Integrated Artificial Intelligent and Physics-Based Models for Unbalance Estimation in Rotating Systems

PurposeRotating systems are one of the most common systems that transfer power in several industrial sectors. Unbalance is a severe fault that contributes to machine downtime and unscheduled maintenance actions and can damage crucial rotary systems. Estimation of unbalance and model parameter update in rotor-bearing systems are crucial for the safe and efficient operation of the machine.MethodThis research paper presents a novel approach utilizing artificial neural networks (ANNs) to identify the mass and location of unbalance in a multidisk system. Additionally, the study explores the model updating process of rotating system parameters using genetic algorithms (GAs).ResultsThe integration of physics-based models with artificial intelligence algorithms led to the development of digital twin-based methodologies for model parameter updating and unbalance estimation. For all the artificial neural networks of this paper, the coefficient of determination (R2) exceeded 0.99, indicating excellent predictive accuracy. Additionally, the implementation of genetic algorithms (GAs) consistently yielded objective function values below 5 x 10-4, enhancing the reliability of the developed systems. These results demonstrate the system's potential for real-time application and its significant advantages in predictive maintenance.ConclusionThe main novelty of this work is the development of digital twin methodologies, for unbalance estimation and parameter assessment, which consist of a mathematical model of the rotor, ANNs, GAs, and real-experimental vibration data from a rotor-kit. The obtained results demonstrate the system's potential for real-time application and its advantages in predictive maintenance.