Investigation on SMT Product Defect Recognition Based on Multi-Source and Multi-Dimensional Data Reconstruction.

Area of Science:

• Manufacturing Engineering
• Quality Control
• Materials Science

Background:

• Defect recognition in solder paste printing is crucial for surface-mounted technology (SMT) production quality.
• Current machine inspection methods for SMT defects exhibit low accuracy, necessitating manual rechecks and increasing production costs.

Purpose of the Study:

• To develop an advanced SMT product defect recognition model using multi-source and multi-dimensional data reconstruction.
• To enhance the accuracy of defect identification in the solder paste printing process for improved SMT quality control.

Main Methods:

• Feature engineering techniques including interaction, selection, and conversion were employed to strengthen feature-defect correlations.
• A novel defect recognition model was constructed based on reconstructed features for the solder paste printing process.
• The model's performance was validated using a SMT production dataset and compared against existing methodologies.

Main Results:

• The proposed defect recognition model achieved a high accuracy rate of 96.97%.
• Comparative analysis demonstrated superior accuracy of the developed model over four other benchmark methods.
• The findings indicate a significant improvement in defect recognition capabilities for SMT production.

Conclusions:

• The developed multi-source, multi-dimensional data reconstruction approach offers a robust solution for SMT defect recognition.
• This new method provides a viable strategy for enhancing defect recognition rates and overall quality control in SMT manufacturing.
• The study presents a novel approach to address the limitations of current automated inspection systems in SMT production.