Prediction of the Solder Rise in Selective Wave Soldering Comparing Decision Tree and Logistic Regression

摘要

Selective wave soldering is a common method for soldering through hole components on PCB boards. With PCB design becoming increasingly more complex and copper layer thickness rises, the solderability becomes difficult due to high heat losses. Hence, the design phase and parameterization of machines on the shop floor is time consuming because the choice of soldering parameters for a reliable connection according to IPC-A-610 is mostly done based on experience and trial and error. In this paper a data framework for machine learning models is presented that allows the storage of experiential knowledge for new designs. The data framework is then used to structure experimental data from a test board. A decision tree model and a logistic regression model for classification of the data are applied and the prediction quality is evaluated. It is shown that the solder rise can be predicted based on the chosen data base. Especially logistic regression performs well for the given dataset.