Scholarlink

<p>Process monitoring plays a vital role in controlling manufacturing processes and ensuring product quality by enabling early detection of process shifts. This study proposes a novel multivariate monitoring method that integrates a stacked gated recurrent unit (S-GRU) with a self-adaptive particle swarm optimization (PSO) called S-GRU-PSO. In this framework, PSO is employed to automatically optimize hyperparameters of the S-GRU architecture, enhancing its learning capability and adaptability. The proposed model was evaluated using synthetic datasets with both normal and non-normal (gamma) distributions across various data dimensions and shift magnitudes. It was also validated on a real-world wine production dataset to assess its industrial applicability. Comparative experiments were conducted against benchmark monitoring methods, including distance-based support vector machine (D-SVM), random forest with real-time contrast (RF-RTC), D-SVM optimized by differential evolution (D-SVM-DE), Stacked long short-term memory (S-LSTM), and stacked gated recurrent unit (S-GRU). The results demonstrate that the S-GRU-PSO model achieves superior performance, with a 26.9% improvement in shift detection for normal data, 11.35% for gamma-distributed data, and 16.6% for the wine dataset, as measured by reductions in average run length (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="10489_2025_6617_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="45" /> </InlineMediaObject> <EquationSource Format="TEX">\(ARL_{1}\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>A</mi> <mi>R</mi> <msub> <mi>L</mi> <mn>1</mn> </msub> </mrow> </math></EquationSource> </InlineEquation>). These findings confirm not only the model’s effectiveness in quickly detecting process changes but also its novel integration of temporal deep learning and adaptive optimization, offering a robust and scalable solution for real-time industrial monitoring applications.</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

Real-time manufacturing process monitoring using GRU with PSO algorithm

  • 期刊论文

  • 发表日期：

    2025年6月18日

查看全文
  • Atinkut Atinafu Yilma,
  • Chao-Lung Yang,
  • Bereket Haile Woldegiorgis

摘要

Process monitoring plays a vital role in controlling manufacturing processes and ensuring product quality by enabling early detection of process shifts. This study proposes a novel multivariate monitoring method that integrates a stacked gated recurrent unit (S-GRU) with a self-adaptive particle swarm optimization (PSO) called S-GRU-PSO. In this framework, PSO is employed to automatically optimize hyperparameters of the S-GRU architecture, enhancing its learning capability and adaptability. The proposed model was evaluated using synthetic datasets with both normal and non-normal (gamma) distributions across various data dimensions and shift magnitudes. It was also validated on a real-world wine production dataset to assess its industrial applicability. Comparative experiments were conducted against benchmark monitoring methods, including distance-based support vector machine (D-SVM), random forest with real-time contrast (RF-RTC), D-SVM optimized by differential evolution (D-SVM-DE), Stacked long short-term memory (S-LSTM), and stacked gated recurrent unit (S-GRU). The results demonstrate that the S-GRU-PSO model achieves superior performance, with a 26.9% improvement in shift detection for normal data, 11.35% for gamma-distributed data, and 16.6% for the wine dataset, as measured by reductions in average run length ( \(ARL_{1}\) A R L 1 ). These findings confirm not only the model’s effectiveness in quickly detecting process changes but also its novel integration of temporal deep learning and adaptive optimization, offering a robust and scalable solution for real-time industrial monitoring applications.