Short Review on Machine Learning Optimization Methods in Surface Mounted Electronics Assembly Technologies

Machine learning (ML) is increasingly adopted to support the production process of surface-mounted electronics assembly technology (SMT). In the literature, extensive research was carried out to enhance automated optical inspection (AOI) by applying ML algorithms, like artificial neural networks. However, predicting production defects based on data from the sensors built into the production equipment is still not that widespread. Furthermore, applying ML-based approaches may allow the optimization of production control by fine-tuning process parameters of stencil printing or component placement, for example. Many research papers were surveyed to identify the strong and weak points of current approaches in this area. A clear overview was given to aid the proper classification of recent results and the applicability of ML-based methods in SMT. Our results contain not only the review of multiple approaches, but a comparison of the numerical results, the complexity of the experimental environment and the extent of applicability are also included.

I. Introduction

As ML algorithms got more efficient in recent years, they got involved in failure detection by AOIs [1]. However, electronics manufacturers aim not only to enhance failure detection but also to reduce production errors. Such novel technologies like ML should also support increasing the first pass yield and approaching zero defect manufacturing in the scope of Industry 4.0 [2]. The process for surface mount assembly (SMA) is highly standardized. Possible inspection points for failure detection are clearly defined after stencil printing and component placement since these are the most critical steps in SMA (Fig. 1). Fig. 1.

Schematic overview of SMT production line [1]