Optimisation of material composition in functionally graded plates using a structure-tuned deep neural network

This study presents a neural network (NN)-based approach for optimising material composition in multi-layered functionally graded (FG) plates to minimise steady-state thermal stress. The focus is on the metal-ceramic composition across the thickness of heat-resistant FG plates, with the volume fractions of the ceramic constituent in each layer treated as design variables. A fully-connected NN, configured with an open-source Bayesian optimisation framework, is employed to predict the maximum in-plane thermal stress for various combinations of design variables. The optimal distribution of material composition is determined by applying a backpropagation algorithm to the NN. Numerical computations on ten- and twelve-layered FG plates demonstrate the usefulness of this approach in designing FG materials. NNs trained with 8000 samples enable the successful acquisition of valid optimal solutions within a practical timeframe.