Neural topology optimization of ship structures under propulsion machinery vibrations

Problem

Dynamic-compliance topology optimization often leads to suboptimal designs that exhibit pathological characteristics near resonance frequencies. This paper addresses this gap by proposing a neural network-based approach to optimize ship structures specifically under the influence of propulsion machinery vibrations. The work is presented as a preprint and has not undergone peer review.

Method

The authors introduce a Convolutional Kolmogorov-Arnold network (KATO) as the core architecture for their optimization framework. The primary objective of the optimization is to minimize active input power (AIP) in ship structures, particularly focusing on two applications: a 100 Hz engine-supporting deck panel and an 18 Hz thruster foundation frame. The methodology incorporates Helmholtz partial differential equation (PDE) filtering and Heaviside projection techniques to ensure appropriate feature sizes and manufacturing tolerances are maintained during the optimization process. For comparison, the authors utilize the Generalized Method of Moving Asymptotes (GCMMA) as a baseline optimization method. Performance metrics evaluated include AIP, static compliance, and binary static compliance.

Results

The results demonstrate significant improvements over the baseline methods:

• AIP is reduced by more than 32 dB compared to the initial AIP.
• Static compliance is reduced by a factor of 22-36x when compared to matched-volume binary re-analysis.
• Binary static compliance shows a reduction of 59x relative to GCMMA.
• The KATO architecture exhibits a speed improvement, running 6.4-10.4x faster than GCMMA.

Limitations

The authors do not report any limitations in their study. However, the absence of peer review may imply that the findings should be interpreted with caution until validated by the community.

Why it matters

This work has significant implications for the design of ship structures, particularly in enhancing their resilience to vibrations from propulsion systems. The proposed neural topology optimization method could lead to more efficient designs that not only reduce vibrations but also improve overall structural performance. This approach may pave the way for further research into neural network applications in engineering design, particularly in fields where dynamic compliance is critical.