Mold Inspection Learned Model Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing mold inspection devices struggle with accuracy, often misclassifying normal molds as defective or vice versa, due to difficulties in predefined pseudo defect feature recognition and actual defect misidentification.
Innovation Solution
The implementation of a machine learning-based inspection system that constructs a learned model through supervised learning using a dataset that includes inspection images, sand information, molding information, conveyance information, and environment information to improve the accuracy of mold surface defect detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If predefined features of pseudo defects are used for inspection, then the inspection process is simplified, but the inspection accuracy deteriorates due to inability to recognize all pseudo defect variations and misidentification of actual defects
Solution Approach 1:
The patent replaces the mechanical rule-based defect recognition system with a machine learning-based inspection system. The learned model automatically learns defect patterns from training data, substituting the manual predefined feature approach with an adaptive intelligent system that can recognize diverse defect types without explicit programming.
Solution Approach 2:
The patent changes the inspection approach from using fixed predefined parameters to dynamically learned parameters. The machine learning model adjusts its recognition criteria based on training data, allowing it to adapt to various defect patterns and conditions rather than relying on static predefined features.
2Measurement precision
If machine learning-based inspection is implemented, then the inspection accuracy is improved, but the device complexity increases due to model construction and data processing requirements
Solution Approach 1:
The patent applies preliminary action by constructing the machine learning model and preparing training data in advance. The model is trained offline with defect and non-defect images before deployment, so that during actual inspection, the system can directly use the pre-learned knowledge without performing complex training operations in real-time.
Solution Approach 2:
The patent uses copying by creating a learned model that replicates expert inspection knowledge. The machine learning model captures and reproduces defect recognition patterns from training data, effectively copying the inspection expertise into an automated system that can consistently apply learned patterns.
Data Source
AI summary
The purpose of the present invention is to improve accuracy in inspection of an appearance of a mold. The inspection device includes at least one processor for performing an inspection step of inspecting an appearance of a mold using a learned model constructed by machine learning. Input into the learned model includes an inspection image obtained by imaging the appearance of the mold. Output from the learned model is information indicating an inspection result of the appearance of the mold.