Predicción de la resistencia a compresión en hormigón simple mediante un modelo de regresión lineal múltiple

Abstract

The construction industry has relied on traditional methods to determine the compressive strength of concrete for the past few decades. These conventional procedures, characterized by their destructive nature and long waiting periods, have generated economic losses and delays in project execution over the years. Faced with this problem, it is appropriate to develop non-destructive predictive models that allow obtaining the compressive strength value of concrete immediately, to optimize construction processes and reduce costs associated with delays in project execution. The objective of this research work was to develop a multiple linear regression model that estimates the compressive strength of plain concrete at 7, 14, and 28 days. This research was structured in four phases to meet the established objectives: in the preliminary phase, a database was built on the physical properties that influence the compressive strength of concrete. Then, in the first phase, the predictors considered for the construction of the MLR predictive model were determined. Subsequently, in the second phase, the predictive capacity of the model was evaluated using evaluation metrics for prediction. Finally, in the third phase, the MLR model was validated by comparing its predictions with the compressive strength values obtained in concrete cylinders made with materials from the area. The multiple linear regression model built from a database with 179 records of factors that influence the compressive strength of concrete contains 13 independent variables. This model proved to have a good fit to the data, with an adjusted coefficient of determination equal to 0.7934. This value indicates that the model explains 79 percent of the behavior of the compressive strength of concrete. To evaluate the predictive capacity of the model during the testing stage, a test with 35 data was carried out, the results obtained with the evaluation metrics such as RMSE and MAPE is equal ± 20.195 kg/cm2 and 7.17 percent, respectively. Likewise, during the validation stage, the multiple linear regression model for the ages of 7, 14 and 28 days presented a MAPE of 10 percent and an RMSE equal to ± 22.148 kg/cm2. These results indicate that the proposed predictive model has an acceptable predictive capacity.