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

Abstract: This research introduces a two-domain physics-informed neural network (PINN) framework designed to simulate contaminant transport within a geosynthetic clay liner (GCL)/soil liner (SL) composite system. In this model, the thin GCL layer is represented by a steady-state advection-dispersion-biodegradation equation, while the underlying soil liner is treated as a transient transport domain. We assess two distinct formulations against analytical and finite-element benchmark solutions under varying leachate-head scenarios: a standard PINN employing soft constraint enforcement (Std-PINN) and a hard-constrained PINN (H-PINN), which integrates selected boundary and initial conditions directly into the trial solutions.

The results indicate that while the Std-PINN accurately captures general breakthrough trends, it exhibits higher errors during the initial transport phase, especially under elevated leachate heads where advection dominates. Conversely, the H-PINN mitigates the optimization challenges associated with penalty-based constraints, yielding more precise and stable concentration predictions. This improvement is evidenced by a reduction in Mean Absolute Error (MAE) from approximately 0.058–0.067 for the Std-PINN to 0.011–0.023 for the H-PINN, and a decrease in Mean Relative Error (MRE) from 9.10%–19.16% to 2.08%–3.14%.

Parametric studies demonstrate that the H-PINN utilizing a tanh activation function, combined with an optimized network architecture, achieves superior predictive accuracy. Furthermore, the H-PINN is applied to inverse modeling to estimate the SL degradation half-life using limited concentration data. The method shows reliable convergence to target values and maintains robustness even under low-to-moderate levels of observation noise.

Source: arXiv Generated at: 2026-06-04 00:00:00 UTC