Defect Cause Localization Using Reduced-Dimension Combination Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The complexity of identifying the cause of defective products in industrial production lines is increased by the large number of process devices and vast amount of generated data, leading to prolonged time consumption in fault localization.
Innovation Solution
A data processing method involving dimension reduction algorithms like PCA, LDA, and LLE to extract combination features from product samples, followed by influence score calculation to rank and identify the cause of defects based on these features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the number of process devices and data amount increase to improve production monitoring capability, then the ability to detect product defects is improved, but the complexity of locating defect causes increases and time consumption increases
Solution Approach 1:
The patent extracts and selects only the key process devices that have the greatest influence on product defects from the large number of process devices in the production line. By using influence degree calculation and threshold filtering, the system extracts the critical subset of devices that contribute most to defect occurrence, thereby reducing the complexity of defect cause location while maintaining reliable defect detection capability.
2Reliability
If the number of process devices and data amount increase to improve production monitoring capability, then the ability to detect product defects is improved, but the time consumption to locate faulty devices increases
Solution Approach 1:
The patent extracts and selects only the key process devices that have the greatest influence on product defects from the large number of process devices in the production line. By using influence degree calculation and threshold filtering, the system extracts the critical subset of devices that contribute most to defect occurrence, thereby reducing the complexity of defect cause location while maintaining reliable defect detection capability.
Solution Approach 2:
The patent performs preliminary calculation of influence degrees for all process devices before actual defect occurrence. By pre-calculating and ranking the influence of each device on potential defects, the system prepares the data in advance so that when a defect occurs, the location of the faulty device can be quickly determined by simply checking the pre-ranked list, significantly reducing the time required for fault localization.
3Productivity
If dimension reduction is applied to reduce data complexity, then the speed of defect cause localization is improved, but the amount of information retained may be reduced
Solution Approach 1:
The patent transforms the original high-dimensional process device data into a lower-dimensional representation by calculating influence degrees as new parameters. This parameter transformation condenses multiple device attributes into a single influential metric that ranks devices by their contribution to defects, reducing data dimensionality while preserving the critical information needed for fast defect cause localization.
Data Source
AI summary
The present disclosure provides data processing methods and apparatuses, an electronic device and a storage medium. The method includes: obtaining a product sample set; obtaining combination features in specified dimensions of the product sample set by processing a second parameter based on a preset dimension reduction algorithm; obtaining influence scores respectively for the combination features in specified dimensions based on a first parameter and the combination features in specified dimensions; obtaining at least one combination feature ranked top by sorting the combination features based on the influence scores, and taking a raw parameter corresponding to the at least one combination feature as a cause of the product defect. In the embodiments of the present disclosure, combination features in R dimensions may be a combination of raw parameters having similarity such that similar parameters are associated while raw information of the product samples is retained, thus helping fast locating the cause of the product defect, and improving the detection efficiency.