Reducing Sweeping Frequencies in Microwave NDT Employing Machine Learning Feature Selection

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Science

Background:

• Nondestructive Testing (NDT) is crucial for assessing material health and detecting flaws in structures.
• Current microwave NDT methods require analyzing a wide range of frequencies, increasing complexity and cost.
• A unified testing method for both metallic and dielectric materials, like microwave testing, is desirable.

Purpose of the Study:

• To reduce the number of sweeping frequencies used in microwave NDT for structural health assessment.
• To develop an intelligent approach for microwave NDT using machine learning feature selection and classification.
• To decrease the complexity and expenses associated with current diagnostic practices.

Main Methods:

• Machine learning feature selection techniques (Information Gain, Gain Ratio, Relief, Chi-Squared) were applied to identify influential frequencies.
• Frequency sweeping was treated as feature extraction for identifying key parameters.
• Classification models including Nearest Neighbor, Neural Networks, Random Forest, and Support Vector Machine were used to validate feature effectiveness.

Main Results:

• Feature selection algorithms successfully identified the most influential frequencies for microwave NDT.
• Experimental validation using a waveguide sensor and cracked metallic plates demonstrated the method's efficacy.
• Good crack classification accuracy rates were achieved after employing the proposed feature selection methods.

Conclusions:

• Machine learning-based feature selection can significantly reduce the operational frequency range in microwave NDT.
• The proposed method offers a more efficient and cost-effective approach to structural health monitoring.
• This intelligent NDT approach is applicable to both metallic and dielectric materials, enhancing diagnostic capabilities.