spread-footing-surrogate

Spread Footing Bending Check — Random Forest Surrogate Model

A machine learning surrogate model that instantly predicts whether a concrete spread footing passes the bending check per CSA A23.3-19 (Canadian concrete design standard). It replaces iterative structural calculations with a near-instant classification, making it suitable for design optimization loops, preliminary sizing tools, and educational applications.

---

Model Description

This is a RandomForestClassifier trained on synthetically generated data produced by a physics-based spread footing calculator implementing CSA A23.3-19. Given the geometric, loading, and material properties of a footing, the model predicts whether the footing will pass or fail the bending design check in both directions.

• Model type: Random Forest Classifier (scikit-learn)
• Design standard: CSA A23.3-19
• Task: Binary classification (bending_ok: pass / fail)
• Training data: Synthetically generated using a validated Python calculator

---

Intended Use

• Rapid preliminary sizing of spread footings
• Design space exploration and optimization
• Integration into structural engineering tools as a fast surrogate
• Educational demonstration of ML in structural engineering

---

Inputs

The model takes the following features as input:

---

Output

---

How to Use

1. Download the model

from huggingface_hub import hf_hub_download
import pickle

model_path = hf_hub_download(
    repo_id="siavashhabibi1993/spread-footing-surrogate",
    filename="spread_footing_random_forest_model.pkl"
)

with open(model_path, "rb") as f:
    model = pickle.load(f)

2. Prepare your input and predict

import numpy as np

# Example footing input — must match the training feature order
# column_shape: 0 = rectangular, 1 = circular (LabelEncoded)
X = np.array([[
    500,    # Ps (kN)
    700,    # Pf (kN)
    0,      # Pup (kN)
    150,    # qss (kPa)
    200,    # qsf (kPa)
    2.5,    # lf (m)
    2.5,    # wf (m)
    500,    # df (mm)
    25,     # fc (MPa)
    75,     # cover (mm)
    1.5,    # dUSF (m)
    100,    # tslab (mm)
    0,      # column_shape (0 = rectangular)
    0.4,    # lc (m)
    0.4,    # wc (m)
    0.4,    # dc (m)
    25,     # fcCol (MPa)
    1.0,    # lambda_
    0.65,   # phiC
    0.85,   # phiS
    400,    # fy (MPa)
    24,     # gammaConc (kN/m³)
    18,     # gammaSoil (kN/m³)
    18,     # gammaFill (kN/m³)
    20,     # bar_size_bot
    15      # bar_size_top
]])

prediction = model.predict(X)
print("Bending OK:" if prediction[0] == 1 else "Bending FAILS")

---

Important Notes

• The column_shape column must be label encoded before prediction: rectangular = 0, circular = 1
• A SimpleImputer (mean strategy) was applied to training data. If your inputs contain missing values, apply the same imputation before predicting.
• Feature order must exactly match the input table above.

---

Training Details

• Algorithm: RandomForestClassifier (scikit-learn)
• Hyperparameter tuning: n_estimators swept from 10 to 90
• Validation: Cross-validation (5-fold) used during tuning
• Training data: Synthetically generated from a CSA A23.3-19 physics-based calculator

---

Limitations

• This model predicts only the bending check (bending_ok). It does not predict one-way shear, punching shear, bearing, or uplift checks.
• Trained on synthetic data — real-world edge cases may not be fully represented.
• Not intended as a replacement for a licensed engineer's judgement.

---

Environment

scikit-learn == 0.22.1
numpy        == 2.3.5
pandas       == 2.3.3
python       == 3.11.14

---

Citation

If you use this model in your work, please credit the repository:

@misc{spread-footing-surrogate,
  author = {Siavash},
  title  = {Spread Footing Bending Check — Random Forest Surrogate Model},
  year   = {2026},
  url    = {https://huggingface.co/siavashhabibi1993/spread-footing-surrogate}
}