Optimum Milling Parameters of Sugarcane Juice Production Using Artificial Neural Networks (ANN). (February 2019)

• Record Type: Journal Article
• Authors: Oktarini, D; Mohruni, A S; Sharif, S; Yanis, M; Madagaskar,
• Is Part Of: Journal of physics. Volume 1167(2019)
• Publication Date: 2019-02
• Subjects: Physics -- Congresses; 530.5
• DOI: 10.1088/1742-6596/1167/1/012016
• Languages: English
• ISSNs: 1742-6588

Abstract

Abstract: High population growth has an impact on increasing the need for sugar cane from year to year. Cintamanis as a sugarcane producer company continues to increase production by reaching the highest production target. In achieving the production target, Cintamanis has a sugarcane milling system consisting of 5 grinding machines that process sequentially. The five machines have three rollers of rollers, the top rollers, the front rollers, and the rear rollers that have a certain distance by the number of cane crops to be processed. The purpose of this research is to optimum milling parameters of sugarcane juice using artificial neural networks. The prediction of this sugarcane milling process uses the input variables of each roll that is found on the milling machine. The approximate procedure begins by calculating each distance between rollers. Then count the amount of sugarcane juice produced. Next, select the input variables that provide a significant correlation to the output variables. Then proceed with designing the optimum network structure and choose the learning rate and momentum. The validation process is performed on the optimum network structure to determine the accuracy level of the sugarcane milling process. The selected backpropagation neural network model is a model with eight inputs, 1 hidden layer (with 8 neurons), and 1 output using binary sigmoid activation in training and linear activation function at the output. Based on testing the maximum number of iterations obtained the lowest MAPE value of 17.85% with the number of iterations 800. Moreover, in the test of learning rate obtained the lowest MAPE value of 17.38% with the value of learning rate 0.4. If the maximum iteration value of 800 and the value of learning rate 0.4 it will result in MAPE value of 16.98%.