PCB Assembly Component Detection With Size-Specific AI Classifiers
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to effectively appraise assemblies of electronic components on a PCB due to their varying sizes and the impracticality of using high-resolution images for training and testing AI classifiers, leading to memory limitations and inaccurate object detection.
Innovation Solution
A system of pre-trained AI classifiers is employed to detect elements of different size ranges by sequentially using classifiers from largest to smallest, preventing overlap and using a top-down approach to identify components accurately.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If high-resolution images are used to train and test AI classifiers for detecting electronic components, then detection accuracy is improved, but memory requirements increase and become impractical
Solution Approach 1:
The patent divides the assembly into multiple size ranges (first size range for larger components, second size range for smaller components) and trains separate AI classifiers for each range. This segmentation allows each classifier to work with appropriately scaled images, reducing memory requirements while maintaining detection accuracy for components of different sizes.
Solution Approach 2:
The patent introduces a size range dimension by creating multiple classifiers tailored to different component size ranges. Instead of using a single classifier on high-resolution images, the system processes images through multiple classifiers with different size range specializations, effectively adding a dimensional approach to handle the complexity of varying component sizes without requiring excessive memory.
2Device complexity
If a single AI classifier is used to detect all electronic components, then device complexity is reduced, but detection precision decreases due to varying component sizes
Solution Approach 1:
The patent segments the detection task into multiple classifiers, each specialized for a specific size range of electronic components. This segmentation improves detection precision for each component size category while keeping each individual classifier relatively simple, balancing overall system complexity with detection accuracy.
Solution Approach 2:
Each AI classifier is trained with local quality in mind, being specialized for a specific size range of components. The first classifier handles larger components (first size range) while the second classifier handles smaller components (second size range), allowing each classifier to optimize its detection precision for its designated size range rather than attempting to handle all sizes equally.
3Measurement precision
If multiple AI classifiers are used to detect components of different size ranges, then detection precision is improved, but device complexity increases
Solution Approach 1:
The patent applies preliminary action by determining the size range of each component before applying the appropriate classifier. The system identifies whether a component falls within the first size range or second size range beforehand, then selects the corresponding pre-trained classifier. This preliminary classification step organizes the complexity systematically, improving detection precision while managing device complexity through structured workflow.
Solution Approach 2:
The patent changes the parameter of classifier selection based on component size range. Instead of using a fixed classifier for all components, the system dynamically selects which classifier to apply based on the size parameter of the component being detected. This parameter-based selection approach improves detection precision for each size category while managing system complexity through conditional logic rather than requiring all classifiers to run simultaneously.
4Reliability
If smaller classifiers search within larger classifiers' bounding boxes, then detection completeness is improved, but false positives increase
Solution Approach 1:
The patent extracts and removes the bounding boxes detected by the first classifier from the search space before applying the second classifier. By taking out the regions already identified as containing larger components, the system prevents the second classifier from redundantly detecting components within those same regions, thereby reducing false positives while maintaining detection completeness for components outside those bounding boxes.
Data Source
AI summary
A processing module for a communication device includes memory for storing a key and processing elements. These elements receive a current sequence number and a random number from a network entity, and check if the current sequence number is within a range based on a previously received sequence number stored in memory. If not, they generate a sequence number encryption key from the random number and stored key, encrypt the previously received sequence number using a block cipher encryption function, and send the encrypted sequence number in a response message to the network entity.