Author/Authors :
Hematian، Rasoul نويسنده Deptartment of Mechanics of Agricultural Machinary, Faculty of Agriculture, Tarbiat Modares University, P.O. Box 14115-111, Tehran 14114, Iran , , Fayyazi، Ebrahim نويسنده Deptartment of Mechanics of Agricultural Machinary, Faculty of Agriculture, Tarbiat Modares University, P.O. Box 14115-111, Tehran 14114, Iran , , Hosseinzadeh، Bahram نويسنده Deptartment of Mechanics of Agricultural Machinary, Faculty of Agriculture, Tarbiat Modares University, P.O. Box 14115-111, Tehran 14114, Iran , , Najafi، Gholamhassan نويسنده Deptartment of Mechanics of Agricultural Machinary, Faculty of Agriculture, Tarbiat Modares University, P.O. Box 14115-111, Tehran 14114, Iran , , Ghobadian، Barat نويسنده Department of Agricultural Machinery Engineering, Faculty of Agriculture, Tarbiat Modares University, Tehran, I.R. IRAN ,
Abstract :
The current research was conducted to study the shearing characteristics of sugarcane stems using nano-coated knifes for shearing the stems. The experiments were conducted at five levels of moisture contents (46, 54, 62, 70 and 78% w.b.), three levels of shear shearing speed (5, 10 and 15 mm/min) with two types of cutting knifes (usual type and nano-coated type). Results of statistical analysis indicated that the effects of moisture content, shearing speed and type of cutting knife on shearing strength of sugarcane stem were significant (P < 0.01). As the moisture content decreased from 78 to 46%, the shearing force and specific shearing energy of stems decreased 20% and 80%, respectively. With increasing the shearing speed from 5 to 15 mm/min, the shearing force and specific shearing energy encountered to 12% and 7% decrease, respectively. Using nano-coated knifes, 32% and 34% decrease was observed in the shearing strength and specific shearing energy, respectively. Several artificial neural networks were studied to model the preliminary experimental results. In order to evaluate the different networks effectively, the database was assigned to three steps of test, validation and train, which 70% of the data was assigned to the train step, 15% of the data was designated for the validation step, and 15% of the data was used for the test step. Several topologies were evaluated to obtain the maximum R2 and minimum MSE values. The results revealed that a network with two hidden layers (15 and 6 neurons in first and second layer, respectively) using Levenberg–Marquardt (LM) learning algorithm and tangent-sigmoid transfer function would provide an efficient response to predict the output parameter. A coefficient of determination (R2) of 0.9999 and a training error of 0.0009 resulted from the network training.
