Physics-Informed Neural Network Modeling of Biodegradable Contaminant Transport through GCL/SL Composite Liners

The purpose of this research is to create a two-domain physics-informed neural network (PINN) framework to model the transport of biodegradable contaminants in a GCL/SL composite liner system. For the GCL layer, we use a steady-state advection-dispersion-biodegradation model, and for the SL, we use a conceptual model represented as a transient transport domain. The research proposes two separate PINN formulations: a standard PINN (Std-PINN) that uses soft constraint enforcement and a hard constrained PINN (H-PINN) that has boundary and initial condition constraints specified directly in the trial solution. The models have been compared against analytical and finite-element model solutions for different leachate-head conditions. The results of the study indicate that

the Std-PINN captures the general breakthrough behavior, but has increased errors during the initial transport phase (especially when leachate head is high and advective transport is dominant). The H-PINN model outperforms the Std-PINN model by reducing the optimization complexity resulting from penalty-based constraints to offer more accurate and more stable concentration predictions. The quantitative results indicate that the H-PINN model reduces the Mean Absolute Error (MAE) from approximately 0.058-0.067 to 0.011-0.023, and the Mean Relative Error (MRE) from 9.10%-19.16% to 2.08%-3.14%. A further parametric investigation shows that the H-PINN with a tanh activation function and optimized neural network configuration provides the best predictive accuracy.

Message: Hi FE, I need help with “Physics-Informed Neural Network Modeling of Biodegradable Contaminant Transport through GCL/SL Composite Liners” in “Environmental Engineering”