Supervised Undergraduate Graduation Project: Shear-Wall Resistance Prediction

This repository contains the dataset and Python workflow from Chengrui Wang's 2020 undergraduate graduation project, supervised by Kun Zhang. The project uses LS-DYNA simulation data to train machine-learning models for predicting the resistance function of reinforced-concrete shear walls.

The work focused on building a cleaned tabular dataset from LS-DYNA outputs, training a regression model, and comparing predicted resistance values against the simulation-derived response data.

Project Scope

• Data source: LS-DYNA numerical simulation results for shear-wall response.
• Target problem: resistance-function prediction for reinforced-concrete shear walls.
• Modeling task: supervised regression from geometric, reinforcement, concrete-stress, and deformation features to wall resistance.
• Baseline model: decision-tree regression with cross-validation.
• Interpretation: optional SHAP analysis for feature contribution review.

Repository Layout

.
|-- all_data.csv                 # Cleaned LS-DYNA-derived dataset
|-- merge_data.py                # Utility for merging raw CSV exports
|-- train_resistance_model.py    # Training, evaluation, plotting, optional SHAP analysis
|-- requirements.txt             # Python dependencies
`-- figures/
    |-- normalized_prediction_comparison.png
    `-- raw_prediction_comparison.png

Dataset

all_data.csv contains the merged training dataset. The columns are:

Column	Meaning
width	Shear-wall width parameter
height	Shear-wall height parameter
span	Wall span parameter
covering	Concrete cover parameter
num_longitude	Number of longitudinal reinforcement bars
diameter_longitude	Diameter of longitudinal reinforcement bars
num_hoop	Number of hoop reinforcement bars
diameter_hoop	Diameter of hoop reinforcement bars
concrete_stress	Concrete stress-related input feature
deflection	Wall deflection response variable used as an input feature
F_resistance	Target resistance value

Results Preview

The original project compared raw-data training and normalized-data training results.

Installation

Create a Python environment and install the required packages:

python -m venv .venv
.venv\Scripts\activate  # Windows PowerShell: .venv\Scripts\Activate.ps1
pip install -r requirements.txt

On macOS/Linux:

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Usage

Train and evaluate the default model:

python train_resistance_model.py

Write generated figures and metrics to a custom directory:

python train_resistance_model.py --output-dir results

Run without feature normalization:

python train_resistance_model.py --no-normalize

Merge raw CSV files into a single dataset:

python merge_data.py --input-dir raw_data --output all_data.csv

Notes

This is a small 2020 undergraduate thesis project, not a production machine-learning package. I keep it here as a record of the supervised work and as a compact example of using LS-DYNA simulation data for structural-resistance prediction.

Attribution

• Student: Chengrui Wang
• Supervisor: Kun Zhang
• Project type: 2020 undergraduate graduation project
• Topic: machine-learning prediction of reinforced-concrete shear-wall resistance functions from LS-DYNA simulation data

The repository includes the project dataset, the cleaned Python workflow, and the original comparison figures used for model-training review.