A reduced nonstationary discrete convolution kernel for multimode process monitoring
A reduced nonstationary discrete convolution kernel for multimode process monitoring
复制标题
DOI：
10.1007/s13042-022-01621-8
复制
发表时间：
2022-08
影响因子：
5.6
通讯作者：
Kai Wang;Caoyin Yan;Xiaofeng Yuan;Yalin Wang;Chenliang Liu
中科院分区：
文献类型：
--
作者：
Kai Wang;Caoyin Yan;Xiaofeng Yuan;Yalin Wang;Chenliang Liu
多模态行为在工业过程中是普遍存在的。由于多模态数据分布可以被看作是一种特殊的非线性，核方法在构建多模态过程监控模型方面是经验有效的。然而，核方法在收集大量数据时具有较高的复杂度。为了提高多模态数据下的故障检测性能，降低计算复杂度，本文借鉴径向基函数（RBF）神经网络的结构设计，提出了一种简化的非平稳离散卷积核，作为RBF核和非平稳离散卷积核的替代.通过删除NSDC核中不必要的累积项，在保证故障检测性能的前提下，有效降低了NSDC核算法的计算复杂度，加快了故障检测的速度。在标准核主元分析框架下，通过数值算例和多峰TE过程验证了该算法的有效性。
The multimodal behavior is common in industrial process. Since multimodal data distribution can be regarded as a special kind of nonlinearity, kernel method is empirically effective in constructing the multimode process monitoring model. However, kernel methods suffer its high complexity when a large number of data are collected. In order to improve the fault detection performance in multimodal data and reduce the computational complexity, we propose a reduced nonstationary discrete convolution kernel which is inspired by the structural design of radial basis function (RBF) neural network, as an alternative to the RBF kernel and the nonstationary discrete convolution (NSDC) kernel. By deleting the unnecessary accumulated terms in the NSDC kernel, the computational complexity of the proposed NSDC kernel algorithm is effectively reduced and the speed of fault detection is accelerated on the premise of ensuring the fault detection performance. The effectiveness of the proposed algorithm is demonstrated on a numerical example and multimodal TE process under the standard kernel principal component analysis framework.