Adaptive Neuro-fuzzy Inference System Trained for Sizing Semi-elliptical Notches Scanned by Eddy Currents

Area of Science:

• Non-destructive testing
• Computational electromagnetics
• Artificial intelligence in engineering

Background:

• Eddy current testing (ECT) is crucial for detecting surface flaws.
• Modeling probe-material interactions is complex, especially with surface defects like electrical discharge machined (EDM) notches.
• Accurate characterization of EDM notches requires robust signal analysis.

Purpose of the Study:

• To evaluate COMSOL Multiphysics for modeling ECT probe interaction with EDM notches.
• To investigate the impact of probe lift-off and tilt on ECT signals.
• To develop and validate an adaptive neuro-fuzzy inference system (ANFIS) for estimating EDM notch length.

Main Methods:

• Finite element modeling (FEM) using COMSOL Multiphysics to simulate ECT probe and EDM notch interaction.
• Experimental validation of simulation results using probe impedance measurements.
• Design and training of an ANFIS using simulated ECT signal features.
• Testing the ANFIS with both simulated and measured signals from EDM notches.

Main Results:

• COMSOL Multiphysics effectively models the interaction between a split-D ECT probe and EDM notches.
• Probe lift-off and tilt significantly influence ECT signals, as confirmed by modeling and experiments.
• The trained ANFIS demonstrated a very low average estimation error for EDM notch length.
• High accuracy was achieved in estimating the length of tested EDM notches.

Conclusions:

• COMSOL Multiphysics is a capable tool for simulating ECT phenomena involving surface flaws.
• ANFIS provides a reliable and accurate method for determining EDM notch length from ECT signals.
• The integrated approach of FEM simulation and ANFIS offers a promising solution for non-destructive evaluation of surface defects.