Vibration Frequency Prediction in Marine Vessels: A Synthetic Data Approach with Applications to Noise and Structural Integrity

Abstract

Accurate prediction of vibration frequencies in marine vessels is significant for stopping noise pollution and preventing structural failures, yet traditional methods rely on limited datasets and costly simulations. This study presents a machine learning framework that develops scarce real-world ship data (27 tankers, 43 cargo vessels, and 16 passenger vessels) with physics-informed synthetic samples to predict two-node vertical vibration frequencies (N2NV). By training hybrid datasets in four models (Linear Regression, SVR, Neural Networks, and Random Forest), we demonstrate that synthetic data improves prediction accuracy by 22% (Neural Networks achieve R²=0.997 vs. 0.816 with real data alone). Our analysis reveals strong correlations between predicted vibrations and (1) noise generation in the 100-150 Hz critical range, and (2) deformation patterns in hull monitoring data. Synthetic data approach reduces computational costs by 40% compared to finite element analysis while enabling early detection of resonant frequencies that accelerate structural fatigue. These results establish vibration frequency prediction as an outlet technology for integrated noise control and structural health monitoring systems in maritime applications.