Compensation of Cutout Effect on Natural Frequencies of Beam Structure Using Deep Learning: A Novel Multi-Disciplinary Approach

This study investigates the effects of cutouts and added masses on the natural frequencies of a beam structure and develops a deep-learning algorithm to predict the optimal location of the added mass for a desired set of natural frequencies. A beam structure with a cutout is analyzed using COMSOL MULTIPHYSICS to obtain the natural frequencies for various cutouts and added mass locations. The resulting data set is used to train a deep learning algorithm, which is tested by providing it with desired natural frequencies and cutout locations and having it predict the optimal location of the added mass. Within the scope of this study, first of all, hyperparameter optimization for the estimation model was performed with a Random Search tuner algorithm. This step was an important step to improve the performance of the model and to find the best configuration. Then the best model architecture was created and the effect of different Keras optimizers was examined in detail. In these evaluations, the performance of the model was evaluated objectively by using the targeted metrics — mean square error (MSE), root mean squared error (RMSE), mean absolute error (MAE), and [Formula: see text]-square. MSE, RMSE, MAE and [Formula: see text]-square values of 0.000001, 0.0011413300, 0.000876, and 0.999799 have been obtained with the Adamax optimizer. Work continued with the best-performing optimizer and necessary adjustments were made to further improve the performance of the model. In addition, the training process and adaptability of the model were carefully observed by subtracting the loss curve and training curves.