Control, Modeling, Identification, and Applications (CoDAlab), Department of Mathematics, Escola d'Enginyeria de Barcelona Est (EEBE), Universitat Politècnica de Catalunya (UPC), Campus Diagonal-Besòs (CDB), Eduard Maristany, 16, 08019 Barcelona, Spain.
Sensors (Basel). 2020 Mar 19;20(6):1716. doi: 10.3390/s20061716.
In this paper, we evaluate the performance of the so-called parametric -distributed stochastic neighbor embedding (P--SNE), comparing it to the performance of the -SNE, the non-parametric version. The methodology used in this study is introduced for the detection and classification of structural changes in the field of structural health monitoring. This method is based on the combination of principal component analysis (PCA) and P--SNE, and it is applied to an experimental case study of an aluminum plate with four piezoelectric transducers. The basic steps of the detection and classification process are: (i) the raw data are scaled using mean-centered group scaling and then PCA is applied to reduce its dimensionality; (ii) P--SNE is applied to represent the scaled and reduced data as 2-dimensional points, defining a cluster for each structural state; and (iii) the current structure to be diagnosed is associated with a cluster employing two strategies: (a) majority voting; and (b) the sum of the inverse distances. The results in the frequency domain manifest the strong performance of P--SNE, which is comparable to the performance of -SNE but outperforms -SNE in terms of computational cost and runtime. When the method is based on P--SNE, the overall accuracy fluctuates between 99.5% and 99.75%.
在本文中，我们评估了所谓的参数分布随机近邻嵌入（P-SNE）的性能，将其与非参数版本的-SNE 进行了比较。本研究中使用的方法是为了在结构健康监测领域检测和分类结构变化而引入的。该方法基于主成分分析（PCA）和 P-SNE 的组合，并应用于带有四个压电换能器的铝板的实验案例研究。检测和分类过程的基本步骤是：（i）使用均值中心化组缩放对原始数据进行缩放，然后应用 PCA 降低其维度；（ii）应用 P-SNE 将缩放和降维后的数据表示为 2 维点，为每个结构状态定义一个簇；（iii）将待诊断的当前结构与使用两种策略的簇相关联：（a）多数投票；（b）逆距离的总和。频域中的结果显示了 P-SNE 的强大性能，其性能可与-SNE 相媲美，但在计算成本和运行时间方面优于-SNE。当该方法基于 P-SNE 时，整体准确性在 99.5%到 99.75%之间波动。
