---
title: DBFT Pavement Strain Predictor
emoji: 🛣️
colorFrom: green
colorTo: gray
sdk: gradio
sdk_version: 5.24.0
app_file: app.py
pinned: false
license: mit
---

# DBFT — Pavement Strain Predictor

Predicts the two critical pavement strains **directly** from a Falling Weight
Deflectometer (FWD) test — no backcalculation step:

- **ε_t** — horizontal tensile strain at the bottom of the asphalt layer
  (fatigue-cracking criterion)
- **ε_c** — vertical compressive strain at the top of the subgrade
  (rutting criterion)

Model: **DBFT (Deflection-Basin Fusion Transformer)**, ~150k parameters,
trained with a combined surrogate + field loss (λ = 1.0) on an extended
layered-elastic surrogate (14,174 PyMastic/EVERSTRESS solutions) plus 7,651
Thai DOH field measurements. Held-out-route field performance:
**R² = 0.96 (AC) / 0.88 (subgrade)**.

Every prediction includes a **local SHAP attribution** (KernelExplainer over
the 12 field-measurable inputs: 9 deflections D0–D1800 + 3 layer thicknesses)
showing how each input pushed the prediction away from the model baseline.

## Inputs

| Input | Unit | Note |
|---|---|---|
| D0 … D1800 | μm | deflection basin, normalized to 707 kPa plate pressure |
| h_AC, h_Base, h_Subbase | mm | h_Subbase = 0 for structures without a subbase |

## Run locally

```bash
pip install -r requirements.txt gradio
python app.py
```