Application of Machine Learning Approaches to Predict Calcium-Aluminate Cement

Calcium-aluminate cement (CAC), also known as high-alumina cement, was commonly used in structural elements from buildings constructed during the period of 1950–1970 in Catalonia (Spain). The most suitable techniques to confirm the presence of hydrated CAC in a concrete sample are chemical analysis...

Descripción completa

Detalles Bibliográficos
Autores: Heras Díez, Raúl, Rodríguez Cantalapiedra, Inmaculada, Ramírez Casas, Judith, Navarro Ezquerra, Antonia, Corominas Murtra, Andreu
Tipo de recurso: capítulo de libro
Fecha de publicación:2023
País:España
Institución:Consejo General de la Arquitectura Técnica de España (CGATE)
Repositorio:RIARTE
OAI Identifier:oai:www.riarte.es:20.500.12251/3185
Acceso en línea:http://hdl.handle.net/20.500.12251/3185
Access Level:acceso abierto
Palabra clave:Cemento aluminoso
Hormigón armado
Patologías - Construcción
Ensayos (propiedades o materiales)
Inteligencia Artificial
Redes neuronales
1203.04 Inteligencia Artificial
3313.04 Material de Construcción
3312.09 Resistencia de Materiales
3312.12 Ensayo de Materiales
3312.08 Propiedades de Los Materiales
Descripción
Sumario:Calcium-aluminate cement (CAC), also known as high-alumina cement, was commonly used in structural elements from buildings constructed during the period of 1950–1970 in Catalonia (Spain). The most suitable techniques to confirm the presence of hydrated CAC in a concrete sample are chemical analysis and X-ray diffraction (XRD), complemented by differential-thermal and gravimetric analysis (DTA/TG). Up to date, other rapid-detection techniques include the oxine test or colour test. In this line, this study has developed a machine-learning method to improve the rapid recognition process of this binder type. Based on two previ-ously labelled datasets, several avenues of resolution have been explored such as the use of data correlation statistics, and the automated preprocessing of forensic photographs. Later on, the algorithm uses binary classification by means of different supervised machine learning algorithms and, finally, through the training of various neural networks powered by transfer learning, the model successfully identifies the cement type. It has been possible to demonstrate the promising future of artificial intelligence (AI) in Architecture, Engineering and Construction (AEC), improving the ability to identify the CAC of a technician specializing in building diagnosis, obtaining a prediction accuracy of 96.5%.