ABSTRACT Ring-stiffened circular cylindrical shells are commonly used in many engineering applications. In the present study, the strength and stability characteristics of ring-stiffened circular cylindrical shells are investigated using self-organising map (SOM) technique and multivariate approaches. Stress and critical pressure analysis of ring-stiffened circular cylindrical shells under hydrostatic pressure are obtained from the design guide. Design of experiment (DOE) technique is employed for generating design samples. The information behind these design samples can be studied by the self-organising map (SOM) approach. Multivariate analysis is adopted for post- processing of SOM's results. The clustering of samples can be obtained from the codebook of SOM with the help of multivariate analysis. The samples in different clusters would have their own characteristics. The variable significance in each cluster is examined. The dimensionality reduction approach is used for finding the structure of the data set. A useful visualisation approach, parallel coordinate plot with clustering information, is proposed for visualisation of design samples. Correlations and clustering of variables are also studied by the hierarchical method. The ranking of samples is conducted by minimising the distance to an ‘ideal’ solution in the data set. A final compromises sample is obtained. A numerical example is provided for the illustration of the present analysis.

中文翻译：

---

使用自组织图 (SOM) 和数据挖掘分析对环形加强圆柱壳进行强度和稳定性研究

摘要 环形加强筋圆柱壳常用于许多工程应用中。在本研究中，使用自组织映射 (SOM) 技术和多元方法研究了环形加强圆柱壳的强度和稳定性特性。静水压力下环形加筋圆柱壳的应力和临界压力分析可从设计指南中获得。实验设计 (DOE) 技术用于生成设计样本。这些设计样本背后的信息可以通过自组织映射 (SOM) 方法进行研究。SOM结果的后处理采用多变量分析。借助多元分析，可以从SOM的码本中获得样本的聚类。不同聚类中的样本会有自己的特征。检查每个集群中的变量显着性。降维方法用于查找数据集的结构。提出了一种有用的可视化方法，即带有聚类信息的平行坐标图，用于设计样本的可视化。变量的相关性和聚类也通过分层方法进行研究。样本的排序是通过最小化到数据集中“理想”解决方案的距离来进行的。获得最终妥协样本。提供了一个数值例子来说明当前的分析。提出了一种有用的可视化方法，即带有聚类信息的平行坐标图，用于设计样本的可视化。变量的相关性和聚类也通过分层方法进行研究。样本的排序是通过最小化到数据集中“理想”解决方案的距离来进行的。获得最终妥协样本。提供了一个数值例子来说明当前的分析。提出了一种有用的可视化方法，即带有聚类信息的平行坐标图，用于设计样本的可视化。变量的相关性和聚类也通过分层方法进行研究。样本的排序是通过最小化到数据集中“理想”解决方案的距离来进行的。获得最终妥协样本。提供了一个数值例子来说明当前的分析。