Real-time model-based condition monitoring of geothermal systems under uncertainties - Case study on electrical submersible pumps

Monitoring the condition of geothermal facilities and equipment (GFE) is crucial for ensuring reliable and costeffective operations. This work emphasizes the importance of real-time data-driven condition monitoring for proactive operation and maintenance (O&M) planning in geothermal assets. Recognizing that operational planning can be significantly impacted by uncertainties, a novel framework is proposed to monitor the performance of geothermal assets under these conditions. The approach combines machine learning (ML), statistical methods, and expert knowledge to account for uncertainty in evaluating the degradation or onset of failure in GFE. This method was applied to field data from a geothermal plant to monitor Electrical Submersible Pumps (ESPs) and tested for the accuracy and robustness of the framework. Additionally, the framework provides explainability, aiding in understanding the factors influencing equipment condition and degradation. The framework was capable of systematically detecting the onset of the ESP degradation up to six months prior to its failure, with an accuracy of more than 95% in estimating the performance of ESP during normal operation. The explainability layer provided insights on the cause of the failure which was not attributed to ESP malfunction but to a restriction in production inflow into the well. The framework's ability to accurately assess equipment condition under uncertainty supports more informed maintenance decisions, ultimately improving GFE operational reliability and efficiency.