Crack Localization of Pipelines Using Machine Learning and Fuzzy Digital Twin

Pipelines are used to transport liquids and gases between different places. In recent years, identifying the location and size of cracks in transmission pipes has received much attention. Thus, in this research, the combination of fuzzy digital twin (FDT), support vector machine (SVM), and backstepping (BS) observer is suggested for leak detection and localization. In the first stage, the proposed fuzzy digital twin is introduced. So, first, the flow rate in the healthy condition is modeled using the fuzzy Gaussian process Regression (FGR) which is a combination of Gaussian process Regression (GR) and fuzzy algorithm. After that, in the second stage, the flowrate signals are estimated using the fuzzy digital twin that is a combination of FGR, BS technique, and fuzzy procedure. The proposed FDT is used to modeling and estimated the flow rate in the pipeline in normal and abnormal conditions. Moreover, it can be used for data generation. After modeling and estimated data using FDT, in the next step, the residual data is generated to improve the accuracy of crack localization. Next, the crack is detected by the combination of the SVM and FDT. To find the location of the crack, the combination of FDT and the backstepping algorithm is suggested. The Ulsan Industrial Artificial Intelligence (VIAl) Lab dataset is used to test the effectiveness of the proposed scheme. The results show that the proposed FDT is 25% improved in the accuracy for fault localization compared with the classical digital twin.