为了识别典型船体结构裂纹损伤，提出基于支持向量机（SVM）的分类方法，采用固有时间尺度分解（ITD）和奇异值分解（SVD）方法对振动信号进行特征向量提取，得到训练样本和验证样本数据，应用SVM算法对训练样本数据进行正确的分类。为评估SVM分类准确性，引入BP神经网络算法进行对比分析，2种算法对验证样本进行分类预测，结果证明SVM方法对小样本试验数据进行模式分类具有更高的准确率。对十字加筋板的损伤模式进行分类研究，可对船体局部结构的维修和管理提供更好辅助决策支持。
In order to identify the crack damage of the typical hull structure, the classification method based on support vector machine (SVM) was put forward, the intrinsic time-scale decomposition (ITD) and singular value decomposition (SVD) method was carried out on the vibration signal to extract feature vector and the training sample and verified sample data were acquired. Classification of the training sample data by the SVM algorithm was carried out correctly. In order to evaluate the accuracy of classification method based on SVM algorithm, BP neural network algorithm is introduced for comparative analysis. The two algorithms are used for classification prediction of verified sample data. The results prove that SVM algorithm has higher accuracy in mode classification of small sample test data. The research of classification of damage modes of cross stiffened plates can provide better decision support for the maintenance and management of local hull structures.
2021,43(9): 60-63 收稿日期：2019-07-02
DOI：10.3404/j.issn.1672-7649.2021.09.011
分类号：U663.2
基金项目：国家自然科学基金资助项目(51779261)，海军工程大学自然科学基金资助项目(20161595)
作者简介：张仲良(1994-)，男，硕士研究生，研究方向为船舶与海洋结构物设计制造
参考文献：
[1] 潘明清. 基于支持向量机的机械故障模式分类研究[D]. 杭州：浙江大学, 2005.
[2] 高伟, 陈伟凡, 杨耿杰, 等. 基于奇异值分解和多级支持向量机的配电网故障类型识别[J]. 电子测量与仪器学报, 2018(2): 62–71
[3] 李辉, 李欣同, 贾嵘. 基于经验模态分解和支持向量机的水电机组振动故障诊断[J]. 水力发电学报, 2016(12): 105–111
[4] 陈勇旗, 赵一鸣, 陈杨. 基于固有时间尺度分解的滚动轴承故障诊断[J]. 电子测量与仪器学报, 2015, 29(11): 1677–1682
[5] 邢占强. 基于非平稳信号时频分析的故障诊断及应用[D]. 重庆：重庆大学, 2017.
[6] 牛海清, 吴炬卓, 郭少锋. 奇异值分解在电缆局部放电信号模式识别中的应用[J]. 华南理工大学学报(自然科学版), 2018, 46; 376(1): 32–38
[7] 船体结构疲劳强度指南[S]. 上海: 中国船级社, 2014.
[8] MEYER D, WIEN F H T. Support vector machines[J]. The Interface to libsvm in package e1071, 2015
[9] 安学利, 蒋东翔, 陈杰, 等. 基于ITD和LS-SVM的风力发电机组轴承故障诊断[J]. 电力自动化设备, 2011, 31(9): 10–13
