Multi-type feature extraction and classification of leakage in oil pipeline network using digital twin technology

Abstract

Digitization revolution plays predominant role in all industrial sector. Due to the more energy demand, oil pipeline extension is growing tremendously with the incorporation of IoT automation for overall monitoring and control of oil substation. Though IoT automates the entire pipeline transportation, still risk estimation and its prediction probability is a challenging task. Hence in the proposed work, digital twin concept is synchronized with high level feature extraction and classification architecture to identify the exact leakage spill in the oil pipeline. Though pipelines prevail to be feasible mode, the leak probability rate is getting increased and maintenance system becomes difficult with attention to the earlier prediction of leakage spills by undertaking entire pipeline. The feature extraction with sample classifier model is proposed based on the Ensemble Local Mean Decomposition (ELMD) supported with K–L method. The ELDM algorithm extracts the features with more higher-level product functions to formulate the high influenced clusters to predict the leakage spill. The time–frequency signals are analyzed through K–L divergence along with formulation of features to form Markov chain to classify the features with more accuracy with minimum datasets. ELMD was applied to disintegrate the spillage signals into a few significant PF parts with minimum dissimilarity. The experimental results are carried out to form the informational index from the test gathered by three pressure transmitters at 16 kg/cm² under 1.5-, 2.5-, and 3.5-mm spillage gaps. Further, the accomplishment of Digital Twin with high-speed data communication with minimum SNR ratio of holding optimal downlink power numerical results are presented. To compare the performance of the proposed system, performance measures like accuracy, recall, precision and delay time of the training and test samples of multiple pressure attributes are analyzed which conforms that the digital twin platform accompanied with ELDM feature extraction provides the accuracy of 91.23% as compared with SVM (Support Vector Machine) and NB (Naïve Bayes).