Digitalización del concreto: propuesta de un modelo generalizado para el diseño de mezcla con la aplicación de machine learning
Cargando...
Fecha
Authors
Título de la revista
ISSN de la revista
Título del volumen
Editor
Universidad de Ingeniería y Tecnología
Resumen
El concreto es uno de los materiales más utilizados a nivel mundial y su demanda sigue en aumento. Su producción genera un gran impacto en la sostenibilidad y la economía, por lo que optimizar sus procesos de producción es prioridad. Uno de los puntos más críticos en la producción del concreto es el diseño de mezclas que, actualmente, se realiza mediante métodos laboriosos. La digitalización genera una oportunidad para mejorar la eficiencia. En consecuencia, la implementación de técnicas de machine learning es un tema relevante en el rubro. Pese a sus ventajas, no hay muchas investigaciones al respecto en Perú. Por ello, el presente trabajo propone un modelo basado en machine learning que puede predecir propiedades del concreto a partir de la composición de la mezcla que busca mejorar el rendimiento obtenido en investigaciones previas. Para ello, se recopilo data existente y se realizó el entrenamiento de cuatro modelos con la técnica de validación cruzada k-fold. Se eligió el algoritmo con mejor desempeño para optimizarse con GridSearch y realizar la validación con la data recopilada. Asimismo, se utilizó SHAP para abordar la explicabilidad de las predicciones. El modelo validado fue XGBoost para predecir slump con R2 = 0.71, RMSE = 20.8, MARD = 0.05, D10 % = 66.2 % y D25 % = 89.5 %, y para predecir f c con R2 = 0.92, RMSE = 9.4, MARD = 0.07, D10 % = 63.64 % y D25 % = 88 %. Finalmente, se desarrolló una interfaz gráfica basada en los modelos.
Concrete is one of the most popular materials worldwide and its demand continues to increase. Its production has a significant impact on sustainability and the economy, therefore, optimizing its production processes is a priority. One of the most critical aspects is the concrete mix design, which is carried out using laborious methods. The digitization presents an opportunity to enhance production efficiency. Consequently, the implementation of machine learning techniques for the optimization of concrete mix design is a relevant research topic. Nevertheless, despite its advantages, there is not much research on this topic in Peru. Therefore, this study proposes a generalized model that can predict concrete properties from the composition of a mixture based on machine learning, which aims to improve the performance compared to prior investigations. To this end, data was compiled from existing literature, and four models were trained using the k-fold crossvalidation technique. The best performing algorithm was selected for model optimization using GridSearch and validated with the collected data. SHAP was also used to asses the explainibility of the predictions. The validated model was XGBoost for slump prediction with R2 = 0.71, RMSE = 20.8, MARD = 0.05, D10 % = 66.2 % and D25 % = 89.5 %, and for f c prediction with R2 = 0.92, RMSE = 9.4, MARD = 0.43, D10 % = 63.64 % and D25 % = 88 %. Finally, a graphical user interface was developed based on the models.
Concrete is one of the most popular materials worldwide and its demand continues to increase. Its production has a significant impact on sustainability and the economy, therefore, optimizing its production processes is a priority. One of the most critical aspects is the concrete mix design, which is carried out using laborious methods. The digitization presents an opportunity to enhance production efficiency. Consequently, the implementation of machine learning techniques for the optimization of concrete mix design is a relevant research topic. Nevertheless, despite its advantages, there is not much research on this topic in Peru. Therefore, this study proposes a generalized model that can predict concrete properties from the composition of a mixture based on machine learning, which aims to improve the performance compared to prior investigations. To this end, data was compiled from existing literature, and four models were trained using the k-fold crossvalidation technique. The best performing algorithm was selected for model optimization using GridSearch and validated with the collected data. SHAP was also used to asses the explainibility of the predictions. The validated model was XGBoost for slump prediction with R2 = 0.71, RMSE = 20.8, MARD = 0.05, D10 % = 66.2 % and D25 % = 89.5 %, and for f c prediction with R2 = 0.92, RMSE = 9.4, MARD = 0.43, D10 % = 63.64 % and D25 % = 88 %. Finally, a graphical user interface was developed based on the models.
Descripción
Citación
Vera Rodriguez, A. E. (2026). Digitalización del concreto: propuesta de un modelo generalizado para el diseño de mezcla con la aplicación de machine learning [Tesis de Título Profesional, Universidad de Ingeniería y Tecnología]. Repositorio Institucional UTEC. https://hdl.handle.net/20.500.12815/561
Colecciones
Aprobación
Revisión
Complementado por
Referenciado por
Licencia Creative Commons
Excepto donde se indique lo contrario, la licencia de este ítem se describe como info:eu-repo/semantics/openAccess

