PDEng Position on Fault diagnosis of professional printers

The ICT post-master designers program is a two-year salaried program in the field of technological design in Electrical Engineering. The program leads to a Professional Doctorate in Engineering (PDEng) degree.

Project description

An important part of the PDEng programme is the design project which will be performed at Océ technologies. The project is spread over two years and aims to develop your design engineers' abilities. In the project, a specific design objective must be reached within a restricted period of time and with limited means. In accordance with the supervising professor the candidate will be given the opportunity to gather specialized knowledge for the design project already in the first year of your Traineeship. You will be coached by experienced design engineers from industry and/or by TU/e staff with clear and relevant design experience. You will acquire independence and learn to make choices and work well in a project-based manner.

Canon Production Printing develops several high-end digital inkjet printer systems for the high-volume and graphic arts professional printing markets. Here, the demands on print quality are stringent, i.e., no print quality artefacts and consistent good quality prints at high speed with little to no downtime of the printing system. Print quality depends on precisely jetting ink droplets on moving media (e.g. paper), how the image is subsequently formed by interactions between droplets and the media, and finally the ink solidification process. Hence, many mechanical, physical and chemical interaction processes play a role in creating a high quality print.

Each printer has a printer health management architecture which attempts to ensure constant printing quality. Currently, near-real-time acoustic wave sensing is applied to establish the condition of jetting nozzles. Decision tree analysis of the acoustic impulse response results in a known nozzle status and an estimation of the print quality.

Canon Production Printing is working on next-generation printer health architecture. Therefore, an advanced imaging sensor is under development to provide real-time data on print quality in addition to the acoustic measurements. Both methods result in large amounts of data that are excellently suited to be classified by deep learning algorithms (e.g., convolutional neural network). Due to the many variables governing the print quality and (expected) availability of sensor data, application of deep learning classification to acoustic and imaging data is necessary.

This PDEng project will consist of; deep learning algorithm development, simulation of print quality artifacts on a virtual printer and hands-on experimentation on printing systems.

For acoustic data, deep learning can be applied directly to the existing datasets. The goal is to replace the decision tree classification process with a deep learning classification algorithm.

In case of the imaging sensor creating a representative training set of high quality using a real printer is infeasible. Not only from the perspective of the time that would be needed, but also because it is difficult to reproduce many of the relevant print quality artifacts with a real printer. In order to obtain training sets the candidate will use a virtual printer model to create a suitable high-quality training set based on simulation data and subsequently train a deep learning model for print artifact detection. The effectiveness of the deep learning model will be tested using real printer data obtained using the scanning system.

The result of both approaches would be a closed loop print quality control for future printing systems.