Automated error logging in the flowmeter design process: Approaches to processing and analysis

In the modern design of variable differential pressure flowmeters, the introduction of reliable automated logging systems is relevant, as conventional logging methods do not provide the required accuracy and stability under load. The purpose of this study was to substantiate and develop methodological approaches to automating logging processes in the design of variable differential pressure flowmeters, considering parametric optimisation, reducing error localisation time, and increasing the accuracy of uncertainty estimation. The study was based on experimental measurements in the SolidWorks 2024 and ANSYS Fluent software environments using the Elasticsearch and Kibana tools, as well as further computational processing in MATLAB 2024a. The evaluation covered the metrics of accuracy, completeness, integrated harmonic mean, area under the performance curve, time to detect a critical event, time to notify an engineer, time to localise an error, average error in flow calculation with bootstrap analysis, and an integrated logging efficiency index. The study found that basic logging provides limited accuracy (≈ 71%) and low stability (≈ 82.5% of failure-free sessions), while heuristic methods increase efficiency to 87.9%, but leave a considerable level of event duplication and lose stability under load. The statistical classification showed better results (integrated F1-score = 0.81, average consumption error = 2.5%, integrated logging efficiency index = 0.78), providing a balance between accuracy and performance. The highest indicators were achieved with the machine learning approach: accuracy exceeded 91%, completeness was over 87%, the average cost calculation error was reduced to 1.7%, the recovery of cause-and-effect relationships reached over 86%, and the integrated logging efficiency index was 0.89. Analysis of variance and the non-parametric Kruskal-Wallis test confirmed the reliability of the differences between the approaches. The practical significance of this study lies in the identification of machine learning algorithms as a basic direction for the development of intelligent logging systems, the findings of which can be used by engineering companies, software developers, and enterprises in the oil and gas, energy, and mechanical engineering industries to improve the reliability, scalability, and adaptability of design systems to real-world operating conditions