Random and harmonic responses of plain woven carbon fiber reinforced conical-conical shell based on machine learning multiscale modelling

Abstract

This study analyzes the frequency-response characteristics under non-stationary base random acceleration and harmonic points force excitations of plain woven carbon fiber reinforced polymers (PW-CFRP) conical-conical shells comprehensively. Artificial neural network (ANN) based machine learning multiscale modelling strategy is introduced in the process of predicting the mechanical properties of PW-CFRP. The dataset is generated by randomization, linear fitting and two-step homogenization and then trained to generate an ANN to predict the mechanical properties of PW-CFRP. The governing equations of the conical-conical shell under random acceleration excitation are obtained according to the first order shear deformation theory (FSDT) and Hamilton's principle. After the structure is split along the circumference at the excitation points, the harmonic points force excitations are introduced based on the discontinuity condition between conical segments. Generalized differential quadrature (GDQ) is applied during the solving procedure. After comparison studies focusing on the frequency-response characteristics under two types of excitations, two examples on harmonic and random responses of conical-conical shell and conical-cylindrical shell are carried out separately. In this process, the effects of excitation type, excitation and response points, geometric parameters on the frequency-response characteristics are extensively analyzed, which proves efficiency and accuracy of the ANN based machine learning multiscale modelling strategy and the constructed GDQ based computational strategy.