The scale predictions of multilayer PCB board by employing machine learning

Abstract

The purposes of this research were to predict and assign a scale value of multilayer PCB. After the PCB board underwent a process of lamination press, it made its be sized according to customers’ needs. The collected data came from KCE electronics public company limited were the configuration data of the scale of multilayer PCB board which went back from January 2018 to June 2019. Variables we studied consisted of two variables were dependent and independent variables. The dependent variables were the percentage scale values of multilayer PCB board and the independent variables were factors affecting determination of the percentage scale values of multilayer PCB board. Those factors were layer count, distance, core thickness, thickness cu side (thickness of copper on laminate core board), thickness Cu Foil, Tg (glass transition temperature of material), cutdirection, Axis, streak (Warp x Fill) and material brand. 

The models we used for analyzing were a multiple linear regression method (MLR), a support vector regression method (SVR) and a decision tree regression method. Those methods were the supervised learning models in machine learning and they were processed by using RStudio program, mean square error (MSE) and mean absolute percent error (MAPE) for comparison of the efficiency of models for scale predictions. The result revealed that the MSE and the MAPE value of the support vector regression model are minimal, which means it is the most suitable model for the data of scale predictions of multilayer PCB board due to helping increase the accuracy in assigning scale values and helping save time of the production process as a consequence of errors in assigning scale values.