Scholarlink

Abstract <p>Statistical process control (SPC) relies on control charts (CCs) to detect process shifts, but their performance can deteriorate for skewed, non-normal data such as Gamma-distributed cycle times and specification-related measures. This study proposes a machine learning-based control-chart <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(CC_{ML}\)</EquationSource> </InlineEquation> framework that uses information from classical CCs to improve monitoring accuracy under non-normal conditions. Features were derived from traditional CC statistics (including Shewhart, EWMA, and CUSUM) and used to train multiple individual machine learning (ML) models for shift detection. A Monte Carlo simulation study was conducted to compare the proposed approach with classical CCs and selected existing methods across a range of Gamma-distributed scenarios. Real-world utility was further assessed using an industrial sensor dataset focused on machine failure risk prediction. In simulated experiments, the proposed <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(CC_{ML}\)</EquationSource> </InlineEquation> consistently detected shifts earlier and more reliably than classical and existing alternatives, achieving up to a 96.23% improvement in early shift detection. On the industrial sensor dataset, the framework produced timelier alerts and maintained stable monitoring performance for skewed process data. The results indicate that integrating ML with CC information can substantially strengthen SPC for non-normal industrial processes, providing a scalable tool for modern quality monitoring and anomaly detection.</p> Graphic abstract <p></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

Data-driven control charts for monitoring Gamma-distributed manufacturing processes using machine learning

  • 期刊论文

  • 发表日期：

    2026年5月22日

查看全文
  • Faraz Mukhtiar,
  • Babar Zaman,
  • Naveed Razzaq Butt,
  • Marva Ajab,
  • Muhammad Iftikhar Faraz

摘要

Abstract

Statistical process control (SPC) relies on control charts (CCs) to detect process shifts, but their performance can deteriorate for skewed, non-normal data such as Gamma-distributed cycle times and specification-related measures. This study proposes a machine learning-based control-chart \(CC_{ML}\) framework that uses information from classical CCs to improve monitoring accuracy under non-normal conditions. Features were derived from traditional CC statistics (including Shewhart, EWMA, and CUSUM) and used to train multiple individual machine learning (ML) models for shift detection. A Monte Carlo simulation study was conducted to compare the proposed approach with classical CCs and selected existing methods across a range of Gamma-distributed scenarios. Real-world utility was further assessed using an industrial sensor dataset focused on machine failure risk prediction. In simulated experiments, the proposed \(CC_{ML}\) consistently detected shifts earlier and more reliably than classical and existing alternatives, achieving up to a 96.23% improvement in early shift detection. On the industrial sensor dataset, the framework produced timelier alerts and maintained stable monitoring performance for skewed process data. The results indicate that integrating ML with CC information can substantially strengthen SPC for non-normal industrial processes, providing a scalable tool for modern quality monitoring and anomaly detection.

Graphic abstract