Genetic programming based compressive strength prediction model for green concrete

Abstract

Machine Learning (ML) has transformed the workplace in all the engineering domains. Traditionally, the lab experiments are conducted to evaluate the compressive strength of concrete however, it’s have been proved to be time-consuming and labour-intensive. The high costs involved in the testing limits the number of trials and thus compromises with reliability of the construction. The study proposes state-of-art novel genetic programming (GP) based soft-computing prediction model for the estimation of the concrete compressive strength. The current focus on sustainable development goals have led to the innovations in high performance concrete from waste materials which can potentially reduce the adverse environmental impact of cement production, however, the non-linear correlation makes it difficult for the empirical relations and basic ML models to arrive at a reliable prediction model. For this purpose, a dataset of 144 trials of fly ash and silica fume concrete is taken from literature for training and testing the GP model. The GP model performs best with ten genes and a maximum tree depth of four. The population and tournament sizes have been set at 100 and 30, respectively. Crossover and mutation probabilities are 0.84 and 0.14, respectively. The GP model is authenticated using statistical parameters (R2 = 0.98 and RMSE = 0.03). The model proposes an easy-to-use equation for the prediction of compressive strength. The proposed model will save capital, labour and time along with allowing better planning since there is no need to wait for 28 long days. The models can be trained on the in-situ data and trained model can be put to use for variety of datasets in future research.