Authors:

Hamdan Ashfaq

Date Created:

Not specified

Course Title:

Professor:

Not specified

About Paper:

Sn-58Bi solder alloys are critical components in electronic packaging applications. Microstructural patterns directly influence mechanical properties and joint reliability. Manual analysis of these microstructures is subjective and time-intensive. The different types of patterns that form under varying thermal conditions affect material behavior. Accurate classification is essential for optimizing processing parameters. This research develops a machine learning framework to automatically categorize microstructural patterns in Sn-58Bi solder based on scanning electron microscopy images. Sn-58Bi samples were processed under four distinct cooling conditions: water quenching, air cooling at 34°C/min, controlled cooling at 6°C/min, and slow cooling at 1°C/min. SEM images were acquired and cropped to create a training dataset. Each image patch was manually labeled according to dominant pattern type: fishbone, lamellar, fine, or coarse structures. A ResNet-50 convolutional neural network architecture was implemented with transfer learning to classify these patterns. Data preprocessing included image augmentation to enhance model robustness. Preliminary results with 310 training images demonstrate 82% validation accuracy in distinguishing between the four microstructure classes. This demonstrates competitive performance using 75% fewer training images than comparable studies. Analysis confirms that faster cooling rates produce finer, more irregular structures, while slower cooling promotes larger, organized patterns. The automated classification system identifies subtle differences that correlate with cooling history. This approach provides an objective, quantitative tool for microstructure analysis in electronic packaging applications. Future work will expand to multiple datasets of over 1000 images while aiming for higher accuracy and integrating mechanical property predictions, enabling potential manufacturing applications. Keywords: Deep Learning; Microstructure Classification; Solder Alloys; Image Processing; Materials Characterization

Source:

Purdue University / 2025

Topics:

No topics listed

Co-authors:

Hamdan Ashfaq