Explainable machine learning model and gene expression programming for predicting reinforced concrete beams moment...

Area of Science:

• Structural Engineering
• Computational Mechanics
• Fire Safety Engineering

Background:

• Reinforced Concrete (RC) beams are critical structural components.
• Assessing their moment capacity under fire conditions is essential for safety.
• Existing methods may lack accuracy or efficiency in fire scenarios.

Purpose of the Study:

• To develop and evaluate novel computational models for estimating the moment capacity (M_r) of RC beams exposed to fire.
• To compare the performance of Gene Expression Programming (GEP) with various Machine Learning (ML) algorithms.

Main Methods:

• Utilized a database of 280 samples for training and validation.
• Employed GEP and ML models including XGBoost, AdaBoost, and LightGBM.
• Input parameters included geometric properties, reinforcement details, fire duration, and concrete strength.
• Performance was assessed using MAE, MSE, RMSE, R², and regression gradients.
• SHAP analysis was used for model interpretability.

Main Results:

• GEP and ML models demonstrated high accuracy in predicting M_r.
• XGBoost exhibited superior performance with the best R² and lowest error rates.
• SHAP analysis provided insights into the influential parameters for XGBoost predictions.

Conclusions:

• The proposed GEP and ML approaches offer accurate and reliable methods for estimating RC beam moment capacity under fire.
• These computational tools can aid in the fire safety design and assessment of reinforced concrete structures.