Minimizing production defect probability in industrial manufacturing using machine learning and bayesian optimization

Reducing defects in manufacturing presents a critical issue in achieving cost-effective, high-quality production. Though Machine Learning (ML) and Bayesian Optimization (BO) have been independently tested against defect prediction and process optimization, their combined use has not been extensively studied. This study seeks to fill the potential gap in the literature by suggesting a hybrid framework that integrates ML models with a BO mechanism to predict the likelihood of production defects and determine the optimal production parameter sets, consequently minimizing the possible defects. The preprocessing of the dataset of key manufacturing variables included using ADASYN to deal with data imbalances in classes and the Las Vegas Filter (LVF) to select important features. Optuna was employed to train and fine-tune nine ML models, with Random Forest (RF) indicating the best performance (accuracy: ۹۴.۶%). BO was subsequently utilized to deduce the best values of ten critical features and reduce defect probability to the minimum possible (up to ۰.۰۰۰۹۲۵). The results supported the effectiveness of the suggested method in predicting with high precision and actionable optimization, offering a scalable solution consistent with the Industry ۴.۰ objectives.

کلیدواژه ها:

Machine Learning ، Bayesian optimization ، Defect Prediction ، Prediction accuracy ، Industrial Manufacturing