CNN Inspection of Absorbent Articles for High-Speed Defect Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current quality control systems for manufacturing disposable absorbent articles are limited by human-generated inspection algorithms, which constrain the speed and scope of in-process inspection operations, making it difficult to conduct sophisticated inspections of assembled articles.
Innovation Solution
The implementation of a convolutional neural network-based inspection algorithm in conjunction with sensors, such as optical or acoustic sensors, to analyze images or sound recordings of absorbent articles, allowing for the detection of defects and execution of control actions like rejection or machine adjustments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If human-generated inspection algorithms are used, then the system is easier to implement, but the speed and scope of inspection operations are limited
Solution Approach 1:
The patent replaces human-generated inspection algorithms with an artificial intelligence-based inspection algorithm that uses machine learning models to automatically analyze sensor data and detect defects in absorbent articles. This substitution enables faster processing of inspection data and expands the scope of detectable defects without requiring manual algorithm development for each inspection scenario.
2Device complexity
If human-generated inspection algorithms are used, then the system complexity is lower, but the scope of sophisticated inspections is constrained
Solution Approach 1:
The patent implements a universal AI-based inspection algorithm that can detect multiple types of defects across different absorbent article components using a single system. The machine learning model is trained to recognize various defect patterns including missing components, improper positioning, and manufacturing defects, enabling the system to perform sophisticated inspections of assembled articles without requiring separate specialized algorithms for each defect type.
3Measurement precision
If AI-based inspection algorithms are implemented, then inspection accuracy and speed improve, but computational requirements and system complexity increase
Solution Approach 1:
The patent performs preliminary training of the machine learning model using labeled sensor data from absorbent articles before deployment. During this offline training phase, the AI algorithm learns to recognize defect patterns and is optimized for real-time inspection. This preliminary action enables the system to achieve high inspection accuracy during production without requiring complex real-time computational resources, as the heavy processing occurs during the pre-training phase rather than during actual inspection operations.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
This approach enables more advanced and efficient in-process inspection, improving the accuracy and speed of defect detection and machine maintenance, reducing the need for human intervention and enhancing the overall quality control process.
Implementation Method 1
providing a sensor, preferably selected from the group consisting of an optical sensor, thermal sensor, and combinations thereof; the sensor being in data communication with the controller
Data Source
AI summary
A method for inspecting disposable absorbent articles, the method comprising: providing a controller comprising an inspection algorithm with a convolutional neural network; providing a sensor, preferably selected from the group consisting of an optical sensor, thermal sensor, and combinations thereof; the sensor being in data communication with the controller; providing a reference database, wherein the reference database comprises a plurality of pre-processed images of disposable absorbent articles and/or components thereof that have been labeled with one or more of: a feature or element of the article and/or component thereof; a non-acceptable defect within a feature or element of the article and/or component thereof; an acceptable defect within a feature or element of the article and/or component thereof; an acceptable faultless feature or element of the article and/or component thereof; and combinations thereof; and the method comprising the steps of: training the convolutional neural network preferably through a plurality of iterations or epochs: optionally validating the training, wherein the training step is repeated until a mean average precision score (MAP) of at least 0.9 is attained according to the formula (I) where Q is the number of queries: advancing one or more substrates through a converting process along a machine direction to form an array of absorbent articles; creating a plurality of images of said array of absorbent articles with the sensor; transmitting the images from the sensor to the controller; determining at least one of the properties for the labeled component with the inspection algorithm; cutting the substrate into discrete articles; and based on the determination of the least one of the properties for the labeled component, executing a control action, wherein the control action is selected from the group of automatic rejection of one or more articles, automatic machine-setting adjustment, a warning signal for machine maintenance scheduling. and a machine stop command.MAP=∑ q=1 QAP(q)Q(I)