American Concrete Institute ACI Materials Journal January, 2019
Artificial Intelligence to Investigate Properties of Concrete Mixtures Incorporating Recycled Aggregate Concrete
Seyedhamed Sadati, Leonardo Enzo Brito da Silva, Donald C. Wunsch II, and Kamal H. Khayat
Newswise — The research presented in this paper establishes a model to estimate the modulus of elasticity (MOE) of concrete made with coarse recycled concrete aggregate (RCA) that is based on the artificial neural networks (ANN) method. Given the variable nature of RCA materials, there is a need to have predictive tools that enable reliable estimations for properties of concrete made with RCA, including the MOE.
The performance of concrete made with RCA can be impacted when using RCA materials of inferior quality compared to that of virgin aggregate. The MOE is one characteristic of concrete that can be highly sensitive to the incorporation and quality of RCA. The availability of residual mortar in RCA particles can reduce the overall stiffness and restraining capacity of the coarse aggregate skeleton and increase the absolute volume of mortar in the hardened state. This reduces the rigidity of the concrete, resulting in a lower MOE compared to the corresponding mixtures prepared without any RCA. Moreover, the presence of micro-cracks in the residual mortar, old virgin aggregate, and the old interfacial transition zone (ITZ) between these two phases (anticipated due to the crushing procedure), can affect the MOE significantly.
A database was developed and analyzed to establish a prediction model. Given the wide range of investigated parameters, the model is intended to estimate the effect of RCA on the MOE of concrete designated for a variety of applications, including infrastructure construction. Developing the model involved three main phases:
- The first phase included the development and screening of a database of published literature on the MOE of concrete made with coarse RCA. A database summarizing more than 480 data series obtained from 52 technical publications was developed, from which 90 percent were used for model development. Various scenarios for input parameters were considered as shown in Table 1. A multilayer perceptron network and Levenberg-Marquardt backward propagation learning algorithm were successfully used to develop and optimize the ANN architecture with one hidden layer for estimating the variations in the MOE of concrete made with RCA.
- The second phase generated laboratory data to validate the model. An additional dataset of 43 concrete mixtures obtained from laboratory investigation of concrete with well-known properties was used to validate the established model.
- The third phase involved testing the model based on ANN. The remaining portion of the database, (10 percent) was employed for testing the model.
A wide range of mixture design parameters, summarized in Table 1, are considered as input parameters. The extent of reduction in 28-day MOE (RMOE) was considered as the output parameter (See output_parameter).
See Table 1.
The developed model was incorporated for a case study on a typical concrete used for rigid pavement construction. The baseline concrete was a mixture made with 0.40 water-to-cementitious materials ratio and 545 lb/yd3 (323 kg/m3) of a binary cement, designated for rigid pavement construction. The simulated mixture was made with virgin coarse aggregate with 0.8 percent water absorption, 170 lbf/ft^3 (2730 kg/m3)-specific gravity, and 28 percent Los Angeles abrasion mass loss.
Based on the model outputs, contour graphs shown in Figs. 1 and 2 were developed to showcase the effect of RCA on the variations in the MOE of the mixture. The results obtained through the case study indicate that the reduction in the MOE of pavement concrete can be limited to 10 percent when coarse RCA with water absorption lower than 2.5 percent or an oven-dry-specific gravity higher than 156 lb/ft3 (2500 kg/m3) is used even at the full replacement rate. The use of 30 and 50 percent RCA can lead to a reduction of up to 20 and 30 percent in the MOE of pavement concrete when low quality RCA with an oven-dry-specific gravity of 131 lb/ft3 (2100 kg/m3) and water absorption as high as 8.5 percent is used.
The research can be found in a paper titled “Artificial Intelligence to Investigate Modulus of Elasticity of Recycled Aggregate Concrete,” published by ACI Materials Journal.
For more information, contact:
Marketing & Communications Coordinator
Always advancing – Founded in 1904 with a headquarters in Farmington Hills, MI, USA, and a regional office in Dubai, UAE, the American Concrete Institute is a leading authority and resource worldwide for the development, dissemination, and adoption of its consensus-based standards, technical resources, educational & training programs, certification programs, and proven expertise for individuals and organizations involved in concrete design, construction, and materials, who share a commitment to pursuing the best use of concrete. For additional information, visit concrete.org.