Coupling Analytical Models and Machine Learning Methods for Fast and Reliable Resolution of Effects in Multifrequency...

Area of Science:

• Materials Science
• Non-Destructive Testing
• Machine Learning

Background:

• Eddy current (EC) measurements are crucial for assessing conductive materials.
• Accurate EC analysis depends heavily on sample properties (conductivity, thickness) and sensor geometry (lift-off).
• Estimating these parameters simultaneously from EC signals is challenging, especially when unknown and variable.

Purpose of the Study:

• To develop a machine learning-based approach for simultaneous estimation of conductivity, thickness, and lift-off in eddy current testing.
• To provide an efficient alternative to time-consuming computational methods for EC parameter estimation.
• To create a model that does not require experimental data for training.

Main Methods:

• Utilized a machine learning framework to analyze eddy current signals.
• Developed a predictive model capable of inferring multiple sample and sensor parameters.
• Validated the approach using both simulated and real-world experimental eddy current data.

Main Results:

• The proposed machine learning approach successfully estimates conductivity, thickness, and lift-off.
• The method avoids complex, time-intensive calculations.
• The model's efficacy was confirmed with independent datasets, demonstrating its robustness.

Conclusions:

• Machine learning offers an efficient solution for complex eddy current parameter estimation.
• The developed approach reduces computational burden and eliminates the need for experimental training data.
• This method enhances the practical application of eddy current testing for material characterization.