Battery Electrode Inspection Using Adaptive Pre-Trained Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional battery electrode inspection devices require lengthy initial application times and continuous redesign when field conditions change, leading to decreased precision and difficulty in determining defect classification accuracy.
Innovation Solution
An apparatus and method using pre-trained learning models, including random forest and convolutional neural networks, to inspect battery electrodes, selecting models based on training data amount and re-training with updated data for improved reliability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional program logic is designed based on defect images obtained from a camera installed in the field during initial application, then the inspection device can detect defects, but it requires a long time for initial application and continuous redesign when field conditions change
Solution Approach 1:
The patent applies preliminary action by pre-training multiple learning models with different amounts of training data before actual inspection. The system prepares model candidates in advance with varying data quantities, so when deployment occurs, the appropriate pre-trained model can be selected immediately without requiring lengthy initial application or field-specific program design.
2Adaptability or versatility
If conventional program logic is designed based on defect images obtained from a camera installed in the field, then the inspection device can detect defects, but the program logic needs to be continuously redesigned by an engineer when field situation or environment changes
Solution Approach 1:
The patent implements self-service by enabling the inspection system to automatically adapt to changing field conditions through automated model selection based on available training data quantity. Instead of requiring engineers to continuously redesign program logic, the system autonomously selects the most appropriate pre-trained model according to the amount of training data available, eliminating the need for manual intervention when conditions change.
3Measurement precision
If a pre-trained deep learning model is used with sufficient training data, then inspection precision is improved, but the system requires more training data resources
Solution Approach 1:
The patent applies parameter changes by varying the amount of training data used to create different model candidates. The system creates multiple pre-trained models with different data quantities (e.g., small, medium, large datasets) and selects the appropriate model based on the available training data at deployment time. This allows the system to achieve high precision when sufficient data is available while also functioning adequately with limited data resources.
Data Source
AI summary
An apparatus for inspecting battery electrodes according to embodiments of the present invention may extract an inspection object image including an area suspected of being defective based on an electrode image obtained from a camera, determine and apply different types of learning models for determining whether an electrode corresponding to the inspection object image is defective according to the amount of learning data.