Intelligenza Artificiale Generativa per la produzione di dati sintetici nella diagnosi di macchine rotanti

The subject of rotating machinery diagnostics is witnessing a growing significance of Artificial Intelligence (AI) as it demonstrates a remarkable capability to establish correlations between diagnostic parameters and the overall health condition. Nevertheless, the use of AI methodologies necessitates a substantial amount of training data, posing challenges in acquiring such data for damaged industrial machineries. This research presents a novel approach for generating synthetic data using Cycle-Consistent Generative Adversarial Networks (cycleGANs). The proposed model is designed to transform wavelet-based images of simulated vibration signals into real data obtained from machines exhibiting bearing faults. The Maximum Mean Discrepancy (MMD) demonstrates a noteworthy resemblance between synthetic and real data. Synthetic data result effective for training Convolutional Neural Networks (CNNs) by means of Transfer Learning (TL). The research is conducted using the test rig for industrial bearing located at the Mechanical and Aerospace Engineering Department of Politecnico di Torino.