Fault detection in pipelines with graph convolutional networks (GCN) method

Pipeline networks have a wide range of applications, from the transportation of energy sources such as oil and natural gas to the conveyance and distribution of water resources. However, leaks and ruptures in pipelines can cause significant harm to the environment. Therefore, it is crucial to accurately detect pipeline faults in order to avoid economic losses and protect the environment. In this study, pipeline networks carrying water fluid are represented using graph structures. The graph convolutional network (GCN)algorithm is employed for the detection of leaks and blockages in pipeline networks. Experimental methodsare employed to collect the necessary data (pressure data) for the GCN algorithm, creating two datasets byconsidering five different scenarios. The fault detection performance of the GCN algorithm is compared withother graph machine learning algorithms, namely, RGCN, HinSAGE, and GraphSAGE. The results of thisstudy indicate that the performance of the GCN model surpasses that of the other algorithms. Reviewing the literature, accuracy rates for fault diagnosis in pipeline networks using machine learning algorithms rangefrom 78.51% to 99%. In this study, it is found that the GCN, GraphSAGE, HinSAGE, and RGCN algorithms achieve fault detection accuracies of 91%, 90%, 87%, and 89%, respectively, in pipeline networks. Classical machine learning SVM model was used to compare the performance of graph-based algorithms. It is seen that the performances of the algorithms face the literature and the results are above the literature average.