Session 1 – 23 March

Real-time print artifact detection using deep neural networks

16:15 – 16:55

Most common artifacts in printing are caused by some form of nozzle failure. We can apply various compensation tactics to improve image quality or we can perform targeted maintenance actions to fix the failing nozzle. However, all of this hinges on the machine’s ability to detect the problem.

With help of a scanner, we can analyze the printed material. The easiest method would be to print a test pattern and look for missing nozzles. However, such patterns aren’t productive and a waste of material. In this talk, we’ll discuss a method for nozzle failure detection using machine learning directly in customer prints.

Pim Hacking is a relatively fresh graduate working at Canon Production Printing. In 2018, he joined the company as a graduate student to develop a classification/parameter extraction algorithm based on nozzle sensor data. After graduation, he continued as an employee to work on the subject adding machine learning elements. In parallel, he co-supervised a PDEng student working on print artifact detection using machine learning.

Pim Hacking

R&D engineer at Canon Production Printing