Scholarlink

<p>Defect prediction in Investment Casting (IC) is inherently a very complex engineering activity, due to the large number of processes simultaneously affecting the casting mechanics to produce defect free castings. The casting process is susceptible to defects, so a reliable method for predicting defects prior to manufacturing is of critical importance. Achieving high yield and consistent casting quality is essential for ensuring favourable production economics. While numerical simulation methods have shown good predictive capabilities, but they often rely on input parameters based on specific assumptions, which may not fully capture the variability of real-world conditions. In contrast, Machine Learning (ML) offers a promising data-driven approach to model the IC process and predict defects more accurately by learning complex relationships from historical production data. ML models are used to predict defects like shrinkage and ceramic inclusions by analyzing historical process data to detect underlying patterns. Supervised ML models like, K-Nearest Neighbor (K-NN), Gradient Boost (GB), Support Vector Machine (SVM), Logistic Regression (LR) and Random Forest (RF) are compared. K-fold cross validation is used for testing and training of the available data from the manufacturing unit. Grid search method is implemented for hyper-parameter tuning of the parameters of ML models. Accuracy, Precision, Recall and F1-score are calculated from confusion matrix for each model. All models had good prediction output accuracy of 80% and above but RF outperformed other models with accuracy 97.59%, precision 97.51%, F1-score 98.79% and recall of 100% for ceramic inclusion and accuracy 96.38%, precision 96.07%, F1-score 98.00% and 100% recall for shrinkage prediction<b>.</b></p> Graphical 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

Comparative study of supervised machine learning-driven defect prediction in investment casting process

  • 期刊论文

  • 发表日期：

    2025年12月20日

查看全文
  • Sumit Kishanlal Sahitya,
  • Janak B. Valaki,
  • Mayur Sutaria,
  • Hamirbhai R. Sapramer

摘要

Defect prediction in Investment Casting (IC) is inherently a very complex engineering activity, due to the large number of processes simultaneously affecting the casting mechanics to produce defect free castings. The casting process is susceptible to defects, so a reliable method for predicting defects prior to manufacturing is of critical importance. Achieving high yield and consistent casting quality is essential for ensuring favourable production economics. While numerical simulation methods have shown good predictive capabilities, but they often rely on input parameters based on specific assumptions, which may not fully capture the variability of real-world conditions. In contrast, Machine Learning (ML) offers a promising data-driven approach to model the IC process and predict defects more accurately by learning complex relationships from historical production data. ML models are used to predict defects like shrinkage and ceramic inclusions by analyzing historical process data to detect underlying patterns. Supervised ML models like, K-Nearest Neighbor (K-NN), Gradient Boost (GB), Support Vector Machine (SVM), Logistic Regression (LR) and Random Forest (RF) are compared. K-fold cross validation is used for testing and training of the available data from the manufacturing unit. Grid search method is implemented for hyper-parameter tuning of the parameters of ML models. Accuracy, Precision, Recall and F1-score are calculated from confusion matrix for each model. All models had good prediction output accuracy of 80% and above but RF outperformed other models with accuracy 97.59%, precision 97.51%, F1-score 98.79% and recall of 100% for ceramic inclusion and accuracy 96.38%, precision 96.07%, F1-score 98.00% and 100% recall for shrinkage prediction.

Graphical Abstract