Suitability Analysis of Machine Learning Algorithms for Crack Growth Prediction Based on Dynamic Response Data.

Area of Science:

• Materials Science
• Computational Materials Engineering
• Machine Learning Applications

Background:

• Machine learning (ML) offers superior damage detection and prediction capabilities in materials science compared to traditional methods.
• ML models provide reliable representations for applications like material design, property prediction, and defect classification.
• Interpretability of ML models can be a challenge in materials science, hindering understanding of prediction reasoning.

Purpose of the Study:

• To analyze the compatibility of material dynamic response data with prominent machine learning approaches.
• To guide researchers in selecting effective and understandable ML models for materials engineering.
• To enhance the comprehension of ML model predictions in the context of material damage.

Main Methods:

• Analysis of requirements and characteristics of common ML algorithms for crack propagation.
• Selection and application of K nearest neighbor, Ridge, and Lasso regression algorithms.
• Evaluation of dynamic response data from aluminum and ABS materials using experimental data for model training.

Main Results:

• Natural frequency identified as the most significant predictor for ABS material damage.
• Temperature, natural frequency, and amplitude identified as key predictors for aluminum material damage.
• Crack location along samples showed no significant impact on damage prediction for either material.

Conclusions:

• The study demonstrates the effectiveness of selected ML algorithms in predicting material damage based on dynamic response.
• Identified key predictors (natural frequency, temperature, amplitude) offer insights into material behavior under dynamic loading.
• Future research should explore these ML techniques across a broader range of materials and dynamic conditions.