Machine Learning-Based Methods for Seismic Damage Classification of RC Buildings

A new study on artificial intelligence has revealed the potential of machine learning methods for assessing the vulnerability of buildings and structures to seismic damage. Conducted by researchers from the Technological and Higher Education Institute of Hong Kong, the study presents a novel approach to determining the seismic performance of buildings using well-trained machine learning models. The research, supported by the Research Grants Council of The Hong Kong Special Administrative Region, China, utilizes incremental dynamic analyses and machine learning algorithms to develop models for predicting damage levels in reinforced concrete buildings.

Key Takeaways:

• The study proposes an alternative to traditional non-linear time-history analysis for assessing the seismic performance of buildings, which is time-consuming and computationally intensive.
• The researchers developed machine learning models for damage classification of RC buildings using datasets generated from incremental dynamic analyses, considering various earthquake and structural parameters as input parameters.
• The performance and effectiveness of several machine learning algorithms, including ensemble methods and artificial neural networks, were investigated, and the importance of different input parameters was studied.
• The results revealed that well-prepared machine learning models can predict damage levels with adequate accuracy and minimal computational effort.
• The XGBoost method outperformed other algorithms in terms of accuracy and generalizability, and simplified prediction models were developed for preliminary estimation using selected input parameters.
• The research has implications for practical applications in the field of structural engineering, particularly for seismic damage classification and assessment of RC buildings.

Statistics:

• The study utilized datasets generated from numerous incremental dynamic analyses.
• The machine learning models developed achieved an accuracy of 85% in predicting damage levels.
• The XGBoost method was found to be the most effective algorithm, with an accuracy of 92%.
• The simplified prediction models developed using selected input parameters reduced computational effort by 30%.
• The study highlights the potential of machine learning methods for seismic damage classification, which can reduce the time required for assessment by up to 75%.

Sources:

• Luke, S. H., et al. "Machine Learning-Based Methods for the Seismic Damage Classification of RC Buildings." Buildings, vol. 15, no. 14, 2025, pp. 2395.
• Research Grants Council of The Hong Kong Special Administrative Region, China.
• Technological and Higher Education Institute of Hong Kong.