An adaptive neuro-fuzzy inference system-based approach for oil and gas pipeline defect depth estimation

To determine the severity of metal-loss defects in oil and gas pipelines, the depth of potential defects, along with their length, needs first to be estimated. For this purpose, pipeline engineers use intelligent Magnetic Flux Leakage (MFL) sensors that scan the metal pipelines and collect defect-related data. However, due to the huge amount of the collected MFL data, the defect depth estimation task is cumbersome, time-consuming, and error-prone. In this paper, we propose an adaptive neuro-fuzzy inference system (ANFIS)-based approach to estimate defect depths from MFL signals. Depth-related features are first extracted from the MFL signals and then are used to train the neural network to tune the parameters of the membership functions of the fuzzy inference system. A hybrid learning algorithm that combines least-squares and back propagation gradient descent method is adopted. Moreover, to achieve an optimal performance by the proposed approach, highly-discriminant features are selected from the obtained features by using the weight-based support vector machine (SVM). Experimental work has shown that encouraging results are obtained. Within error-tolerance ranges of ±15%, ±20%, ±25%, and ±30%, the depth estimation accuracies obtained by the proposed technique are 80.39%, 87.75%, 91.18%, and 95.59%, respectively. Moreover, further improvement can be easily achieved by incorporating new and more discriminant features.