An Improved Momentum Rate in Artificial Neural Networks for Estimating Product Cycle Time at Semi-automatic Production

Among all the prediction techniques, the Artificial Neural Networks (ANN) shows excellent performance. The ANN technique has a momentum rate to slow down the ANN learning process. However, the value of the momentum rate has no restriction since it is commonly based on the experiment with different values as presented in the previous studies. In this regard, the objective of this study is to formulate a momentum rate to achieve a better prediction result. The proposed momentum rate equation was tested on three ANN models. Subsequently, the 3-2-1 network emerged as the best network based on the smallest mean square error. To evaluate the proposed momentum rate, a problem based on a real company situation in producing audio products was considered. Cycle time of the new audio products at its semi-automatic production line was predicted based on several factors, which were manpower shortage, material preparation time and machine breakdowns through the 3-2-1 network. As a result, the best cycle time to complete new audio products can be estimated accurately. In conclusion, the proposed momentum rate can improve the convergence of the ANN learning process for a better prediction result. Consequently, audio products delivery is smooth and fulfil customer’s demands.