GAN-UTILIZING GENERATIVE ADVERSARIAL NETWORKS FOR ENHANCING MULTI-SENSOR DATA IN PREDICTING WEAR OF CNC MACHINE TOOLS

Abstract

This paper explores the use of data augmentation techniques to address the challenge of data scarcity in predicting wear of CNC machine tools. Due to the difficulty in acquiring extensive experimental and real-world factory data, developing precise machine learning models for supervised prediction has been problematic. To overcome this obstacle, our study focuses on employing a multi-sensor strategy to classify the states of tool wear during the milling process, where data fusion from multiple sensors is conducted in the frequency domain for feature extraction. To augment the limited training data available, we introduce a generative adversarial network (GAN) specifically designed for data augmentation. An early stopping mechanism is implemented to enhance the efficiency of the GAN. Our findings indicate that incorporating GAN-generated synthetic data significantly boosts the accuracy of tool wear classification models, thereby reducing the reliance on data from real industrial settings. This research presents a promising avenue for supplementing scarce experimental data with synthetic data generated by GANs for the prediction of tool wear.