Simultaneous identification of unbalance and cracks in rotating systems using ANN

Transverse breathing cracks and rotor unbalance are two of the most critical failure modes that compromise the reliability and safe operation of rotating machines. This study introduces an integrated physics-based and data-driven framework for the simultaneous and early identification of unbalance and transverse cracks in a multi-disk rotor-bearing system. A finite-element model incorporating a weight-dominant breathing crack formulation is developed to provide an accurate yet computationally efficient prediction of vibration response across operating conditions. Signal processing methods – fast Fourier transform and Hilbert–Huang transform – together with machine learning algorithms, enhance the detectability of early-stage cracks, overcoming the limitations of traditional signal indicators. The signal features are used to train an artificial neural network capable of estimating crack depth and location, as well as unbalance magnitude and position, with high accuracy. The proposed hybrid condition-monitoring scheme achieves R² values above 0.99 for all outputs and detects cracks as small as 10% of the shaft diameter. The results demonstrate the method’s potential to enhance system reliability, enable predictive maintenance, and support effective online health monitoring strategies for rotating systems.