Prediction of mechanical solutions for a laminated LCEs system fusing an analytical model and neural networks

This paper presents a convenient and efficient method to predict the mechanical solutions of a laminated Liquid Crystal Elastomers (LCEs) system subjected to combined thermo-mechanical load, based on a back propagation (BP) neural network which is trained by machine learning from a database established by analytical solutions. Firstly, the general solutions of temperature, displacement, and stress of any single layer in the LCEs system are obtained by solving the two-dimensional (2D) governing equations of both heat conduction and thermoelasticity. Then, the unknown coefficients in above general solutions are determined by a transfer-matrix method based on the continuity condition at the interface of adjacent layers and the combined thermo-mechanical loads condition at the surface of the LCEs system. The formula derivation and calculator program are verified through convergence studies and comparisons with FEM results. Finally, a database with displacements of LCEs system in a temperature field subjected to 561 sets of mechanical loads is established based on the presented analytical model. The BP neural network based on above database is further applied to establish the relationship between deformation and mechanical load to predict the elastic deformation of the LCEs system in a temperature field subjected to a mechanical load. Moreover, the BP network can also inverse the coefficients of mechanical load which induces the specific deformation in a temperature field. The numerical examples show that: (1) The deformation of a laminated LCEs system due to thermal load is limited within the range of human temperature changes from 36 °C to 40 °C. (2) The thickness of the LCE is a sensitive parameter on the deformation at the bottom surface of the system. (3) The accuracy of predicted displacements induced by the thermo-mechanical load and the inversed mechanical load based on deformation of the LCEs system in a temperature field using BP neural network reaches 99.6% and 98.5% respectively.