Sudirman Rizki Ariyanto, Retno Wulandari, Suprayitno
Artificial Neural Network (ANN) is a modeling approach that is widely applied in various fields. However, the main problem in using ANN is on the tuning ANN performance parameters because there is no definite reference to choose the definite parameters. In this study, the Taguchi method is offered as the alternative problem solving. The Taguchi method was chosen because of its ability to find robust parameter combination. There are five ANN parameters that are used as the inputs where each parameter consists of three levels. The tuning process of parameters of ANN performance criteria was carried out using the Taguchi method with the L18 Orthogonal Array (OA) design with five replications for each combination. Signal to noise ratio (S/N) is adopted as the quality measure for robustness index. 231 data set of metal catallitic converter designs are used as training sample. The optimization results show that the optimum combination of ANN performance parameters consists of (1) three hidden layers; (2) ten neurons of each hidden layer; (3) log sigmoid transfer function for hidden layer; (4) tan sigmoid transfer function for output layer; and (5) learning algorithm using Bayesian Regularization. © 2023 Author(s).
Department of Mechanical Engineering, Universitas Negeri Malang, Malang, Indonesia