Abstract
AbstractThis study presents the development and validation of a mix design determination procedure for geopolymer concrete to achieve the desired compressive strength. The procedure integrates artificial neural network (ANN) model developed based on a comprehensive data base from literature, data clustering, and parameter optimization techniques to enhance accuracy and reliability. Experimental validation is undertaken to demonstrate the mix design determination procedure’s capability to accurately predict mix designs for geopolymer concrete based on the target compressive strength, validating its efficacy for mix proportion determination. The integration of chemical oxide content in fly ash, curing time, curing temperature, and activator properties results in a 15.9% improvement in prediction accuracy for the training dataset and a 68.3% enhancement for the testing dataset, compared to the base ANN model that includes only the weight of fly ash and activator properties. Employing data clustering techniques enables the identification of prior estimates for the mix design parameters related to specific fly ash types and target compressive strength, streamlining the mix design process by analyzing pertinent data subsets. Parameter optimization ensures refined mix proportions, achieving the desired target strength economically while minimizing material waste and cost. The development of a user interface facilitates easy manipulation of mix designs, catering to users of varying expertise levels. Additional options for deeper insights into geopolymer concrete characteristics can be integrated into the mix design determination procedure. To assess the mix design determination procedure's ability to generalize effectively, a variety of fly ash samples with distinct chemical compositions were utilized, differing from those already present in the database. This approach allows for a thorough evaluation of the mix design determination procedure's performance when presented with fly ash compositions it has not encountered before. By doing so, this provides insights into the adaptability of the mix design determination procedure beyond the limitations of the training and testing datasets.
Funder
Royal Melbourne Institute of Technology
Publisher
Springer Science and Business Media LLC