Predicting Model of Quasi-Static Mechanical Properties of Almond using Wave-let Neural Network

Dimensions, shape and mechanical properties for the design of agricultural machinery for separation, harvesting and classification are important. The neural network as one of the main components of computational intelligence has important properties and wavelet transformation improves the accuracy of the model. In Research, by horizontal position of almonds, power failure, static brain, three varieties of almond named Mamaee, Rabie and Shahrood 12, in three levels of moisture (5.5%, 15% and 25% wb) and three speed loading (5,15 and 25 millimeters per minute) were determined. In the neural network, input parameters including three levels of humidity and three levels of speed as an independent variable were introduced into the model. The variables of fracture force, energy consumption and modulus of elasticity were considered as dependent variables of the model output. For more precision, various wave-lets such as Haar, db4, Sym2, Coif4 were first applied to these data. The R-criteria was used to evaluate the accuracy of the model. The cross validation results indicated the superiority of the Coif4 wavelet. In this algorithm, a feed forward network with a Levenberg-Marquardt algorithm with a sigmoid tangent function in the hidden layer and a linear function in the output layer were used. The results show the high accuracy of the neural network-wavelet. For the mamaee Variaty neural network with an arrangement of 1-5-2 with R = 0.9523, the arrangement 1-7-2 with R = 0.9745 and the arrangement of 1-4-2 with R = 0.8374 the best prediction for the modulus Elasticity has the power of failure and energy consumption. The R-criteria was used to evaluate the accuracy of the model. The cross validation results indicated the superiority of the Coif4 wavelet. In this algorithm, a feed forward network with a Levenberg-Marquardt algorithm with a sigmoid tangent function in the hidden layer and a linear function in the output layer were used. The results show the high accuracy of the neural network-wavelet.