Generalized Regression Neural Network and Empirical Models to Predict the Strength of Gypsum Pastes Containing Fly Ash and Blast Furnace Slag


Erdem T. K., Cengiz O., Tayfur G.

ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING, cilt.45, sa.5, ss.3671-3681, 2020 (SCI-Expanded) identifier identifier

  • Yayın Türü: Makale / Tam Makale
  • Cilt numarası: 45 Sayı: 5
  • Basım Tarihi: 2020
  • Doi Numarası: 10.1007/s13369-019-04210-0
  • Dergi Adı: ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING
  • Derginin Tarandığı İndeksler: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Aerospace Database, Communication Abstracts, Metadex, Pollution Abstracts, zbMATH, Civil Engineering Abstracts
  • Sayfa Sayıları: ss.3671-3681
  • Marmara Üniversitesi Adresli: Hayır

Özet

Gypsum is widely used in constructions owing to its easy application, zero shrinkage, and excellent fire resistance. Several parameters can affect the properties of gypsum pastes. To study the strength of the gypsum pastes experimentally by trying all these parameters is time-consuming and costly. Therefore, artificial intelligence methods can be very useful to predict the paste strength, which, in turn, can reduce the number of trial batches. Based on experimental data, the generalized regression neural network (GRNN) and empirical models were developed to predict strength of gypsum pastes containing fly ash (FA) and blast furnace slag (BFS). Gypsum content, pozzolan content, curing temperature, curing duration, and testing age constituted the input variables of the models while the paste strength was the target output. The trained and tested GRNN model was found to be successful in predicting strength. Sensitivity analysis by the GRNN model revealed that the curing duration and temperature were important sensitive parameters. In addition to the GRNN model, empirical models were proposed for the strength prediction. The same input variables formed the input vectors of the empirical models. The same dataset used for the calibration of the GRNN model was employed to establish the empirical models by employing genetic algorithm (GA) method. The empirical models were successfully validated. The GRNN and GA_based empirical models were also tested against the multi-linear regression (MLR) and multi-nonlinear regression (MNLR) models. The results showed the outperformance of the GRNN and the GA_based empirical models over the others.