Process disturbance identification through integration of spatiotemporal ICA and CART approach
| 详细信息 | 推荐本文 | |
- 作者：Chih-Chou Chiu ; Shin-Ying Hwang ; Deborah F. Cook and Yuan-Ping Luh
- 关键词：Process disturbance ; Statistical process control ; Engineering process control ; Spatiotemporal independent component analysis ; Classification and regression tree
- 刊名：Neural Computing & Applications
- 出版时间：July, 2010
- 出版年：2010
- 期刊代码：113_0941-0643
- 类别：cp
- 卷：19
- 期：5
- 页码：677-689
- 数据来源：sp
摘要
Many studies have been conducted over the past several years evaluating the integrated use of statistical process control (SPC) and engineering process control (EPC). The majority of these studies reported that combining SPC with EPC outperforms the use of only SPC or EPC. Basically, the former aims to rapidly detect assignable causes and time points for abnormalities that take place during process; and the latter is a method in which input variables are adjusted against process outputs through a feedback control mechanism. Although combining SPC with EPC can effectively detect time points when abnormalities occur during process, their combination can also cause an increased occurrence of false alarms when autocorrelation is present in the process. In this study, to increase the accuracy of process disturbance identification, we propose the integration of spatiotemporal independent component analysis (stICA) with the classification and regression tree (CART) approach to improve our capability to identify process disturbances and recognize shifts in the correlated process parameters. The integration of the CART methodology results in the development of decision rules that can provide valuable information related to the impact of variation in process variable values. These decision rules can provide an increased understanding of process behavior and useful information for process control. For comparison, the integration of traditional principle component analysis (PCA) with CART (called PCA-CART), ICA with CART (called ICA-CART) and cumulative sum chart approaches were applied to evaluate the identification capability of the proposed approach. As the results reveal, the proposed approach is more effective for monitoring correlated process.