Fig. 4  | Scientific Reports

Fig. 4 

From: Performance analysis of a three-dimensional micromixer with baffles using a flexible physics-informed neural network

Fig. 4 

Architecture of the enhanced Physics-Informed Neural Network (FlexPINN) employed in this study. Spatial coordinates \(({x}^{*},{y}^{*},{z}^{*})\) are input into a main network with parallel subnetworks to predict 14 output quantities. The total loss \({L}_{total}={L}_{GE}+{L}_{BC}+{L}_{PT}\)​ enforces governing equations, boundary conditions, and penalty term (see Eqs. 36–38 for definitions). Hyperbolic tangent (tanh) activation functions and a hybrid Adam/L-BFGS optimization strategy are used.

Back to article page