Automatic defect detection for TFT-LCD array process using quasiconformal kernel support vector data description.

Area of Science:

• Materials Science
• Computer Science
• Electrical Engineering

Background:

• Defect detection is crucial for improving yield rates in thin-film transistor liquid crystal display (TFT-LCD) manufacturing.
• The array process is a key stage where defects can significantly damage LCD panels.
• Traditional Support Vector Data Description (SVDD) methods have limitations in generalization performance for LCD defect detection.

Purpose of the Study:

• To propose a novel one-class machine learning method for robust defect detection in the TFT-LCD array process.
• To address the generalization performance limitations of existing SVDD methods.
• To enhance the accuracy and efficiency of defect detection in LCD manufacturing.

Main Methods:

• Development of a quasiconformal kernel Support Vector Data Description (QK-SVDD) method.
• Integration of quasiconformal transformation into a predefined kernel to improve SVDD generalization.
• Application and testing of the QK-SVDD method on real LCD images from a Taiwanese manufacturer.

Main Results:

• The proposed QK-SVDD achieved a high defect detection rate of 96%.
• QK-SVDD demonstrated a significant improvement in generalization performance over traditional SVDD, exceeding 30%.
• The QK-SVDD defect detection system processed LCD images in under 60 milliseconds.

Conclusions:

• QK-SVDD offers a robust and effective solution for defect detection in TFT-LCD array manufacturing.
• The method significantly enhances generalization capabilities compared to conventional SVDD.
• QK-SVDD provides a fast and accurate defect detection system suitable for industrial application.