This repository contains the implementation of an Artificial Neural Network (ANN) surrogate model for predicting multilayer pavement resilient moduli from FWD-like surface deflection data. The ANN is trained on datasets generated using nonlinear Finite Element Method (FEM) simulations performed in ILLI-PAVE 2005.
This work was carried out as part of a Master of Technology (MTP) project at
Indian Institute of Technology Bhubaneswar.
Conventional backcalculation methods based on Layered Elastic Theory (LET) suffer from:
- Linear elastic assumptions
- Non-uniqueness of solutions
- Inability to capture stress-dependent behavior of granular and subgrade layers
To overcome these limitations, this project combines:
- Nonlinear FEM-based forward modeling
- ANN-based surrogate modeling for fast inversion
This framework enables near-instantaneous prediction of layer moduli while retaining FEM-level realism.
- FEM simulations performed using ILLI-PAVE 2005
- Synthetic FWD deflection basins (Uz) generated for:
- Asphalt Concrete
- Granular Base
- Lime-Stabilized Soil / CTB
- Fine-Grained Subgrade
- ANN trained to map deflection basins → layer moduli
- Model evaluated using MAE, MSE, and %AAE (MAPE)
Refer to the full project report for detailed methodology and theory.
ANN-FEM-Pavement-Modulus-Backcalculation/ │ ├── src/ ├── data/ │ └── raw/ ├── models/ ├── results/ ├── report/