基于联合训练的浮法玻璃缺陷预测性监控方法及系统
By using a joint training method, the frequency mismatch between high-frequency process data and low-frequency defect data in float glass production was solved, achieving high-precision defect prediction and realizing the transformation from post-production detection to pre-production early warning, thereby improving production efficiency and product quality rate.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- QINHUANGDAO DONGCHEN TECH CO LTD
- Filing Date
- 2025-11-13
- Publication Date
- 2026-07-17
AI Technical Summary
In existing technologies, float glass defect detection mainly relies on post-processing inspection, which leads to lag issues. Furthermore, due to the frequency mismatch between high-frequency process data and low-frequency defect data, it is difficult to establish an accurate mapping relationship, resulting in poor defect prediction accuracy.
A joint training-based approach is adopted, which generates high-quality pseudo-labels by iteratively training the initial time series prediction model and the defect soft measurement model. Combined with multi-task learning and adaptive weighted loss, the model parameters are optimized to achieve effective fusion of high-frequency process data and low-frequency defect data.
It has enabled a shift from post-inspection to pre-inspection, predicting defects tens of minutes to an hour in advance, improving prediction accuracy and interpretability, and achieving a leap from passive quality control to proactive quality assurance.
Smart Images
Figure CN121328341B_ABST