非线性多工序批生产过程监控：最近组标准化KPCA方法

133 浏览量更新于2024-08-28 收藏 268KB PDF 举报

"基于最近组标准化KPCA的非线性多工序批生产过程监控方法" 本文是一篇研究论文，由Siyuan Fu、Yajun Wang和Bo Li合作完成，发表于辽宁科技大学电子与信息工程学院。研究的主要目标是提出一种针对非线性多工序批生产过程的新型监控方法——基于最近组标准化的核主成分分析（NGS-KPCA）。这种方法旨在处理生产过程中遇到的非线性和非高斯分布特性。文章首先介绍了问题背景，指出在多工序批生产过程中，由于工艺的复杂性和随机性，数据往往呈现出非线性和非高斯分布的特点，这给传统的统计过程控制（SPC）方法带来了挑战。为了解决这些问题，作者提出了一个包含三个主要步骤的方法： 1. **阶段划分**：根据生产操作的不同阶段对过程数据进行划分。这一步旨在识别不同工序间的差异，以便针对性地进行分析和监控。 2. **最近组标准化（NGS）**：在每个阶段内，采用k-means聚类算法生成最优的簇解决方案。然后，利用NGS方法将属于同一簇的数据转化为符合高斯分布的形式。这种方法可以有效地处理由于操作的随机性和不确定性导致的非高斯特性。 3. **核主成分分析（KPCA）**：为了应对非线性问题，论文选择了KPCA作为建模工具。KPCA是一种非线性降维技术，能够揭示数据中的非线性结构，并用于构建每个聚类的监控模型。KPCA通过核函数将数据映射到高维空间，使得在该空间内的线性变换能反映原始数据的非线性关系。实验部分，作者通过一个120吨炼铝炉的过程数据来验证所提方法的有效性。结果显示，NGS-KPCA方法在异常检测和过程监控方面表现优越，能够准确识别出异常情况，提高生产效率和产品质量。总结来说，这篇论文提出的NGS-KPCA方法为非线性多工序批生产过程提供了有效的监控手段，解决了传统方法在处理非线性和非高斯数据时的局限性，对于提升工业生产过程的智能监控水平具有重要意义。

Nearest Group Standardization KPCA Based Monitoring Method for

Nonlinear Multi-operation Batch Process

Siyuan Fu, Yajun Wang, Bo Li

School of Electronic and Information Engineering, Liaoning University of Technology, Jinzhou 121001, China

E-mail: wyj_lg@163.com

Abstract: A nearest group standardization kernel principal component analysis (NGS-KPCA) based monitoring method specifically

designed for nonlinear multi-operation batch process is proposed in this paper. The key feature of the proposed method can deal

with nonlinearity and non-Gaussian characteristics in the process. To start, phase partition is implemented according to operation.

Then, k-means clustering algorithm produces an optimal ensemble clustering solution at each phase. Meanwhile, to deal with

non-Gaussian characteristic produced by randomness and uncertainty of the operation, nearest group standardization (NGS)

method is exploited to make the data belonging to the same cluster obey Gaussian distribution. To deal with nonlinear characteristic,

KPCA is chosen to build monitoring model for each clustering. The proposed method is illustrated through the 120t ladle furnace

(LF) steelmaking process. Simulation results have shown better performance than multiway kernel principal component analysis

(MKPCA).

Key Words: Nonlinear multi-operation batch process, monitoring, NGS-KPCA, non-Gaussian characteristics, 120t ladle furnace

(LF) steelmaking process



1 Introduction

To meet the increasing demand of process efficiency and

safety, process monitoring has recently prompted a lot of

interest due to its practical importance. In the absence of

prior process knowledge, multivariate statistical process

monitoring (MSPM) such as principal component analysis

(PCA) and partial least-squares (PLS) have been actively

investigated [1-3]. It should be noted that the traditional

MSPM approaches require the process data to follow

multivariate Gaussian distribution approximately in order for

deducing their confidence limits. For complex industrial

process with multiple operations, such methods become

inappropriate and may lead to erroneous and costly

monitoring results.

To overcome the shortcomings of MSPM schemes for

nonlinear multi-operation batch process, a number of

approaches have been proposed. Tong et al. [4] proposed a

comprehensive multimode process monitoring scheme in

which mode clustering and mode unfolding are integrated

within an adaptive strategy. Liu [5] modeled a large-scale

nonlinear system containing many operating regions using

adaptive Takagi-Sugeno fuzzy model on PCA subspace.

Whereas the monitoring results more or less depend on and

be restricted to PCA modeling. Alternatively, Fisher

dis-criminant analysis (FDA) is applied to distinguish

between normal and faulty process operations. However, the

initial model are required to build by labeled training data

from both normal and faulty operating conditions [6].

Meanwhile, FDA may not detect the new faults unless they

already exist in the training data set. A localized Fisher

discriminant analysis (LFDA) method has been applied to

continuous chemical process monitoring. The method can

This work is supported by National Natural Science Foundation (NNSF) of

China under Grant 61503169, 51679116; Innovative Talents Project in Liaoning

Province under Grant LR2016057 and Nature Science Foundation of Liaoning

Province under Grant 2015020102.

effectively retain the multi-Gaussianity within normal or

faulty data sets while separating the faulty samples from

normal operation. Yu [7] extended localized Fisher

discriminant analysis (LFDA) method to multiwa y kernel

localized Fisher discriminant analysis (MKLFDA) to capture

the nonlinearity, batch dynamics, and within-class

multimodality. The MLKFDA approach cannot only retain

and capture the multi-Gaussianity within the batch process

data, but also identify the best separating hyperplanes with

data set involving both normal and faulty batches. Whereas

the method is not valid in a larger-scale and more

complicated bioprocess monitoring including several

different operating conditions along with various types o f

faults. Recently, independent component analysis (ICA) is

adopted to decompose independent components (IC) and

applied to detect the faults from non-Gaussian distributed

data [8-10]. ICA imposes independency on latent variables

beyond second-order statistics and thus can extract the

non-Gaussian features of process data. However, the

multimodality of process data is not necessarily reflected by

the negentropy index, which is very common in industrial

processes due to the change of operations. Liu et al. [11]

proposed the moving window kernel PCA for adaptive

monitoring of nonlinear process. However MWPCA

approach will encounter difficulty in the case of only limited

samples available to initialize the monitoring procedure. On

the other hand, Yu and Qin [12] developed a multimode

process monitoring approach based on Finite Gaussian

Mixture Model (FGMM) and Bayesian inference strategy.

However, it assumes that the data in each operating mode

follow multivariate Gaussian distribution.

For several complex industrial processes with

non-linearity and multi-operation conditions, the process data

often show significantly non-Gaussian distribution. Each

batch process runs under different operating conditions. Due

to the complexity of a sequence of operations, the durations of

batch data are obviously unequal. To handle nonlinearity and

non-Gaussian distribution characteristic in multi-operation

Proceedings of the 37th Chinese Control Conference

Jul

25-27, 2018, Wuhan, China

8037

下载后可阅读完整内容，剩余5页未读，立即下载

weixin_38682518

粉丝: 3
资源: 935

非线性多工序批生产过程监控：最近组标准化KPCA方法

kpca非线性方法特征值提取

基于KPCA-ICA的新型非线性过程监控

并行PCA–KPCA用于非线性过程监控

一种基于KPCA的非线性故障诊断方法 (2005年)

20494855sime-kpca.rar_KPCA_核PCA程序 KPCA_非线性PCA

基于FRFT-KPCA的J模拟电路非线性故障特征提取 (2013年)

基于核主成分分析(KPCA)的分类数据降维可视化，matlab代码，多特征数据降维 基于核主成分分析(KPCA)的分类数据降维

KPCA.rar_改进KPCA_核 线性_核主成分分析_特征提取算法_非线性特征

基于组合核函数KPCA的人脸识别研究.pdf

kpca (2)_KPCA核心思想_KPCA_分类_非线性降维_compositionaad_

最新资源

基于核主成分分析(KPCA)的分类数据降维可视化，matlab代码，多特征数据降维基于核主成分分析(KPCA)的分类数据降维

KPCA.rar_改进KPCA_核线性_核主成分分析_特征提取算法_非线性特征