Application of artificial neural network model to predict limiting current for cobalt magneto-electrodeposition

Abstract

Magneto-electrodeposition (MED) is one of cobalt electrodeposition technique which able to produce more uniform, denser, and finer deposition on cobalt surface. MED is an electrodeposition technique carried out under the influence of a magnetic field. This technique was also able to increase the mass transfer which indicated by the increase of limiting current. The method to determine the limiting current is very important in MED because the optimum mass transport happens at the limiting current. One model which able to predict the limiting current simply and easily is a neural network model. Another alternative in predicting the limiting current (iB) is using artificial neural networks (ANNs), one of the ANNs used in this study is the feed forward neural network (FFNN) with the multiple-input-single-output (MISO) model. This MISO FFNN has eight input variables and one output. The data was obtained from the results of semi- empirical model experiments which are then modeled with the best FFNN model. In order to get the best model of FFNN, three different identification algorithms (Levenberg-Marquardt, Bayesian Regularization, and Scaled Conjugate Gradient algorithm) were used. In this work, the number of hidden nodes were varied from 10 to 50. The best model obtained is the FFNN which uses the Levenberg-Marquardt algorithm with 20 hidden nodes. The result show that the FFNN model has a good performance to simulate the limiting current which shown by the small means square error (MSE) value when it compared with the limiting current form experiment. The final step in this study is to create an FFNN model using Simulink to make it easier to run the model.