Inkjet Nozzle Ejection Failure Detection Using Autoencoders
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting inkjet printer nozzle ejection failures require extensive parameter setting and collection of scarce failure data, making them time-consuming and inefficient.
Innovation Solution
An unsupervised learning method using an autoencoder to generate a trained model from normal ejection state images, allowing detection of ejection failures without requiring parameter setting or extensive data collection, by comparing input and output data errors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional numerical analysis methods are used for detecting ejection failures, then detection capability is provided, but extensive parameter setting is required making the process time-consuming
Solution Approach 1:
The patent replaces conventional numerical analysis methods with machine learning-based automated detection. The system uses an imaging device to capture print images and automatically processes them through machine learning algorithms to identify ejection failures, eliminating the need for manual parameter setting and significantly reducing detection time while maintaining high accuracy.
2Measurement precision
If supervised learning methods are used for ejection failure detection, then detection accuracy is improved, but collection of scarce failure data becomes extremely difficult and time-consuming
Solution Approach 1:
Instead of collecting scarce failure data and using supervised learning with labeled failure examples, the patent inverts the approach by using unsupervised learning. The system learns from normal print images and automatically detects abnormalities (ejection failures) without requiring any pre-labeled failure data, thus eliminating the time-consuming data collection process while maintaining detection accuracy.
Solution Approach 2:
The machine learning model performs self-learning from normal print images and automatically identifies ejection failures without requiring external intervention or pre-labeled data. The system serves itself by autonomously detecting anomalies in the print quality, eliminating the need for manual failure data collection and labeling.
3Measurement precision
If extensive parameter setting is performed for detection methods, then detection capability is achieved, but the process becomes complex and time-consuming
Solution Approach 1:
The patent replaces complex manual parameter setting with automated machine learning-based detection. The system uses imaging devices and computational algorithms to automatically identify ejection failures without requiring users to configure multiple parameters, thereby reducing operational complexity while maintaining high detection capability.
Data Source
Figure 1~3
Figure 4
Figure 5
AI summary
Regarding an inkjet printing apparatus, it is possible to accurately detect an ejection failure of a nozzle without requiring much time and effort for generating and collecting learning data and setting a parameter. By performing unsupervised learning by an auto encoder using only images representing a favorable ejection state, a trained model for detecting an ejection failure of a nozzle is generated (S20). A plurality of partial images are cut out from a captured image of a print image of a test pattern (S50). The partial image is given as input data to the auto encoder constituting the trained model (S60), and a mean square error between the input data and output data is calculated (S70). Based on a result of comparing the mean square error with a threshold, it is determined whether or not an ejection failure nozzle is included in one or more nozzles corresponding to the input data (S80).