Impact of thermal loads on silica fume-modified lightweight concrete: Machine learning approach to assess compressive strength evolution

Abstract

Fire resistance of concrete is complex to comprehend as it is a composite material with ingredients of varied thermal properties. As the behaviour of concrete subjected to thermal loads is complex and required exhaustive experimentation, the use of soft computing techniques can be beneficial for predicting the residual compressive strength of concrete. Leveraging the power of machine learning (ML), specifically Artificial Neural Network (ANN), Random Forest (RF), and AdaBoost (AdB) models, this study aims to overcome the limitations of conventional prediction methods. Models were developed for the prediction of residual compressive strength of light weight concrete with silica fume (SF) exposed to thermal loads. Cement, silica fume, lightweight pumice aggregate, water-to-cement ratio, superplasticizer, and temperature were taken as input attributes while residual compressive strength was taken as output. Firstly, the data were subjected to extensive statistical analysis before the models were created. Additionally, hyperparameter tuning using grid search was done to increase the models' efficiency. The coefficient of determination (R2), root mean squared error (RMSE), and mean absolute error (MAE) as model evaluation parameters were used to evaluate the performance of various developed ML models. When the performance of models was compared, it was found that the AdaBoost (AdB) model outperformed all other ML models. Hence, this study highlights the significance of ML techniques in addressing complex engineering challenges and underscore the practical implications for optimizing structural design and performance.