ISSN 2658–5782
DOI 10.21662
Electronic Scientific Journal


© Институт механики
им. Р.Р. Мавлютова
УФИЦ РАН

Яндекс.Метрика web site traffic statistics

Salikhov L.S. Modification of algorithms for neural network-based shape recognition of ESP components in the automation of inspection operations. Multiphase Systems. 19 (2024) 2. 86–91 (in Russian).
2024. Vol. 19. Issue 2, Pp. 86–91
URL: http://mfs.uimech.org/mfs2024.2.013,en
DOI: 10.21662/mfs2024.2.013
Modification of algorithms for neural network-based shape recognition of ESP components in the automation of inspection operations
L.S. Salikhov
Ufa State Petroleum Technological University, Ufa

Abstract

In recent decades, the primary oil extraction technology in Russia has been the mechanized method, implemented by installing Electric Submersible Pump (ESP) units in wells. Due to the remote locations of oil fields and the economic impracticality of transporting units that have exhausted their service life, downtime and accumulation of ESPs occur directly near the fields. Given the possibility of repairing and restoring the functionality of failed pumps, the development of a mobile robotic module for sorting, defect detection and storage of ESP components is proposed. This module would be part of a mobile robotic complex for conducting repairs of the units. This article examines modern methods of shape inspection of components using neural networks and machine learning, which also improve defect recognition accuracy. Based on the analysis of various control methods, their disadvantages have been identified, which may manifest themselves when they are used in the developed robotic module at the preprocessing stage. To eliminate the identified shortcomings, it is proposed to use additional training of the neural network model directly during its operation. Based on the results of the analysis and conducted computer experiments, the optimal way to solve the problem was determined in the form of methods and algorithms for modifying the neural network method for recognizing the shape of ESP parts, which made it possible to increase the accuracy of detecting the most common defects.

Keywords

neural network model,
machine learning,
object recognition,
ESP repair,
shape inspection,
defect detection

Article outline

Most of the oil is extracted through the operation of an electric centrifugal pump (ECP) installation. The use of such installations makes it possible to extract about 8 trillion rubles worth of products from wells annually. At the same time, the main active fields are located in remote areas (Siberia, the Arctic), which significantly complicates preventive and repair work in order to prevent downtime of oil production due to premature equipment failure. Currently, the repair of failed ESPs is carried out at enterprises in the central and European parts of the country, therefore, the equipment that has been put out of operation is stored at the field. Therefore, based on the above and guided by the plans for the digital transformation of the fuel and energy complex, it was decided to develop a mobile robotic complex for carrying out certain types of repair work directly at the field. The problem is complex, so the complex is built in the form of three modules.

The article shows the solution of one of the tasks of the preprocessing module of the ESP parts, after which they are sent to the surface and dimensional parameters defect. Automation of preprocessing, where obvious defects in the shape of products are revealed, is not a trivial scientific and technical task, therefore, a fairly thorough analysis of known solutions and techniques was required. The results of a comparative analysis of existing methods of washing and drying operations of objects, a review of patent literature and existing equipment for getting rid of pollution allowed us to correctly select the necessary equipment and synthesize the optimal architecture of the site. Then, the analysis of pollutants and methods of their purification was carried out, since each variant of the operating conditions (deposits) leads to a certain type of deposits. The data given in the tables allow you to choose the optimal cleaning agent and process parameters. As a possible technical implementation of the shape control of parts, the option of using a neural network algorithm for recognizing obvious shape defects based on the ResNet18 architecture with the introduction of partial modifications in order to increase the accuracy and speed of defect recognition was considered.

The purpose of this work is to generate a training sample based on three-dimensional CAD models of the objects under study, which will identify and justify the modifications necessary to increase the accuracy of defect recognition. In order to achieve this goal, the following solution technology is proposed:

  1. Based on drawings, defect maps and technological documentation, create ideal CAD models of the objects under study (Solidworks), models with various types of defects.
  2. Create a sample of several hundred images for further training of the model.
  3. Train a ResNet type model to classify and control shape defects, checking the accuracy of the trained model.
  4. Modify the model on the pre-trained ImageNet model and compare the results obtained on the model without modification.
  5. To evaluate the possibility of using a pre-trained model to control real details and the possibility of its further training during operation, with an increase in the data set due to the expansion of the sample due to objects that have undergone preliminary shape defection in the robotic module.

Based on the parts selected for testing, the working wheel and the guide device, three-dimensional CAD models of objects were developed, a training sample was formed, a neural network model was trained and object recognition was verified by the model without taking into account modifications and taking into account modifications, respectively, estimated results of the accuracy of defect detection were obtained. Experimental verification has shown the effectiveness of the proposed approach, which ensures the required recognition accuracy even on a model dataset. A way to increase this parameter is shown by using the results of the real application of the neural network technique, i.e., supplementing the dataset with real images obtained during the operation of the robotic complex. Based on the results of testing, it can be concluded that modifications of the chosen method can increase the accuracy of the neural network model by about 16%, i.e. the values that meet the requirements. The use of a neural network at the preliminary stage of product control, high-quality rejection of parts made it possible to reduce the flow and load of other modules, allowing us to effectively filter out obvious defects already in the input stream.

References

  1. Исполнение федерального бюджета и бюджетов системы Российской Федерации за 2022 год. М.: Минфин России. 2023.
    [Execution of the federal budget and budgets of the Russian Federation system for 2022] Ispolnenie federal’nogo byudzheta i byudzhetov sistemy Rossijskoj Federacii za 2022 god. [Moscow: Ministry of Finance of Russia] Minfin Rossii. 2023.
    https://minfin.gov.ru/common/upload/library/2023/08/main/Illustrirovannoe_izdanie_za_2022_god.pdf
  2. Рудницкий С.В., Зацепин А.Ю., Демин Е.В., Ющенко Т.С. Перспективное скважинное оборудование для добычи сланцевой нефти баженовской свиты // PROНефть. Профессионально о нефти. 2021. Т. 6, № 1. C. 64–75.
    Rudnitsky S.V., Zatsepin A.YU., Demin E.V., Yushchenko T.S. Prospective artificial lift equipment for shale oil production from the Bazhenov formation. PRONeft. Professionally about oil. 2021. Vol. 6, No. 1. Pp. 64–75 (in Russian). DOI: 10.51890/2587-7399-2021-6-1-64-75
  3. Конторович А.Э., Эдер Л.В., Филимонова И.В., Мишенин М.В. Роль уникальных и крупных месторождений в нефтяной промышленности России: ретроспектива, современное состояние, прогноз // Энергетическая политика. 2016. № 2. С. 34–43.
    Kontorovich A.E., Eder L.V., Filimonova I.V., Mishenin M.V. The role of unique and large deposits in russian oil sector: retrospective analysis, current state and forecast. Energy policy. 2016. No. 2. Pp. 34–43 (in Russian).
    EDN: WXMMYV
  4. Мотина С.И., Копейкин И.С. Техническая экспертиза промышленной безопасности эксплуатационных колонн нефтяных скважин. // Материалы Международной научно-технической конференции «Современные проблемы нефтегазового оборудования». 2019. C. 436–440.
    Motina S.I., Kopeikin I.S. Technical expertise of industrial safety of production columns of oil wells. Materials of the International Scientific and Technical Conference «Modern problems of oil and gas equipment». 2019. Pp. 436–4408 (in Russian).
    EDN: QCPZZA
  5. Галиев К.Р., Ямалиев В.У. Методы повышения безотказности УЭЦН в осложнённых условиях эксплуатации // Материалы Международной научно– технической конференции «Современные проблемы нефтегазового оборудования». 2019. C. 209–218.
    Galiev K.R., Yamaliev V.U. Methods for improving the operation reliability of electric centrifugal pump installation in complicated operating conditions. Materials of the International Scientific and Technical Conference «Modern problems of oil and gas equipment». 2019. Pp. 209–218 (in Russian).
    EDN: SMTDPL
  6. Гиндуллин И.И, Лонщаков А.А., Салихов Л.С. Структура робототехнического модуля сортировки деталей // Завалишинские чтения: XVII Международная конференция по электромеханике и робототехнике (СПб., 18–19 апреля 2023 г.). Молодёжная секция: сб. докл.: в 2 ч. Ч. 1. СПб.: ГУАП, 2023. С. 41–44.
    Gindullin I.I., Lonshakov A.A., Salikhov L.S. [The structure of the robotic module for sorting parts] Struktura robototekhnicheskogo modulya sortirovki detalej. Struktura robototekhnicheskogo modulya sortirovki detaley. Zavalishinsky readings: XVII International Conference on Electromechanics and Robotics (St. Petersburg, April 18–19, 2023). Youth section. Vol. 2, No. 1. St. Petersburg: GUAP. 2023. Pp. 41–44 (in Russian).[7] Zhu H., Zhu J., Zhou Z., Rutter R., Forsberg M., Gunter S., Zhang H.-Q. Experimental Study of Sand Erosion in Multistage Electrical Submersible Pump ESP: Performance Degradation, Wear and Vibration // International Petroleum Technology Conference: Day 1 Tue, March 26, 2019. Paper Number: IPTC-19264-MS.
    DOI: 10.2523/iptc-19264-ms
  7. Волошин А.И., Гусаков В.Н., Фахреева А.В., Докичев В.А. Ингибиторы для предотвращения солеотложения в нефтедобыче // Нефтяное хозяйство. 2018. № 11. С. 60–72.
    Voloshin A.I., Gusakov V.N., Fahreeva A.V., Dokichev V.A. Scaling prevention inhibitors in oil production. Neftepromyslovoe delo. 2018. No. 11. Pp. 60– 72 (in Russian).
    DOI: 10.30713/0207-2351-2018-11-60-72
  8. Bulgarelli N., Biazussi J., Verde W., Perles C., de Castro M., Bannwar A. Experimental investigation of the Electrical Submersible Pump’s energy consumption under unstable and stable oil/water emulsions: A catastrophic phase inversion analysis // Journal of Petroleum Science and Engineering. 2022. Vol. 216. Paper 110814.
    DOI: 10.1016/j.petrol.2022.110814
  9. Seif A., Salut M.M., Marsono M.N. A hardware architecture of Prewitt edge detection // 2010 IEEE Conference on Sustainable Utilization and Development in Engineering and Technology. 2010. Pp. 99–101.
    DOI: 10.1109/STUDENT.2010.5686999
  10. Huang R., Gu J., Sun X., Hou Y., Uddin S. A rapid recognition method for electronic components based on the improved YOLO-V3 network // Electronics. 2019. Vol. 8, no. 8. Paper 825.
    DOI: 10.3390/electronics8080825
  11. Tian Y., Yang G., Wang Z., Wang H., Li E., Liang Z. Apple detection during different growth stages in orchards using the improved YOLO-V3 model // Computers and electronics in agriculture. 157. 2019. Vol. 157. Pp. 417–426.
    DOI: 10.1016/j.compag.2019.01.012
  12. He K., Zhang X., Ren S., Sun J. Deep Residual Learning for Image Recognition // 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016. Pp. 770–778.
    DOI: 10.1109/cvpr.2016.90
  13. Liu Z., Mao H., Wu C.-Y., Feichtenhofer C., Darrell T., Xie S. A ConvNet for the 2020s // Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2022. Pp. 11976–11986.
    DOI: 10.48550/arXiv.2201.03545