Synthetic Data Classifier for Magnetic Media Defect Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for analyzing surface defect patterns on magnetic media are time-consuming and unreliable due to changing densities and overlapping patterns, and rely heavily on laborious and error-prone labeling of real data samples.
Innovation Solution
A method using synthetic data generated by parameterized rules to train a classifier model, combined with deep learning Convolutional Neural Networks (CNN) and manual feature algorithms, to identify and classify defect patterns on magnetic media, with a rule-based fusion model to resolve conflicts and improve processing speed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If real defect data samples are used for training classifier models, then the model can learn from actual manufacturing defects, but the labeling process becomes laborious and error-prone
Solution Approach 1:
The patent creates synthetic copies of defect patterns through parameterized rules instead of using real defect data. The system generates artificial defect patterns that replicate the characteristics of real defects without requiring actual defective media samples, thereby eliminating the time-consuming and error-prone manual labeling process while maintaining training effectiveness
Solution Approach 2:
The patent performs preliminary generation of defect pattern data before the actual classification task. By pre-generating synthetic defect data with known parameters and patterns, the system prepares training datasets in advance without needing to collect, capture, and label real defective media during the manufacturing process
2Reliability
If traditional defect analysis methods are used, then the process can handle simple defect patterns, but it becomes time-consuming and unreliable when dealing with changing densities and overlapping patterns
Solution Approach 1:
The patent changes the parameters of defect patterns through parameterized rules that can dynamically adjust defect characteristics such as density, size, shape, and spatial distribution. This allows the synthetic data generation to reflect varying manufacturing conditions and overlapping patterns, making the classifier model adaptable to different defect scenarios without retraining on new real data
Solution Approach 2:
The patent replaces traditional mechanical/manual defect analysis methods with an automated computer-implemented system. The classifier model, trained on synthetic data, automatically identifies and classifies defect patterns without requiring manual inspection or complex mechanical analysis systems, thereby improving both speed and reliability
3Adaptability or versatility
If multiple defect patterns are present on magnetic media, then comprehensive defect detection is achieved, but conflicting patterns make classification difficult and error-prone
Solution Approach 1:
The patent applies a hierarchy of classification actions where the system first attempts to identify the most prominent defect pattern, then handles conflicting patterns through a rule-based fusion model. This partial classification approach processes defects in stages, resolving conflicts by applying specific rules that prioritize certain pattern types over others based on their characteristics and manufacturing significance
Data Source
AI summary
A method includes generating synthetic data related to known defect patterns on surfaces of magnetic media using parameterized rules. A classifier model is trained with the synthetic data so that the classifier model learns how to detect and identify defect patterns on magnetic media. Performance of the classifier model is validated by using real defect pattern data. The classifier model is deployed for use in identifying defective data patterns on magnetic media test specimens. The classifier may be used before or after clustering defect data points on surfaces of magnetic media.