Study on Machine Learning Based Automatic Defect Detection in Non Stationary Thermal Wave Imaging

Machine learning is a branch of Artificial intelligence is used to evaluate the solution to a problem with
efficiently and accurately. Detection of subsurface non uniformity is crucial in deciding the strength of
objects for various industrial applications. Non stationary thermal wave imaging is emerging as a
reliable qualitative assessment procedure to detect anomalies in a wide range of materials. This paper
proposes a supervised machine learning based classification modality to detect the subsurface
defects using quadratic frequency modulated thermal wave imaging and experimentation has been
carried over glass fibre reinforced polymer material (GFRP) with 10 Teflon patches having different
depths and sizes and Carbon fibre reinforced polymer (CFRP) with 25 bottom holes having different
sizes and depths. In this paper three well known supervised machine learning techniques Decision
tree (DT), Support vector machine (SVM) and k-nearest neighbour (KNN) classifiers are used for
defect detection. Detection capability and reliability of defect detection have been assessed using
signal to noise ratio and probability of detection respectively. Among various supervised learning
methods, decision tree classifier provides better detection capability, sizing and reliability estimation
compared to remaining processing methods.