Ensemble learning based sustainable approach to rebuilding metal structures prediction

Area of Science:

• Materials Science
• Civil Engineering
• Computer Science

Background:

• The European Green Deal emphasizes product reuse and durability, particularly for construction steel.
• Life Cycle Assessment (LCA) is crucial for environmental impact but has gaps in predicting end-of-life steel properties.
• Predicting steel reusability at the end-of-life stage (C1-C4, D) is challenging.

Purpose of the Study:

• To develop a machine learning model for predicting construction steel reusability.
• To determine steel yield strength using a non-destructive magnetic method for reuse assessment.
• To support informed decision-making for reusing construction steel.

Main Methods:

• Utilized machine learning (ML) approaches, focusing on regression problems.
• Employed ensemble learning to combine multiple models for improved prediction accuracy.
• Applied the WeightedEnsemble method, integrating 8 models for predicting tensile strength.

Main Results:

• The WeightedEnsemble method achieved the highest prediction accuracy (MSE = 441 MPa, RMSE = 21 MPa).
• The model demonstrated high accuracy and low conclusion delay (IL = 0.119 s) in predicting tensile strength.
• Non-destructive magnetic testing data was effectively used for property prediction.

Conclusions:

• Machine learning, specifically ensemble methods, significantly enhances the prediction of construction steel reusability.
• An automated tool based on this ML model can aid construction professionals in making informed decisions for steel reuse.
• This approach represents a significant advancement in managing steel reuse processes at the building's end-of-life stage (Stage D).