Scholarlink

<p>Melt pressure and melt temperature are crucial indicators of product quality and process safety in polymer extrusion. However, real-time monitoring and control of these parameters remain challenging due to the complexity of process variables and the non-linear dynamics of the process. This study proposes an explainable machine learning (ML) framework to model these parameters using real-world data from a single-screw extrusion process. Key innovations include the estimation of sensor lags via the cross-correlation function, noise reduction using a fifth-order Butterworth filter, and the design of engineered features that reproduce thermomechanical interactions. Amongst several ML models evaluated, the Light Gradient Boosting Machine (LightGBM) achieves the best performance for melt temperature prediction (RMSE: 0.0367&#xa0;°C, R<sup>2</sup>: 0.9954), improving accuracy by 79.7% over downsampled data. Likewise, the Gradient Boosting Regressor (GBR) yields optimal melt pressure prediction (RMSE: 0.0032&#xa0;MPa, R<sup>2</sup>: 0.996), suggesting an 85% improvement. Explainability is ensured through Shapley Additive Explanation (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Partial Dependence Plot (PDP) analyses, revealing that barrel zone temperatures and screw speed are the dominant predictors. The approach is computationally efficient, supporting real-time deployment and offering a transparent path for integrating Artificial Intelligence (AI) into industrial extrusion monitoring and control.</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

Machine learned and explainable prediction of melt pressure and temperature for polypropylene extrusion

  • 期刊论文

  • 发表日期：

    2025年8月7日

查看全文
  • Mohammad Basit Akram,
  • Collin Gilarno,
  • Kazuo McCoy,
  • Sam Osten,
  • Michael Rumps,
  • Utsav Thota,
  • Daniel Campbell,
  • Ganesh Balasubramanian

摘要

Melt pressure and melt temperature are crucial indicators of product quality and process safety in polymer extrusion. However, real-time monitoring and control of these parameters remain challenging due to the complexity of process variables and the non-linear dynamics of the process. This study proposes an explainable machine learning (ML) framework to model these parameters using real-world data from a single-screw extrusion process. Key innovations include the estimation of sensor lags via the cross-correlation function, noise reduction using a fifth-order Butterworth filter, and the design of engineered features that reproduce thermomechanical interactions. Amongst several ML models evaluated, the Light Gradient Boosting Machine (LightGBM) achieves the best performance for melt temperature prediction (RMSE: 0.0367 °C, R2: 0.9954), improving accuracy by 79.7% over downsampled data. Likewise, the Gradient Boosting Regressor (GBR) yields optimal melt pressure prediction (RMSE: 0.0032 MPa, R2: 0.996), suggesting an 85% improvement. Explainability is ensured through Shapley Additive Explanation (SHAP), Local Interpretable Model-Agnostic Explanations (LIME), and Partial Dependence Plot (PDP) analyses, revealing that barrel zone temperatures and screw speed are the dominant predictors. The approach is computationally efficient, supporting real-time deployment and offering a transparent path for integrating Artificial Intelligence (AI) into industrial extrusion monitoring and control.