Feature space monitoring for smart manufacturing via statistics pattern analysis
Feature space monitoring for smart manufacturing via statistics pattern analysis
复制标题
DOI：
10.1016/j.compchemeng.2019.04.010
复制
发表时间：
2019-07-12
影响因子：
4.3
通讯作者：
Shah, Devarshi
中科院分区：
文献类型：
--
作者：
He, Q. Peter;Wang, Jin;Shah, Devarshi
统计过程监控（SPM）是任何系统长期可靠运行的重要组成部分，在智能制造（SM）时代，其重要性只会变得更加重要。以前，我们提出了统计模式分析（SPA）的基础上使用各种统计量来量化过程特性的想法，并监视这些统计量，而不是过程变量本身来执行过程监控。在这项工作中，我们提供了一个全面的审查最近取得的进展，在SPA框架，这是我们最近概述的SPM路线图的基础。讨论了样本特征提取和变量特征提取，以及在故障检测和诊断以及软测量开发中的新应用。具体来说，我们提供了第一个系统的检查，在处理过程特性，包括动态，非线性和数据非高斯的SPA的能力，并比较其性能的代表国家的最先进的SPM方法，突出增强功能的功能为基础的监测。此外，SPA的性能进行了测试，使用基准工业模拟器TEP的故障检测和诊断，加上湿实验室和软测量开发的工业案例研究。最后，讨论了SPA在应对智能制造大数据带来的新挑战方面的优势和潜在局限性。(C)2019爱思唯尔有限公司版权所有。
Statistical process monitoring (SPM) is an important component in the long-term reliable operation of any system and its importance can only become greater in the era of smart manufacturing (SM). Previously we proposed statistics pattern analysis (SPA) based on the idea of using various statistics to quantify process characteristics, and monitoring these statistics instead of process variables themselves to perform process monitoring. In this work we provide a comprehensive review on recent progresses made in SPA framework, which underpins a roadmap of SPM we outlined recently. Both sample-wise feature extraction and variable-wise feature extraction are discussed, with new applications in both fault detection and diagnosis, and soft sensor development. Specifically, we provide the first systematic examination on the SPA's capability in handling process characteristics including dynamics, nonlinearity and data non-Gaussianity; and compare its performance to representative state-of-the-art SPM methods to highlight the enhanced capability of feature-based monitoring. In addition, the performance of SPA is tested using the benchmark industrial simulator TEP for fault detection and diagnosis, plus a wet lab and an industrial case studies for soft sensor development. Finally, the advantages and potential limitations of SPA in addressing the new challenges presented by smart manufacturing big data are discussed. (C) 2019 Elsevier Ltd. All rights reserved.