Deep Learning Model for Automated Visual Inspection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated visual inspection systems for pharmaceuticals face challenges such as large footprint, high complexity, and high costs due to the need for multiple camera stations and complex image processing algorithms, which also require substantial resources for maintenance and adaptation to new product lines.
Innovation Solution
The implementation of deep learning techniques to reduce the number of camera stations, improve defect detection accuracy, and simplify the adaptation process by training models to detect various defects simultaneously, using techniques like heatmap analysis and synthetic image generation to enhance training image libraries and reduce processing demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If multiple camera stations with specialized image processing algorithms are used to detect various defects, then defect detection accuracy is improved, but device complexity and footprint increase
Solution Approach 1:
The patent combines multiple specialized image processing algorithms into a single deep learning model that can detect multiple defect types simultaneously. The neural network is trained to identify container defects, cosmetic defects, and drug product defects in one unified system, replacing the traditional multi-station approach with 15+ separate camera stations.
Solution Approach 2:
The deep learning model serves multiple functions by detecting various defect categories (container integrity, cosmetic defects, foreign particles, liquid characteristics) through a single inspection station. This universal detector eliminates the need for specialized algorithms for each defect type, reducing system complexity while maintaining comprehensive inspection capabilities.
2Adaptability or versatility
If multiple camera stations are deployed to cover full range of defects, then defect detection coverage is improved, but manufacturing cost increases
Solution Approach 1:
The patent merges the functionality of 15 separate camera stations into a single inspection station equipped with a deep learning model. This consolidation reduces the number of cameras, lighting systems, and mechanical components required, thereby significantly lowering manufacturing costs while maintaining comprehensive defect detection coverage across multiple categories.
Solution Approach 2:
The system uses synthetic training images generated through computer graphics to create virtual representations of various defect types. These synthetic images allow the deep learning model to be trained on a diverse range of defect scenarios without requiring physical samples of every possible defect, reducing the need for expensive physical test artifacts and enabling broader defect coverage at lower cost.
3Reliability
If specialized AVI equipment is used for comprehensive inspection, then inspection capability is improved, but maintenance burden and resource requirements increase
Solution Approach 1:
The patent consolidates multiple specialized inspection functions into a single deep learning-based system. This unified architecture reduces the number of separate components that require maintenance, simplifies troubleshooting, and eliminates the need for highly specialized engineers to qualify and commission each individual camera station. The system can be adapted to new product lines by retraining the model rather than reconfiguring multiple specialized stations.
4Productivity
If traditional image processing algorithms are used at each station, then processing speed meets production requirements, but system footprint and complexity increase
Solution Approach 1:
The patent combines multiple sequential inspection stations into a single station with a deep learning model that processes images in real-time. The neural network's parallel processing capability allows it to analyze multiple defect types simultaneously from a single image, achieving the same throughput as multiple stations while occupying minimal space and eliminating the need for extensive mechanical conveyance systems.
Data Source
AI summary
Techniques that facilitate the development and/or modification of an automated visual inspection (AVI) system that implements deep learning are described herein. Some aspects facilitate the generation of a large and diverse training image library, such as by digitally modifying images of real-world containers, and/or generating synthetic container images using a deep generative model. Other aspects decrease the use of processing resources for training, and/or making inferences with, neural networks in an AVI system, such as by automatically reducing the pixel sizes of training images (e.g., by down-sampling and/or selectively cropping container images). Still other aspects facilitate the testing or qualification of an AVI neural network by automatically analyzing a heatmap or bounding box generated by the neural network. Various other techniques are also described herein.