather than dynamic compliance, the research introduces a more reliable objective that avoids common pitfalls associated with vibration tuning at specific frequencies.

Key Findings: Reductions in Active Input Power

The optimization framework was applied to two test cases: an engine-supporting deck panel operating at 100 Hz and a thruster foundation frame designed for 18 Hz operation. Results showed that KATO-produced designs significantly reduced AIP, outperforming traditional engineering references by notable margins. This advancement not only improves structural performance but also demonstrates the ability to create designs that are more efficient in terms of their material use while boasting improved vibration suppression.

Efficiency and Speed: KATO vs. Traditional Methods

One of the standout features of the KATO framework is its efficiency. The study showcased how the KATO method runs 6.4 to 10.4 times faster than traditional optimization methods like the globally convergent method of moving asymptotes (GCMMA). The neural approach facilitates a quicker evaluation process, making it feasible to undertake extensive analyses without compromising accuracy. This performance gain is particularly valuable in large-scale engineering problems where computational resources are a concern.

A Future of Ship Design

This research signifies a crucial step towards modernizing ship design practices through advanced computational techniques. The ability to effectively minimize AIP while ensuring structural integrity and performance creates significant potential for enhancing the resilience and efficiency of marine vessels. As the maritime industry continues to confront the challenges of environmental regulations and operational efficiency, adopting such cutting-edge methodologies could well define the future of ship design.

Authors: Shengyu Yan, Muhammad Muztahidul Hakim Zareer, Jasmin Jelovica