Deep Learning-Based Inkjet Droplet Detection for Jetting Characterizations and Multijet Synchronization.

Area of Science:

• Materials Science and Engineering
• Artificial Intelligence
• Manufacturing Technology

Background:

• Inkjet printing offers precise material deposition but faces challenges in predicting jetting due to complex ink-nozzle-air interactions.
• Current manual monitoring by engineers is labor-intensive and limits scalability in industrial applications.
• Ensuring process reliability in high-volume manufacturing, such as electronics, requires automated characterization methods.

Purpose of the Study:

• To develop a deep learning-based method for autonomous, real-time jetting characterization in inkjet printing.
• To overcome the limitations of manual monitoring and reduce labor costs associated with process reliability checks.
• To enable synchronized multinozzle jetting for advanced manufacturing applications.

Main Methods:

• Utilized an in situ CCD camera to record inkjet printing processes.
• Employed YOLOv5, a convolutional neural network (CNN)-based object detection model, to identify individual droplets.
• Developed a regression analysis to quantify droplet parameters like velocity, diameter, length, and translation.

Main Results:

• The YOLOv5 model achieved high performance with precision (0.86), recall (0.89), and mean average precision (mAP) of 0.90 at a 0.5 IoU threshold.
• Quantified droplet characteristics were accumulated chronologically for each nozzle and droplet class.
• Demonstrated the ability to synchronize multinozzle jetting based on real-time characterization data.

Conclusions:

• The deep learning approach enables autonomous real-time process testing for inkjet printing.
• This method significantly reduces labor costs and enhances process reliability in large-scale manufacturing.
• Facilitates high-resolution patterning for applications like biosensor electrodes and QD display pixels by leveraging big data from jetting characterizations.