Screen printing has been adopted for fabricating a wide variety of electronic devices. However, the printing defects and reliability have been an obstacle for industrialization of printed electronics. In this research, the artificial intelligence (AI) model was developed and integrated with the in-house roll-to-roll screen printing system to detect smearing defect, which is one of the main defects of screen printing. The U-Net architecture was adopted, and a total of 19 models were designed with model sizes ranging from 8E + 3 to 3E + 7 number of parameters. Their performances as validation mean Intersection over Union (IoU) were analyzed, and the optimal model was chosen with a validation mean IoU of 95.1% and a number of parameters of 8E + 6. The printed line images were evaluated by the AI model for various printing conditions, such as printed line widths, printing paste premixing, printing speeds, and printed line directions, which showed that the model could effectively detect the smearing defects. Also, the AI model capabilities were investigated for repeated printing, which demonstrated that it can be used for the reliability assessment of the screen printing process.

This study presents a hybrid fabrication process for LED matrix on flexible multilayer printed circuit boards (PCBs). To prepare for the flexible PCB, a roll-to-roll screen printing process was developed to create a core circuit board by printing on both sides of the flexible substrate. From the core PCB, the lamination and etching processes are then used to form four layers in total. By using the proposed method, we successfully implemented the passive LED matrix display. This approach could be used for other applications that require flexible and multilayer circuits.

Roll-to-roll (R2R) printing methods are well known as additive, cost-effective, and ecologically friendly mass-production methods for processing functional materials and fabricating devices. However, implementing R2R printing to fabricate sophisticated devices is challenging because of the efficiency of material processing, the alignment, and the vulnerability of the polymeric substrate during printing. Therefore, this study proposes the fabrication process of a hybrid device to solve the problems. The device was created so that four layers, composed of polymer insulating layers and conductive circuit layers, are entirely screen-printed layer by layer onto a roll of polyethylene terephthalate (PET) film to produce the circuit. Registration control methods were presented to deal with the PET substrate during printing, and then solid-state components and sensors were assembled and soldered to the printed circuits of the completed devices. In this way, the quality of the devices could be ensured, and the devices could be massively used for specific purposes. Specifically, a hybrid device for personal environmental monitoring was fabricated in this study. The importance of environmental challenges to human welfare and sustainable development is growing. As a result, environmental monitoring is essential to protect public health and serve as a basis for policymaking. In addition to the fabrication of the monitoring devices, a whole monitoring system was also developed to collect and process the data. Here, the monitored data from the fabricated device were personally collected via a mobile phone and uploaded to a cloud server for additional processing. The information could then be utilized for local or global monitoring purposes, moving one step toward creating tools for big data analysis and forecasting. The successful deployment of this system could be a foundation for creating and developing systems for other prospective uses.

Electrohyrodynamic (EHD) jet printing has been widely used in the field of direct micro-nano patterning applications, due to its high resolution printing capability. So far, vector line printing using a single nozzle has been widely used for most EHD printing applications. However, the application has been limited to low-speed printing, to avoid non-uniform line width near the end points where line printing starts and ends. At end points of line vector printing, the deposited drop amount is likely to be significantly large compared to the rest of the printed lines, due to unavoidable acceleration and deceleration. In this study, we proposed a method to solve the printing quality problems by producing droplets at an equally spaced distance, irrespective of the printing speed. For this purpose, an encoder processing unit (EPU) was developed, so that the jetting trigger could be generated according to user-defined spacing by using encoder position signals, which are used for the positioning control of the two linear stages.

In printed electronics, flawless printing quality is crucial for electronic device fabrication. While printing defects may reduce the performance or even cause a failure in the electronic device, there is a challenge in quality evaluation using conventional computer vision tools for printing defect recognition. This study proposed the computer vision approach based on artificial intelligence (AI) and deep convolutional neural networks. First, the data set with printed line images was collected and labeled. Second, the overall printing quality classification model was trained and evaluated using the Grad-CAM visualization technique. Third and last, the pretrained object detection model YOLOv3 was fine-tuned for local printing defect detection. Before fine-tuning, ground truth bounding boxes were analyzed, and anchor box sizes were chosen using the k-means clustering algorithm. The overall printing quality and local defect detection AI models were integrated with the roll-based gravure offset system. This AI approach is also expected to complement more accurate printing reliability analysis firmly.

This study investigates dataset‐adaptive optimisation strategies for a Single Shot MultiBox Detector (SSD) network, focusing on a critical case study of parts detection on a manufacturing line. We address the challenges of achieving high‐performance object detection in real‐time industrial environments by implementing a structural optimisation approach that dynamically reduces feature maps and prior boxes based on dataset characteristics. The network self‐identifies and prunes redundant features by analysing the statistical properties of the images and objects in the datasets, leading to a leaner and more efficient SSD network. This dataset‐informed reduction of feature maps and anchor boxes accelerates training and inference speeds and enhances detection precision, which is crucial for accurate part identification in high‐throughput production. Experimental results on a custom manufacturing line parts dataset, consisting of multi‐coloured and differently sized bolts and nuts, demonstrate a mean Average Precision (mAP@50) of approximately 96%‐99%, while the inference time was reduced by more than four times (to approximately 3–8 ms in the study case) according to feature extraction backbones, compared to the original network.