Abstract
Water injection is widely utilized technique for secondary Oil recovery. Aquifer water is produced from Water supply wells in aquifer zones and pumped out through Electrical Submersible Pumps (ESP). In ADNOC Onshore Water supply ESP are operated on run to fail mode. Average Mean time between failure method was leading to ineffective work over planning and rigs utilization. This paper illustrates the condition-based failure prediction model established in ADNOC Onshore matured field to predict ESP failure based on Realtime monitoring.
Due to criticality of the water injection requirement for sustainable oil production, ESP down time and Meantime to recover reduction was highly envisaged. Artificial intelligence-based Machine learning Solution was utilized for predicting probability of failures based on real time data monitoring, anomaly detection and historical failures. Real time data of past 6 years, actual down time events and all ESP failure root cause analysis were collected, assessed, and detailed data base was developed. The solution provides a probability of ESP failures based on the anomalies (anomaly severity) detected from unsupervised machine learning model (individual cluster based), ESP performance MTBF & number of starts. The probability is normalized based on weight-based approach.
Unavailability of any predictive model was leading to ineffective work over planning and impacting rigs availability, which results in elongated down times and significant shortfalls of reservoir pressure support.
Blind tests have been conducted historical failures and 70% of prediction accuracy has been achieved. The deep dive analysis has resulted that unavailability critical real time parameters such as motor temperatures, intake pressure & vibration data has impacted the prediction accuracy. Models have been deployed real time mode and prediction probability was provided to engineers on daily basis. The high probability indicates that failure incident can happen in 3 months prior to actual failure events. After successful deployment of this pilot, full scale ESP failure prediction analysis is being implemented on all remaining facilities in AON asset. Following benefits shall be materialized after full scale implementation.
Build the automated predictive solution and alert custodian.
Enabling better planning and workover candidate selection prioritization and scheduling.
Expedite meeting increased Oil production mandates.
Sustain oil production and recovery ratios.
MTTR of ESP shall be reduced by 73%.
Facilities availability shall be improved up to 96%.
Effective utilization of Rigs.
Increase run life of ESP
Develop a support and advisory solution
Despite the unavailability of real time vital parameters, first of its kind, Predictive model based on Artificial intelligence and Machine Learning was developed and deployed which helped in building an automated predictive solution to alert asset custodian prior to actual failures of ESP. Based on the encouraging results of piloted failure prediction tool, deployment of solution will be extended to all other assets of the company as well.