On machine learning assisted data-driven bridging of FSDT and HOZT for high-fidelity uncertainty quantification of laminated composite and sandwich plates

Abstract

First-order shear deformation theory (FSDT) is less accurate compared to higher-order theories like higher-order zigzag theory (HOZT). In case of large-scale simulation-based analyses like uncertainty quantification and optimization using FSDT, such errors propagate and accumulate over multiple realizations, leading to significantly erroneous results. Consideration of higher-order theories results in significantly increased computational expenses, even though these theories are more accurate. The aspect of computational efficiency becomes more critical when thousands of realizations are necessary for the analyses. Here we propose to exploit Gaussian process-based machine learning for creating a computational bridging between FSDT and HOZT, wherein the accuracy of HOZT can be achieved while having the low computational expenses of FSDT. The machine learning augmented FSDT algorithm is referred to here as modified FSDT (mFSDT), based on which extensive deterministic results and Monte Carlo simulation-assisted probabilistic results are presented for the free vibration analysis of shear deformation sensitive structures like laminated composite and sandwich plates considering various configurations. The proposed algorithm of bridging different laminate theories is generic in nature and it can be utilized further in a range of other static and dynamic analyses concerning composite plates and shells for accurate, yet efficient results.