A supervised artificial neural network representational model based prediction of contact pressure and bulk density

Hamid Taghavifar


Artificial neural networks are fitting replacements to traditional statistical modeling techniques in scientific disciplines. The attempt was to find a supervised feed-forward artificial neural network (ANN) trained by back propagation algorithms to predict contact pressure and bulk density from readily available data obtained from experiments conducted in soil bin facility and a single wheel-tester. A total of 648 samples were used for training, validating and testing the neural networks. Evaluating a three layered architecture (i.e. one hidden layer) with a four layered one (i.e. two hidden layers), the optimal topology to yield better performance on the criteria of lower root mean squared error (RMSE), T value and coefficient of determination (R2) was a four-layered one. Then among 2 hidden layers, each of layers was increased from 0 to 20 to realize the best number of neurons by the best performance. It was divulged that a 6-12-10-2 provided the best performance of RMSE and T were 0.027, 0.977, and R2 for bulk density and contact pressure were 0.9950 and 0.9982, respectively, and that a proper trained ANN could model and predict output variable wherein mathematical and statistical methods may fail to have the equal efficiency owing to unidentified behavior of soil due to its unknown nature. 

Full Text:



Ansorge D, Godwin R J. The effect of tyres and a rubber track at high axle loads on soil compaction, Part 1: Single axle-studies. Biosyst. Eng. 2007, 98(1): 115-126.

Patel S K, Mani I. Effect of multiple passes of tractor with varying normal load on subsoil compaction. J of Terramechanics. 2011, 48(4): 277-284.

Keller T. A model for the prediction of contact area and the distribution of vertical stress below agricultural tyres from readily available tyre parameters. Biosyst. Eng. 2005, 87: 375-386.

Thompson M L, Kramer M A. Modeling chemical processing using prior knowledge and neural networks. AIChE J. 1994, 40(8): 1328-1340.

Jaiswal S, Benson E R, Bernard J C, Van Wicklen G L. Neural network modelling and sensitivity analysis of a mechanical poultry catching system. Biosyst. Eng. 2005, 92(1), 59-68.

Roul A K, Raheman H, Pansare M S, Machavaram R. Predicting the draught requirement of tillage implements in sandy clay loam soil using an artificial neural network. Biosyst. Eng. 2009, 104(4), 476-485.

Zhang Z X, Kushwaha R L. Application of neural networks to simulate soil tool interaction and soil behaviour. Canadian Agricultural Engineering. 1999, 41(2), 119-125

Mardani A, Shahidi K, Rahmani A, Mashoofi B, Karimmaslak H. Studies on a long soil bin for soil-tool interaction. Cercetări Agronomice în Moldova. 2010, 142(2), 5-10.

Black, C.A., 1965. Methods of soil analysis, Part I. American Society of Agronomy, Madison, WI.

Weiss M, Baret F, Leroy M, Hautecoeur O, Prevot L, Bruguier N. Validation of neural network techniques for the estimation of canopy biophysical variables from vegetation data. Vegetation-2000, Lake Maggiore, Italy


  • There are currently no refbacks.