Machine learning and Artificial Intelligence applications in artificial lift systems have seen a growth in importance recently and are no longer a “nice to have” but essential tools for design, optimisation and failure prediction. Real time optimisation techniques help to optimise the production, however, do not necessarily holistically consider the equipment reliability and best operating range.
Whenever a failure occurs, the reasons could be attributed to more than one condition and hence, Root Cause Analysis is often a complicated process involving visual inspection and laboratory analysis to confirm the reason for the failure. This can lead to production deferment and lost run time. It also may involve workovers and inspection time adding to the operational expenses. The best operating envelopes for the lift systems is also up to the discretion of the optimisation engineer and may not be visited often enough to account for changing operating conditions. This may lead to the system operating in a conservative fashion leading to reduced production. As an example, a rod pump might operate at a lower speed anticipating high rod stresses based on historical operation. In some instances, the systems might not be designed for the desired operating environment and may pose a threat to its reliability. There is a need for a technology which would serve as a guide to overcome these challenges using real time diagnosis and provide foresight into future operations and potential problems that may increase operator costs. Optimisation platforms are mostly based on first principles and therefore can be successfully applied to different kinds of wells.
However, the same may not be true for predictive analytics as it uses a combination of first principles and data driven technology. The predictive analytics approach therefore, should be exclusive for each well depending on the available data, its operating conditions, its potential and its contribution to the overall asset performance.
Emerson’s predictive analysis for artificial lift using the Knowledge Net (KNet) Machine Learning Platform is an engineered solution for predicting failures or abnormal working conditions before their onset and providing actionable insights to mitigate the same. The solution utilises historical data from the wells to build a solid offline well model which then gets trained on the real time data as the well comes on to production. Easy access to third party data sources makes the process convenient. As the lift systems are subject to many complex events that might lead to a potential failure, the principal component analysis helps in the elimination process. With a comprehensive Failure Mode Effect Analysis and Root Cause Analysis library, the solution captures, in real time, the abnormality and translates it into a potential run time deviation. With a prior indication, the condition can be corrected or interventions planned more efficiently. The dynamic modelling of key performance indicators based on system intelligence helps in driving asset performance, to identify the priorities relevant to the existing conditions. This widens the scope from mere well performance to complete asset performance enhancement.
An attractive return on investment (ROI) of less than a year for nominal production values with normal operating conditions makes this an attractive option for operators. The results are even better in harsh operating conditions. With the current market dynamics and restricted field movement, this technology becomes a necessity for seamless operations and uninterrupted production.