Scholarlink

<p>Predicting the remaining time of production orders is a critical task in the manufacturing sector, particularly for professionals involved in production planning and control (PPC). These managers frequently face substantial uncertainty when estimating delivery commitments to customers. Numerous techniques have been applied to address this challenge, including neural networks, time series models, and non-parametric statistical approaches. Among the emerging methodologies, process mining has proven particularly effective in leveraging event logs to extract insights into the actual execution of business processes. This study introduces a hybrid predictive model that integrates annotated transition systems with machine learning techniques to enhance the accuracy of remaining time predictions for ongoing production orders in industrial settings. The combination of predictive models is achieved through the optimization of a linear programming formulation that minimizes the aggregate absolute error. The proposed approach was evaluated using both synthetic event logs and real-world data from a manufacturing company. Experimental results demonstrate that the proposed method outperforms all benchmarked techniques in terms of prediction accuracy across test scenarios. Across the three tested products, the hybrid model consistently obtained the lowest MAE, improving accuracy up to 12% relative to baseline and state-of-the-art machine-learning models.</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

A process mining and machine learning based approach for remaining time prediction of production orders

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月6日

查看全文
  • João Gabriel Santin Botelho,
  • Alexandre Checoli Choueiri,
  • José Eduardo Pécora Júnior,
  • Eduardo Alves Portela Santos

摘要

Predicting the remaining time of production orders is a critical task in the manufacturing sector, particularly for professionals involved in production planning and control (PPC). These managers frequently face substantial uncertainty when estimating delivery commitments to customers. Numerous techniques have been applied to address this challenge, including neural networks, time series models, and non-parametric statistical approaches. Among the emerging methodologies, process mining has proven particularly effective in leveraging event logs to extract insights into the actual execution of business processes. This study introduces a hybrid predictive model that integrates annotated transition systems with machine learning techniques to enhance the accuracy of remaining time predictions for ongoing production orders in industrial settings. The combination of predictive models is achieved through the optimization of a linear programming formulation that minimizes the aggregate absolute error. The proposed approach was evaluated using both synthetic event logs and real-world data from a manufacturing company. Experimental results demonstrate that the proposed method outperforms all benchmarked techniques in terms of prediction accuracy across test scenarios. Across the three tested products, the hybrid model consistently obtained the lowest MAE, improving accuracy up to 12% relative to baseline and state-of-the-art machine-learning models.