Tracking and tracing are a key technology for production process optimization and subsequent cost reduction. However, several industrial environments (e.g. high temperatures in metal processing) are challenging for most part-marking and identification approaches. A method for printing individual part markings on metal components (e.g. data matrix codes (DMCs) or similar identifiers) with high temperatures and chemical resistance has been developed based on drop-on-demand (DOD) print technology and special ink dispersions with submicrometer-sized ceramic and glass particles. Both ink and printer are required to work highly reliably without nozzle clogging or other failures to prevent interruptions of the production process in which the printing technology is used. This is especially challenging for the pigmented inks applied here. To perform long-term tests with different ink formulations and to assess print quality over time, we set up a test bench for inkjet printing systems. We present a novel approach for monitoring the printhead's state as well as the print-quality degradation. This method does not require measuring and monitoring, e.g. electrical components or drop flight, as it is done in the state of the art and instead uses only the printed result. By digitally quantifying selected quality factors within the printed result and evaluating their progression over time, several non-stationary measurands were identified. Some of these measurands show a monotonic trend and, hence, can be used to measure print-quality degradation. These results are a promising basis for automated printing system maintenance.

Ceramic inks can be used to mark metal sheets in hot forming for track-and-trace purposes. However, the ceramic pigments in the inks can lead to clogging of printer nozzles which results in loss of print quality. Here we report on a predictive maintenance concept including different machine-and deeplearning models as the basis of a print quality assurance strategy. Pixelwise image segmentation leads to detailed information about the printing results. The information is used to train a model, classifying the remaining useful lifetime until insufficient printing results.