Self-Healing Performance Assessment of Bacterial-Based Concrete Using Machine Learning Approaches

Area of Science:

• Civil Engineering
• Materials Science
• Computational Science

Background:

• Bacterial-based self-healing concrete (BSHC) offers excellent crack repair but lab evaluation is costly and time-consuming.
• Predicting healing performance (HP) can optimize bacteria selection and healing mechanism adoption.
• Machine learning (ML) offers a data-driven approach to predict BSHC HP.

Purpose of the Study:

• To develop and evaluate ML models for predicting the HP of BSHC.
• To identify key factors influencing BSHC HP.
• To compare the predictive accuracy of different ML algorithms.

Main Methods:

• Three BSHC types (UBHC, ABHC, NBHC) and five ML algorithms (SVR, DTR, DNN, GBR, RF) were used.
• 22 influencing factors were employed as variables in ML models.
• A dataset of 797 BSHC tests from 2000-2021 was used for model verification. Grid Search Algorithm (GSA) tuned parameters.

Main Results:

• The Gradient Boosting Regression (GBR) model demonstrated superior prediction ability with R² = 0.956 and RMSE = 6.756%.
• Sensitivity analysis identified key variables influencing BSHC HP within the GBR model.
• All ML models were validated against extensive literature data.

Conclusions:

• ML models, particularly GBR, can accurately predict BSHC HP, saving time and resources.
• The study provides insights into factors affecting self-healing concrete performance.
• This approach facilitates the practical application and design optimization of BSHC.