Development and Test of a Data Framework for Prediction of Soldering Quality in Selective Wave Soldering Applying K-Nearest Neighbors

摘要

Machine Learning has been proven to be a powerful tool to model and predict complex applications. Selective wave soldering is a widely applied interconnection technology for THT components. It is mostly used if components are not substitutable by surface mount devices due to high thermal load or mechanical stress. Especially in power electronic circuit boards, large copper layers or high thermal mass components lead to critical soldering situations. This paper suggests a Machine Learning framework to identify thermally challenging solder joints. The hybrid approach consisting of an analytical thermal description of THT components and solder joints in the multilayer circuit board and the ML analysis allows the prediction of arbitrarily complex solder joint configurations. The data framework represents electronic components and solder joints. Utilizing solder joint, component and soldering process parameters as input, the K-Nearest Neighbors algorithm predicts the probable hole fill following IPC-A-610 with an overall accuracy of about 75%.