A novel physics-informed neural operator for thermochemical curing analysis of carbon-fibre-reinforced thermosetting composites

The temperature field during the cure process significantly influences the final quality of thermosetting composites. It is essential to ensure temperature histories within specifications by cure optimisation, of which the essence equals solving parametric coupled PDEs with varying boundary conditions. Recently, the physics informed neural network (PINN) has shown promising potential for solving PDE unsupervised. Conventional PINN approximates the solution function based on the point-to-point manner, which requires vast collocation points and suffers from an unacceptable training burden. In comparison, this paper proposes a novel physics informed neural operator (PINO) framework that directly constructs the solution operator between the whole cure cycles and temperature or DoC histories in a function-to-function manner. Through enforcing global constraints on the field outputs, PINO can simultaneously solve parametric coupled PDEs unsupervised and significantly accelerate the training process. Experiments under deterministic and parametric settings are conducted to exhibit the notable superiority of the proposed method.