Optimization of dynamic behavior of thin-walled laminated cylindrical shells by genetic algorithms and deep neural networks supported by modal shape identification

This paper presents a novel method for the optimization of the dynamic behavior of a laminated cylinder through stacking sequence optimization. The number of layers reaches 48 and the optimized parameters are either the value of the fundamental natural frequency or the width of the frequency gaps around the excitation force frequencies. The proposed procedure involves automatic mode shape identification and a combined genetic algorithm-deep neural network procedure along with a Curriculum Learning loop, enabling the improvement of accuracy and reduction of computational costs. The proposed optimization algorithm is accurate, robust, and significantly faster than typical genetic algorithm optimization, in which the objective function values are calculated using the finite element method. A rough rule for the estimation of the necessary number of patterns is proposed and the optimal results obtained using the proposed approach are compared to the results obtained using a standard approach (without mode shape identification), in order to demonstrate the effectiveness and robustness of the new method.