413 / 2022-07-28 15:16:09

Parameter Estimation based on Deep Neural Network Model for a Damped Euler-Bernoulli Beam

Accelerance function; Cantilevered beam; Deep neural network; Material property estimation

Special Sessions > Applications of machine learning in vibration and noise problems

In this study, we propose a deep neural network model-based parameter estimation method to estimate the mechanical properties of cantilevered beams such as elastic modulus and damping modulus. The proposed method estimates unknown parameters by using the latent vector generated in the process of optimizing the weights of the deep neural network model. The beam used for estimating the mechanical properties was analytically modeled as a damped Euler-Bernoulli beam. Herein, the magnitude and phase of the accelerance function at the end of the beam were used as training data for the deep neural network model. In order to investigate the effect of noise that may occur in the experiment, the estimation accuracy of the proposed method was analyzed when the target accelerance function is in the presence of noise. The proposed method shows high estimation accuracy even in the presence of noise. Further, we compared the estimation accuracy of the proposed method and the existing methods (gradient descent method, pattern search method) which are used conventionally for parameter estimation in terms of various initial values and noise levels. Finally, we applied our method to the results of real vibration experiments to estimate the stiffness and damping coefficients of various materials (steel, aluminum, copper).