Artificial neural network-based parameter identification of a beam-3D tip attachment system

Abstract

In this study, an artificial neural network (ANN)-based approach is proposed for estimating the tip mass attachment and structural damping parameters in a beam-3D mass system undergoing combined bending and torsional vibrations. The study begins with a detailed explanation of the calculation of the frequency response functions (FRFs) for this specific system. Subsequently, a difference vector is defined, based on the discrepancy between experimental and numerical FRF curves. This vector is dependent on ten parameters: the mass of the tip attachment, its mass moments of inertia, and the coordinates of the mass center, and the structural damping ratios of the beam. An orthogonal design method is then employed to create a design space for these parameters, and the elements of the difference vector are calculated for each point within this space. The points in the design space and the computed difference vectors are utilized as input and output data for training the ANN. The optimization process conducted with the obtained ANN model allows for realistic estimation of the tip mass parameters and damping values. The analysis reveals that as the design space widens, the parameter estimation process becomes increasingly challenging. This, in turn, necessitates a larger number of points in the design space and more neurons in the hidden layers of the trained network. In cases where the design space is small to medium in size, the parameter estimation errors are observed to be <5%. However, for wider design spaces, the estimation errors tend to increase.