AI Summary & Key Takeaways
AI InsightPCA and Isomap improved the accuracy and Type-I error measures of all models; however, no such improvement was obtained on the Type-II error rates.
📌 3 Key Takeaways
Özet
Significant advances in digital technology and advanced analytical tools have had a substantial impact on the production environment and laid the foundation for Industry 4.0 and intelligent production concepts. Predictive engineering is one of the key pillars of smart manufacturing that necessitates the collection and analysis of real-time data with an anticipatory point of view through advanced analytical techniques. In the literature, machine learning-based methods have received a great deal of attention to extract valuable information from process data for fault detection. In this study, fault prediction problem was addressed in a molding process that includes successive steps by applying machine learning methods with dimension reduction techniques. The techniques of Principal Component Analysis (PCA), and Isometric Feature Mapping (Isomap) were first utilized for dimension reduction. Then, the data was analyzed for fault prediction with several machine learning techniques, namely, Support Vector Machine (SVM), Neural Network (NN), and Logistic Regression (LR). The dataset for our analysis includes sensor data captured during the molding process of a wheel rim manufacturer. Several criteria, including accuracy, area under curve (AUC), Type I, and Type II error, were employed to assess the predictive performance of the methods applied, including and the model variants reinforced with PCA and Isomap. Our study demonstrates that all predictive model variants have performed with high accuracy, ranging between 92.16% (LR) and 98.04% (PCA-NN). PCA and Isomap improved the accuracy and Type-I error measures of all models; however, no such improvement was obtained on the Type-II error rates.