ANN Based Prediction Models for Consistency and Compaction Characteristics of Bentonite-Sand Mixtures

Yücel M., Akbay Arama Z., Gençdal H. B. , Başbuğ B., Seçkin E.

Mediterranean Geoscience Union Annual Meeting-21, İstanbul, Turkey, 25 - 28 November 2021, pp.1-10

  • Publication Type: Conference Paper / Full Text
  • City: İstanbul
  • Country: Turkey
  • Page Numbers: pp.1-10
  • Istanbul University-Cerrahpasa Affiliated: Yes


Since clay soils are formed as a result of chemical degradation, they do not preserve the chemical structure of the bedrock, and they have different physical and mineralogical properties than silt, sand, and gravel. Clayey soils, which contain a large proportion of montmorillonite minerals, formed as a result of the interaction of serpentine, feldspar, or sputtering with air and water, are called 'bentonite'. Due to its high swelling, shrinkage, and consistency limit properties, bentonite causes various damages such as settlement, swelling especially in structures with low self-weight like highways, railways, water canals. On the other hand, sandy soils are granular materials composed of fragmented rocks and mineral particles.  While the composition of the sand varies depending on the rock sources and conditions, it is most commonly in the form of quartz. Sand is a non-renewable resource over time, and suitable for making concrete is in high demand. Although its usage area in civil engineering is very common, it is preferred as an embankment and filter material in geotechnical engineering. Nevertheless, natural and pure materials are not always suitable for construction purposes. For this reason, mixtures for construction work involving the use of natural materials such as roads, embankments, barriers are designed and used for different purposes. Accordingly, within the scope of this paper, the geotechnical properties of bentonite and sandy soils, both pure and mixed, will be investigated and revealed through the utilization of the ANN logic with the use of collected experimental data from the literature. In other words, this study is fictionalized as an attempt to explain the effects of the consistency characteristics of the bentonite on the behavior of bentonite-sand mixtures that are mixed in various proportions. For this purpose, 19 different literature sources are selected to generate a data set including sieve analysis, consistency limit test, and standard compaction test results. 365 experimental tests are utilized and integrated into an artificial neural network-based model that was generated to predict the consistency and compaction characteristics. In this context, the liquid limit and fine content ratio of the evaluated data are individually and together used as the input parameters and the plasticity index, the optimum water content and maximum dry density are searched as the output parameters. This training is also performed by using Matlab R2018a software and in this respect, be benefited from the "Neural Net Fitting" application, which performs analysis of ANNs as similar to the nonlinear regression principle. Consequently, the accuracy of the suggested relations was examined by conducting a comparative analysis with the use of the presented expressions in the literature. To compare the analysis results, R2, RMSE, MAE, and MAPE, which can be stated as the international metric measurements, were used. The outcomes of the comparative analyses show that the suggested ANN models can be utilized to predict the consistency and compaction properties of bentonite-sand mixtures.