Non-Destructive Detection of Pipe Line Cracks Using Ultra Wide Band Antenna with Machine Learning Algorithm

Abstract

In this article, an Ultra-Wide Band (UWB) antenna for the pipeline crack detection process is proposed. A UWB antenna has been designed with the dimension of 32 x 32 mm2 and it resonates from 3 GHz to 10.8 GHz. The designed antenna produces a peak gain of 4.36 dB. A pair of UWB antennas are employed in various pipeline scenarios and the received pulse from antenna 1 to antenna 2 is used for further processing and detection of pipeline cracks. Through the suitable machine learning data classifier algorithm the dimension of the crack has been detected. The various features such as mean, standard deviation (σ), mean average deviation (mad), skewness, and kurtosis have been extracted from the received pulse. Then the three different machine learning algorithms namely Support Vector Machine (SVM), k-Nearest Neighbor (kNN), and Naïve Bayse (NB) were trained and tested using extracted features, and the dimension of the void has been identified. Out of these three machine learning algorithms, kNN provides better accuracy and precision. It predicts the small cracks with 100% accuracy having a dimension as small as 1 mm width.