Abstract: The extraction and analysis of ship flow field characteristics have long been a critical research direction in hull form design and ship condition analysis. However, data obtained through numerical simulation methods often remain underutilized, and traditional methods suffer from limited feature representation capabilities. This paper proposes a deep learning-based approach for analyzing the characteristics of pressure time-series data on ship surfaces. A clustering autoencoder model is constructed to perform both feature dimensionality reduction and feature classification. Using the spectral curves of pressure time-series data from propeller blade surfaces as training data, the model extracts low-dimensional features for subsequent classification. This process effectively visualizes the spatial distribution of flow field pressure time-series characteristics and accurately identifies high-frequency measurement points. The findings validate the utility of deep learning in the analysis of ship flow field characteristics and provide a methodological framework for understanding flow field characteristics induced by propeller excitation.

Key words: analysis of ship flow field characteristics, deep learning, flow field pressure time history, clustering autoencoder, feature dimensionality reduction and classification

摘要： 船舶流场特征的提取分析一直是船型设计与船舶状态分析的重要研究方向,然而不仅数值模拟方法获取的数据缺少充分的利用,传统方法的特征表达能力也有限。该文提出了基于深度学习的船舶表面压力时程数据特征分析方法,建立了包含特征降维和特征分类作用的聚类自编码器模型;以螺旋桨桨叶表面压力时程频谱曲线为训练数据,提取低维特征后进行分类,展示了流场压力时程数据特征的空间分布,准确地识别出了高频测点。该研究结果验证了深度学习在船舶流场特征分析方面的价值,同时为掌握螺旋桨激励的流场特征提供了方法支撑。

关键词: 船舶流场特征分析, 深度学习, 流场压力时程, 聚类自编码器, 特征降维与分类