Neural model of the Adaptive Tuned Particle Impact Damper

The paper presents a novel approach for the modeling of the Adaptive Tuned Particle Impact Damper (ATPID) using Multilayer Perceptron (MLP). The main motivation was the recognition that such an approach can support the development of novel neural modeling and the optimal determination of damper parameters in terms of mechanical vibration attenuation. The training data were obtained using a theoretical model validated experimentally. The optimally selected MLP was compared with other regression models using 10 different metrics. A hyperparameter tuning of the determined neural network architecture was conducted based on the input parameters such as excitation amplitude, grain mass, and ATPID damper height. The analyses show that the proposed neural network could quickly and accurately estimate the system’s vibration amplitude and efficiently predict the optimal damper height. The ability to effectively determine the correct optimal height is crucial for ATPID damper control. The high efficiency in predicting the system’s vibration amplitude allows for the replacement of the theoretical model with applied time-consuming contact forces. The MLP accurately estimated vibration amplitudes with 1%–10% error for interpolated data and up to 15% for extrapolated cases. The issue raised is particularly important from the perspective of real-time damper control. It was found that computing a single case using the artificial neural network is more than ten times faster compared to the theoretical model. Therefore, the proposed ATPID damper model based on a neural network forms the basis for further considerations and scientific research to finally propose a control algorithm in the future.