Adapting Semi-Empirical Ship Vibration Analysis: A Hybrid ML Approach to Generalized Vibration Prediction

In marine engineering, ship vibration analysis is crucial for ensuring structural integrity, operational safety, and environmental sustainability. Traditional analysis, following classical paradigms established by early contributors such as Todd, Kumai, and Schlick, relies primarily on costly simulations and empirical tests. This study seeks to overcome these limitations by integrating machine learning (ML) methodologies with semi-empirical models to develop a predictive hybrid model, thereby advancing vibration analysis toward a data-driven paradigm. The research is significant for improving ship design, mitigating vibration-related risks, and reducing reliance on resource-intensive approaches, aligning with global efforts to promote energy-efficient and sustainable maritime operations. The proposed hybrid model combines Random Forest (RF) and Logistic Regression (LR), leveraging RF’s capacity for modeling nonlinear relationships and LR’s interpretability for linear adjustments. Trained on Kumai’s seminal dataset and validated on ۳۷۳ cases spanning ۳۴ ship types, the model accurately predicts critical parameters (α, τ₂, N₂, N₃, and c̄) with exceptional precision. Performance metrics demonstrate strong results, including near-perfect R² values (۰.۹۹۳۸ for α) and minimal MSE (۰.۰۰۰۰ for α, ۰.۰۷۰۱ for N₃). Natural frequency predictions exhibit less than ۳% error, as validated against empirical data for crude oil tankers. Feature importance analysis identifies structural parameters (length, displacement, block coefficient) as key predictors, enhancing interpretability for engineering applications. This work bridges the gap between classical vibration theory and modern ML, offering a cost-effective, scalable alternative to conventional simulations. By enabling precise vibration predictions across diverse vessels, the model facilitates predictive maintenance, design optimization, and operational safety. The findings highlight the transformative potential of hybrid ML in maritime engineering, paving the way for digital twins and sustainability-driven ship design.

کلیدواژه ها:

Hybrid Machine Learning ، Random Forest ، Logistic Regression ، Ship Vibrations ، Predictive Maintenance