A case study of SOS-SVR model for PCB throughput estimation in SMT production lines

This paper presents a symbiotic organism search (SOS) algorithm based support vector regression (SVR) model for predicting the production throughput in a high-mix and low-volume setting. A case study of the printed circuit broad (PCB) throughput in surface mount technology (SMT) production lines is conducted. The accurate throughput estimation of PCB order is essential to optimize the SMT production schedule, which is more critical in high-mix and low-volume environment. However, it is difficult to estimate the throughput due to the flexibility, complexity, and uncertainty of the PCB assembly process. Therefore, the SVR model, one of the most efficient machine learning tools, is proposed to solve this problem. To improve the performance of SVR model, the SOS algorithm is applied to optimize SVR parameters, which include the penalty, the width of kernel function, and the loss function. The training and testing data sets are collected from a local electronic manufacturer. Compared with the industrial solution and classic SVR, the proposed model shows its efficiency.