Intelligent Flaw Detection in Eddy Current Inspection Data through Machine Learning Model

Eddy current (EC) testing is the most widely used method for inspecting heat exchanger tubes in industries such as petrochemicals, refineries, and nuclear power plants. Heat exchangers typically consist of hundreds to thousands of tubes, and the data from EC inspections is analysed manually. This manual process is error-prone due to operator fatigue and leads to increased downtime. Thus, there is a need for an intelligent, automated flaw detection system. Although machine learning (ML) methods for this problem exist, they are often either computationally expensive or less accurate. The paper presents a robust machine learning model for automated classification of flaw signals from eddy current inspection data of heat exchanger tubes. The proposed model employs four sliding window based ingenious features namely variance, template correlation, template dynamic time warping distance and area under the signal with Random Forest supervised machine learning model, to identify flaws. The efficacy of the model is evaluated on tube inspection data acquired in a heat exchanger by comparing its performance against expert analysis. The machine learning model exhibits an impressive accuracy of 99.94% for classification of flaw signals in addition to higher desirable metrics such as precision, recall and F1-score. This work lays a strong foundation for developing a real-time, robust and reliable flaw detection system.