ML-Based Classification Models for Assessing Workpiece Dimensional Accuracy

Abstract

The integration of machine learning (ML) into manufacturing processes has significantly improved predictive maintenance and quality assessment, particularly in Computer Numerical Control (CNC) machining. This study presents the development and evaluation of four types of ML classification models, namely Support Vector Machine (SVM), k-Nearest Neighbors (KNN), Naïve Bayes, and Artificial Neural Network (ANN) models for assessing the dimensional accuracy for workpieces produced via step drilling on a horizontal CNC machining center. Vibration signal features were extracted during the machining process, resulting in 27 statistical features per workpiece. The models were trained on a dataset from 2019 and tested on an independent dataset from 2021 in order to evaluate their temporal robustness. The Medium Gaussian SVM model and the ANN model with the 27-21-2 architecture achieved the highest training accuracy, namely 98.77%, and the latter showed a perfect generalization ability with a 100% accuracy for the 2021 test dataset. These findings confirm the suitability of ML-based models for quality assessment in the context of machining processes, and their potential for integration into real-time smart manufacturing systems.