Deep learning for zero-defect inkjet-printing of electronics

We present a vision system for automatic calculation of inkjet printed electronic structures. By adapting the printed structures to individual parts, the system is able to correct misalignments and deviations from previous process steps simply by adjusting the printed image. We propose a vision system that acquires high-resolution images after the pick & place step. A deep neural network is used to read this information and output a suitable print image for the subsequent step of inkjet printing. We evaluate the vision system in a set of experiments. The system reaches an intersection over union (IoU) of 82% on our data set. This shows the potential for future zero-defect additive manufacturing in electronics industry.