High-mix/low-volume productions, i.e., the manufacturing of many customized part series, each in relatively low quantity, will become substantially more important in the future as part of industry 4.0 alongside traditional mass production. One problem with such production is automated, computer-assisted quality control. Current QC software normally has to be recalibrated for every new part in order to achieve satisfactory results. The latest algorithms for detecting defects through machine learning also require extensive, annotated training datasets that, when producing lower quantities, are difficult to generate.
Fraunhofer Singapore is addressing these problems by developing new machine learning processes that only require very little annotated training data and can still achieve high detection accuracy. The developed processes are also flexible enough to be applicable to many different components (such as PCBs and soldered electrical parts) without costly calibration.