Improvement of correlation using artificial neural networks technique for the prediction of resistivity against soil strength properties

Abstract

Objectives: To investigate the relationship of laboratory electrical resistivity with soil shear strength properties at a controlled moisture content (30%). However, this research was carried out to collect the forty (40) samples from different locations of Perak state, Malaysia under different atmospheric conditions incorporating characterization of soil tests, laboratory 1D resistivity method using fabricated sand box. Methods/Statistical Analysis: This work was carried out through direct shear test by using DS7 software to evaluate cohesion and angle of friction, the values ranges from 21.22kPa to 87.25kPa and 5.17° to 42.85° respectively. Findings: The obtained correlation models were compared to the co-efficient of R2 predicted from artificial neural network system. Lavenberg-Marquardt learning rule was utilized up to 20 hidden neurons which showedthe higher accuracy in all soil samples. Relationship developed between laboratory electrical resistivity with cohesion and angle of friction were found out to be more significant with co-efficient of regression values for cohesion enhanced from 0.54 to 0.70 (all soils), 0.265 to 0.515 (silty-sand soils) and 0.027 to 0.540 (clay soils). Similarly, for angle of friction the R2 improved from 0.56 to 0.66 (all soils), 0.57 to 0.763 (silty-sand) and 0.21 to 0.447 (clay soils) respectively. Therefore, the findings reported in this study has been improved which could be helpful because ANNproduces promising results and its advantages can be utilized by developing or using new algorithms in future studies which can produce more precise evaluations. Applications/Improvements: Artificial Neural Networks (ANN) are the algorithms that can be used to perform non-linear statistical modeling and able to provide a new alternative regression values. Therefore, in the current this technique is applied on the same data extracted from least-square regression analysis in order to achieve better and improved correlation models.