Scholarlink

<p>To achieve Zero Defect Manufacturing, Industry 4.0 firms ought to adopt AI analytics in understanding the underlying factors that cause defects. While there is less focus on data-centered approaches in defect detection and origins, this study investigated the causal factors, their magnitude and direction. The approach used a hybrid Extreme Gradient Boosting and Feed Forward Neural Networks (XGBoost-FNN) on the Predicting Manufacturing Defects Dataset from Kaggle. It achieved an accuracy of 95.37%, ROC-AUC of 87.38%, which was superior to the compared KNN and Random Forest models. From the analysis, higher maintenance hours had the highest influence in increasing the rate of defects followed by higher production volumes. The quality score of the production process and the stockout rate decreased the defect status in the products. While the AI model was based on analysis of quantitative data, the future models should focus on both qualitative and quantitative data to further analyze the causes and the trends in the causal factors. Such data-driven approaches are critical for the industry to ensure the right optimization of factors to achieve ZDM. It is also essential for academic research to develop real-world solutions for the manufacturing industry.</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

Could industry 4.0 benefit from hybrid AI analytics to enhance smart manufacturing, defect detection, and prediction?

  • 期刊论文

  • 发表日期：

    2025年12月8日

查看全文
  • Philipp Goetzinger,
  • Karsten Huffstadt,
  • Sebastian Noy

摘要

To achieve Zero Defect Manufacturing, Industry 4.0 firms ought to adopt AI analytics in understanding the underlying factors that cause defects. While there is less focus on data-centered approaches in defect detection and origins, this study investigated the causal factors, their magnitude and direction. The approach used a hybrid Extreme Gradient Boosting and Feed Forward Neural Networks (XGBoost-FNN) on the Predicting Manufacturing Defects Dataset from Kaggle. It achieved an accuracy of 95.37%, ROC-AUC of 87.38%, which was superior to the compared KNN and Random Forest models. From the analysis, higher maintenance hours had the highest influence in increasing the rate of defects followed by higher production volumes. The quality score of the production process and the stockout rate decreased the defect status in the products. While the AI model was based on analysis of quantitative data, the future models should focus on both qualitative and quantitative data to further analyze the causes and the trends in the causal factors. Such data-driven approaches are critical for the industry to ensure the right optimization of factors to achieve ZDM. It is also essential for academic research to develop real-world solutions for the manufacturing industry.