Machine Learning Defect Detection in Liquid Food Packaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing defect detection methods in packaging containers for liquid food are inefficient and disrupt production lines, making it difficult to accurately identify defects at high speeds without significant resource allocation.
Innovation Solution
A method and system that captures image data of packaging containers, uses machine learning to detect defects, determines timestamps for defect occurrence, and correlates defects with production parameters, enabling on-the-fly analysis with minimal production disruption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional defect detection methods are used in high-speed filling machines, then production throughput is maintained, but defect detection accuracy and reliability deteriorate
Solution Approach 1:
The system performs preliminary actions by capturing multiple images of packaging containers at different positions and orientations before final defect classification. The machine learning model is trained in advance with extensive defect data to enable rapid classification during production, allowing accurate defect detection without slowing down the high-speed production line.
Solution Approach 2:
The system creates optical copies (images) of the packaging containers using imaging devices, and creates digital copies of defect patterns in the machine learning model's training data. This allows virtual analysis and classification of defects without physically interrupting the production flow, maintaining throughput while improving detection accuracy.
2Reliability
If comprehensive defect analysis is performed on all packaging containers, then defect detection reliability is improved, but resource consumption and production disruption increase
Solution Approach 1:
The system applies local quality by focusing detailed image analysis and machine learning classification only on regions of the packaging containers that show defect indicators. The machine learning model is trained to identify and focus on specific defect patterns rather than analyzing every pixel of every container, reducing computational resources while maintaining high reliability for actual defect detection.
Solution Approach 2:
The system performs partial action by implementing defect detection at strategic points in the production line rather than continuous analysis of every container. The machine learning model processes only the necessary image features required for defect classification, avoiding excessive computation while maintaining sufficient reliability for quality control.
3Measurement precision
If multiple imaging devices and complex analysis systems are deployed, then defect detection capability is improved, but device complexity and implementation difficulty increase
Solution Approach 1:
The machine learning model serves multiple functions: it classifies different types of defects (sealing defects, contamination, packaging damage), works with images from multiple imaging devices, and can be trained on various defect patterns. This universal approach consolidates what would otherwise require multiple specialized systems into a single flexible platform, improving defect detection capability while managing system complexity.
Solution Approach 2:
The system merges multiple imaging devices and their data streams into a unified machine learning analysis pipeline. Rather than having separate analysis systems for each imaging device, the patent combines all image inputs and defect detection functions into an integrated machine learning model, simplifying the overall system architecture while enhancing comprehensive defect detection capability.
4Productivity
If real-time defect detection is implemented without production disruption, then productivity is maintained, but measurement precision and defect identification accuracy deteriorate
Solution Approach 1:
The system ensures continuity of useful action by implementing defect detection that operates continuously during production without stopping the filling machine. The machine learning model processes images in real-time as containers move through the production line, maintaining uninterrupted production flow while achieving accurate defect identification through advanced image analysis and classification algorithms.
Data Source
AI summary
A method of defect detection in packaging containers for liquid food is disclosed, where packaging containers are produced in a machine. The method comprises capturing image data of the packaging containers, defining image features in the image data representing defects in the packaging containers, associating the image features with different categories of defects, inputting the image features to a machine learning-based model for subsequent detection of categories of defects in packaging containers based on the image features, determining time stamps for the occurrence of defects in said subsequent detection, determining associated production parameters of the packaging containers in the machine for the occurrence of defects based on the time stamps, and correlating said occurrence and category of the defects with said production parameters. A system for defect detection in packaging containers is also disclosed.