A Small Database with an Adaptive Data Selection Method for Solder Joint Fatigue Life Prediction in Advanced...

Area of Science:

• Materials Science
• Mechanical Engineering
• Computer Science

Background:

• Predicting solder joint fatigue life in advanced packaging is crucial for reliability.
• Machine learning (ML) offers efficient prediction but requires substantial training data.
• Large datasets increase computational costs, posing a challenge for ML model development.

Purpose of the Study:

• To develop accurate and efficient ML models for predicting solder joint fatigue life.
• To investigate the effectiveness of adaptive sampling methods for ML model training with limited data.
• To explore ensemble learning for further performance enhancement of ML models.

Main Methods:

• Utilized machine learning to create metamodels for approximating system attributes.
• Applied adaptive sampling techniques to train ML models using a small subset of existing data.
• Visualized model performance using predefined criteria and explored ensemble learning strategies.

Main Results:

• Adaptive sampling enables the development of effective ML models with reduced datasets.
• The study demonstrates a viable approach to improve prediction accuracy while managing computational expenses.
• Ensemble learning shows potential for boosting the performance of trained AI models.

Conclusions:

• Adaptive sampling is an efficient strategy for building accurate ML models for solder joint fatigue life prediction.
• This research offers a cost-effective method for enhancing the reliability of advanced packaging.
• Further improvements in AI model performance can be achieved through ensemble techniques.