การประยุกต์ใช้โครงข่ายประสาทเทียมเพื่อพยากรณ์อาการเสียของเครื่องจักรการทำชั้นในตัวเก็บประจุ

Abstract

Due to machine breakdown incident to loss in manufacturing process. One of maintenance activity to reduce breakdown loss is the preventive maintenance. It is effectively activity in normal production. When demand is higher than machine capacity, machine loss is uncontrollable. This study offers the methodology of the optimal of Backward propagation neural network for forecasting the next daily summation of machine lost time at a Capacitor Stacking Machine. This neural network has 7 input layers and an output layer. It is the next daily sum of machine lost time at the Capacitor Stacking machine and find-out the optimal layer 3,4 and 5 layers. And the optimal node in each hidden layer. There are 10 and 100 nodes in the hidden layer.Finally, the optimal training method that machine with the machine gathering data. There are 3 methods in this study split-test training, 5-fold Cross Validation and 10-fold Cross Validation. The proper neural network structure makes accuracy forecasting value in the next period. In the study, the 5 layers of Backward propagation neural network at 7-100-10-10-1 with the split-test training give RMSE at 51.293 minutes the computational runtime is 5.405 second. This study shown good git to decision maker form data-driven framework must trader-off the smallest error or short computational runtime depend on the purpose of application.