Scholarlink

<p>Using multivariate control charts instead of univariate ones for all process variables offers significant advantages in quality management. However, in the control chart, the statistical calculation treats the measured values for all variables as a single value, making it crucial to pinpoint which variable(s) are causing the out-of-control signal. Effective corrective actions can only be devised when the fault(s) are accurately identified. This study aimed to determine machine learning techniques capable of precisely estimating fault types. The Hotelling T<sup>2</sup> chart was employed to identify out-of-control signals and specify the types of faults influenced by the variables. Various machine learning techniques were compared for classification performance. The developed model was applied to evaluate the variables of ammonia gas in a process of neutralisation of ammonia by nitric acid, with artificial neural networks (ANNs) emerging as the most successful technique, achieving the highest classification accuracy (97.43 %), the lowest standard error (0.023), and the lowest root mean square error (0.150). A key novelty of this study lies in categorizing faults by their types rather than those of the variables, simplifying the process of implementing effective corrective measures.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Fault detection using hotelling T2 chart and artificial neural networks

  • 期刊论文

  • 发表日期：

    2025年12月6日

查看全文
  • Mohamed Elamine Benrabah,
  • Ouahab Kadri,
  • Abdelghani Lakhdari,
  • Naâmane Benhassine

摘要

Using multivariate control charts instead of univariate ones for all process variables offers significant advantages in quality management. However, in the control chart, the statistical calculation treats the measured values for all variables as a single value, making it crucial to pinpoint which variable(s) are causing the out-of-control signal. Effective corrective actions can only be devised when the fault(s) are accurately identified. This study aimed to determine machine learning techniques capable of precisely estimating fault types. The Hotelling T2 chart was employed to identify out-of-control signals and specify the types of faults influenced by the variables. Various machine learning techniques were compared for classification performance. The developed model was applied to evaluate the variables of ammonia gas in a process of neutralisation of ammonia by nitric acid, with artificial neural networks (ANNs) emerging as the most successful technique, achieving the highest classification accuracy (97.43 %), the lowest standard error (0.023), and the lowest root mean square error (0.150). A key novelty of this study lies in categorizing faults by their types rather than those of the variables, simplifying the process of implementing effective corrective measures.