Multi-scale conditional diffusion model for deposited droplet volume measurement in inkjet printing manufacturing

Abstract

The volume consistency of deposited droplets is regarded as a core quality evaluation to achieve zero-defect manufacturing (ZDM) in inkjet printed organic light-emitting diode (OLED). In order to improve the product quality, the causal links between printing defects and its causes should be determined. However, the collection of the defects quantified data remains a challenge due to the micrometer scale and few texture features of droplets. In this paper, an accurate measurement method for deposited droplets based on coherence scanning interferometry (CSI) is proposed. Firstly, time series signals for topography reconstruction are recorded when CSI scans droplets vertically at a constant speed. Then, a deep learning method, multi-scale conditional diffusion model (MS-CDM), is proposed to restore the time series signals’ distortions led by the chromatic dispersion. Furthermore, a self-supervised learning method is proposed to efficiently train the MS-CDM, since the labeled data is hard to be acquired in industrial environment. Finally, the printing experiment indicates the collaboration of different modules in intelligent close-loop to reduce Mura defects. The volume measurement experiments demonstrate the ability to accurately measure the volume of deposited droplets, and the measurement error reaches 1.3%. The performance experiment on several public datasets demonstrates the method’s promising feature representation ability for time series mainstream task, and the MS-CDM achieves the best performance compared to those baselines methods.