Synthetic Vibration Data Generation and Fault Classification in CNC Machines Using Transformer GANs and ConvLSTM Networks

This paper introduces a novel approach using Transformer-based Time-Series Generative Adversarial Networks (TTS-GAN) and 2D Convolutional Long Short-Term Memory Networks (ConvLSTM) to generate and classify synthetic vibration data for CNC (Computer Numerically Controlled) machines. Addressing prevalent issues of data scarcity and imbalance in CNC machine maintenance, our methodology synthesizes faulty vibration signals that are statistically indistinguishable from real data. This synthetic data generation significantly enhances the volume and diversity of training datasets, critical for developing robust fault detection algorithms. We validate the authenticity and applicability of our synthetic data through Principal Component Analysis (PCA) and t-Distributed Stochastic Neighbor Embedding (t-SNE) analyses, demonstrating high similarity with real datasets. Two configurations of the ConvLSTM model, one trained exclusively on real data and another on a mix of real and synthetic data, are tested against real datasets, demonstrating improved fault classification accuracies with the inclusion of synthetic data, achieving approximately 99% accuracy across various operations. This research not only advances the field of synthetic data generation but also has practical implications for improving predictive maintenance strategies in industrial settings.