可解释性
异常检测
离群值
故障检测与隔离
计算机科学
数据挖掘
断层(地质)
过程(计算)
可靠性工程
工程类
实时计算
人工智能
操作系统
地质学
地震学
执行机构
作者
Aurora Kuras,Tzahi Y. Cath,Amanda S. Hering
出处
期刊:ACS ES&T water
[American Chemical Society]
日期:2023-04-10
卷期号:4 (3): 925-935
被引量:8
标识
DOI:10.1021/acsestwater.3c00061
摘要
Early and effective fault detection in water and wastewater treatment plants is important to maintain water quality and prevent process disruptions. Some faults, such as spike faults, are easily detected with traditional fault detection methods that identify extreme values, while other faults, such as drift faults, are difficult to identify due to their slowly changing behavior. In addition, proposed methods should assist operator decision making with straightforward interpretability. This paper applies a method in functional data analysis (FDA) for fault detection to drift faults observed in a sequencing batch membrane bioreactor and a closed-circuit reverse osmosis system. FDA enables analysis of cyclic data, which are curves or functions produced by a system with repetitive behavior over a period of time or space. Unusual functions can be identified by comparing statistics describing each function’s shape and magnitude to a reference set. In addition, functional plots visually supplement alarm results to assist operators. In this paper, we modify a retrospective FDA outlier detection method to handle the nonstationary, real-time setting required for tracking cyclic water and wastewater process data. In two case studies, we demonstrate its ability to identify drifts faults in early stages as well as a shift fault.
科研通智能强力驱动
Strongly Powered by AbleSci AI