System identification for a chain conveyor based on physics-dominated deep learning: System identification for a chain conveyor

This study introduces an innovative physics-dominated deep learning methodology applied to the analysis of a chain conveyor, aimed at unraveling the intricate dynamic behavior exhibited by the system. Traditional physics-based modeling methodologies encounter limitations in comprehensively capturing dynamic information due to cognitive constraints, inherent system uncertainties, and the influence of external disturbances. Moreover, a purely data-driven modeling approach is hampered by issues of diminished interpretability and computationally intensive requirements. To surmount these challenges, a hybrid modeling approach is proposed, integrating mechanism cognition principles with data-driven techniques. Within the framework of physics-based modeling, deep neural networks, coupled with neural network differentiation techniques, are employed to elucidate intricate system dynamics. Furthermore, the amalgamation of Bayesian optimization and the particle swarm optimization algorithm is introduced to advance system identification, utilizing time series similarity as the criterion for identification evaluation. Through the experimental results of the chain conveyor platform, the method proposed in this paper can not only realize the identification of the time-varying parameters but also achieve an average identification accuracy of more than 90%.