基于 SVM 的双层圆柱壳体机械振动噪声数据特征提取方法比较

鉴于某双层圆柱壳体的机械振动噪声数据结构复杂、维数较高，工程上不宜直接分析，文章提出先对其进行特征提取后再进一步分析的思路，可有效简化数据结构，提高数据分析的准确度。选择工程上常用的主成分分析法（PCA）、核主成分分析法（KPCA）与独立成分分析法（ICA）对文中高维机械振动噪声数据进行特征提取。利用支持向量机（SVM）的分类识别能力，对经特征提取后不同工况下的噪声数据进行分类识别。依据正确识别率大小比较三种方法的特征提取效果，以选择针对某双层圆柱壳体机械振动噪声数据合适的特征提取方法。结论可为深入分析某双层圆柱壳体机械振动噪声数据的规律特点打下良好基础。

Comparison of feature extraction methods on Mechanical Vibrating Noise of ribbed cylindrical double-shells based SVM

Mechanical vibrating noise data structure of ribbed cylindrical double-shells is complex and mul-ti-dimensional, so it is not to be dealt directly. An improving approach to this problem is reducing dimen-sion firstly and then analysis. Some common methods in engineering, such as PCA (Principal Component Analysis), KPCA (Kernel Based Principal Component Analysis) and ICA (Independent Component Analy-sis) are dealt with those mechanical vibrating noise data. After the feature extracting, the vibrating noise da-ta which come from different working-modes are classified by using to the classified ability of SVM (Sup-port Vector Machine). Comparing the effects of those methods according to the recognition percentage, the right method to ribbed cylindrical double-shells is chosen. The method will be an important groundwork for researching the characteristic of mechanical vibrating noise data in future.