On the use of convolutional neural networks to detect structural modifications in rotating machinery foundations

Structural vibrations in the supporting foundations of rotating machinery can significantly affect system performance and longevity, yet damage to these structures is often overlooked as a key factor in reducing the operational life of rotating systems. In this study, structural damage is experimentally introduced through small modifications affecting stiffness and proportional damping. Vibrational data measured at the bearings is then analyzed in both the time and frequency domains to detect and classify these faults. Using convolutional neural networks (CNNs), high classification accuracies are achieved, particularly with acceleration signals in both domains. Simpler CNN models, such as a three-layer network, show performance comparable to more complex architectures while offering faster training times and greater practicality. These results demonstrate that CNN-based methods can effectively automate structural health monitoring, providing a robust alternative to traditional techniques like operational modal analysis, and enhancing real-time maintenance strategies for rotating machinery.