Presenting a Novel Approach for Predicting the Compressive Strength of Structural Lightweight Concrete Based on Pattern Recognition and Gene Expression Programming

• Published in: Arabian journal for science and engineering (2011) 2023-10, Vol.48 (10), p.14169-14181
• Main Authors: Hosseini, Seyed Azim, Maleki Toulabi, Hossein
• Format: Article
• Language: English
• Subjects: Algorithms; Artificial neural networks; Back propagation networks; Compressive strength; Construction costs; Correlation coefficients; Engineering; Entropy (Information theory); Environmental protection; Gene expression; Humanities and Social Sciences; Lightweight concretes; Model accuracy; multidisciplinary; Neural networks; Pattern recognition; Research Article-Civil Engineering; Science; Static loads
• ISSN: 2193-567X; 1319-8025; 2191-4281
• DOI: 10.1007/s13369-023-07996-2

Because of the ever-increasing demand for lightweight concrete (LC), especially in structural applications where it is used to reduce dead load of the structure and construction costs while contributing to environmental protection thanks to decreased use of aggregates and cement, presenting optimal mixing design for preparing structural LC (SLC) is of paramount importance. In this article, application of artificial neural networks in predicting the compressive strength of SLC was investigated. For this purpose, we used some pattern recognition neural network and genetic expression programming (GEP) with experimental data from tests to predict the compressive strength of SLC. A total of 10 input parameters from the SLC mix design were utilized for predicting the compressive strength by means of a scaled conjugate gradient backpropagation algorithm in the form of a neural network. The outputs were classified into 5 strength groups of M1, M2, M3, M4 and M5. The simulation results were 98.8% accurate in classifying SLC samples under different predefined ranges of compressive strength using the confusion matrix diagram. Moreover, the cross-entropy error obtained from testing the neural network (NN) model and correlation coefficient (R 2 ) of GEP for predicting compressive strength of the SLC were evaluated at 0.0076082 and 0.9912, respectively, indicating high accuracy of the model. Application of this model could greatly help relevant practitioners formulate SLCs with desired compressive strength.