Remote Real-Time Gas Data Monitoring with Predictive Maintenance of Turbine Flow Meters

Abstract In modern gas distribution practices, the ability to remotely access operational data and monitor plant conditions in real-time is becoming paramount in assessing operational intricacies. In this study, a robust system that provides real-time access to turbine meter operation data and predicts the time until failure, enhancing maintenance schedules and reducing unexpected downtimes is presented. To achieve this, a comprehensive methodology was employed, starting with the design and development of a real-time data collection system using a Raspberry Pi 4 and various communication components to monitor gas flow parameters being emitted from a simulation software environment. Data collected from the system was then used to train predictive models. Regression models, including Random Forest Regressor, Gradient Boosting Regressor, and Support Vector Regressor, were compared with time-series models like SARIMA. The models were evaluated based on metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R2. Additionally, anomaly detection models like Isolation Forest were implemented to identify deviations in gas flow data. The results demonstrated that the Random Forest Regressor provided the best result when compared with other regression models used. However, the time-series model, SARIMA slightly outperformed regression techniques, showing a 2.14% reduction in RMSE and a 2.02% increase in R2 at the cost of higher computational demands. The real-time monitoring device achieved high reliability with a latency of 5 seconds and a system reliability of over 95%. These findings underscore the effectiveness of both regression and time-series models in predicting the time until failure of turbine meters. The study provides a framework for predictive maintenance in the gas industry that can minimize operational disruptions and optimize general maintenance practices.