Neural network eddy current non-destructive evaluation of conductive coatings thickness

Abstract

The proposed study is a machine learning application using a Neural Network for the prediction and identification of the thickness of aluminum placed over a steel plate. Two thousand and five hundred datasets with the eddy current method of different aluminum plate thicknesses above a steel plate and working frequencies of EC-sensor were generated using experimentally validated analytical models in our previous research. The data has three input parameters (normalized resistance, normalized reactance, and frequency) and one output (thickness). The ANN architecture involves careful consideration of the number of hidden layers and neurons within the model. The acquired data was split into two sections: the first section was used to train and test the selected model, and the second section was used to test the model on untrained data to demonstrate its high accuracy. The results obtained, as mentioned in the article, prove the validity and sensibility of the chosen model.