İha ve yapay zekâ tabanlı sensör füzyonu kullanarak petrol tesislerinin otomatik görsel denetimi için yeni bir hibrit yöntem

The facilities that deal with oil and gas are one of the essential components of the geography of every business and industrial zone in the world. This facet of geography is essential. With the continued growth in the various areas of commerce and industry, there is a growing demand for energy and, therefore, oil refineries. When there is a rupture in the oil transport network, the primary overriding concerns are maintaining public and ecosystem safety as well as protecting and preserving the integrity of the system. This thesis proposes a new pipeline rupture detection system that integrates three coordinated techniques: a DCAMFL algorithm, a Convolutional Neural Network, and a real-time, three-dimensional virtual drone. This multi-faceted approach is designed to find and trace the boundaries of minute fractures that would otherwise progress unnoticed. Pairs of DCAMFLs and CNNs, which are well-known for their superior visual acuity, are optimized to driven high performance. These complementary features are the more to equilibrium, and subsequently, the more to surpass their respective detection thresholds. Benchmarking and refinements to the model within ANSYS Fluent have ensured the approach is well-grounded, with the simulated outputs showing a high degree of correspondence to those of the physical model, leading to a reduction in certainty and confidence margin of the various performance parameters. The fluid dynamics inside the pipeline system were simulated in ANSYS Fluent. Using that information, we were able to determine the interactions of temperature, pressure, and flow speed in relation to the formation of and the ability to detect cracks. For an expanded analysis, we also looked at the ability of the algorithm to detect leaks due to potential bypassed cracks in the system. Next, we applied the technique to an extensive set of fractured oil pipelines, and the results were remarkable. For the crack detection, the F1-score was 95.6\%, the recall was 97.3\%, and the precision was 96.5\% respectively. These performance metrics provide evidence that the system accurately and reliably detected and classified the fractures. During the research work, we have shown that significant advancements in diesel pipelines crack detection can be achieved through the application of deep learning, fuzzy logic, and sophisticated modeling techniques. The approach described could yield various results, including improved pipeline monitoring, greater safety, and increased operational and structural longevity. To continue assisting developments within the algorithm, this study will try to determine the value and efficiency of the tool, as well as assessing and re-optimizing the architecture of the convolutional neural networks the range of data being processed will be also increased to a larger and complex pipeline.