This paper describes the opportunity to use high frequency real-time well information for the improvement of well integrity monitoring workflows and presents the results of a pilot workflow application for a large onshore asset. The typical well integrity monitoring workflow is a conditional monitoring workflow based on thresholds monitoring of parameter values or parameter derivatives. An example of such parameters is the well annulus pressure. In this work authors are trying to extend the conditions monitoring concept by adding pattern recognition technology for transient well periods and calculating accumulated flow assurance risks over the well's life, such as corrosion and wax depositions. The concept has been employed for an onshore, but remote field. Due to the field's location, service availability is limited and therefore the use of early diagnostics and the prediction of potential issues is critical for maintenance planning.
The field has a modern data gathering system, thus the use of high frequency data has been considered as an opportunity. After initial data analysis, a two-tier concept has been developed with an online machine learning system implemented for the detection of anomalies and early recognition of well events, as well as offline transient multiphase simulations for the characterization and differentiation of various cases. An anomalies pattern library, which describes the different anomalies detected, was developed manually for a selected number of wells and dynamic modeling in a multiphase transient simulator was used to create the additional patterns that were not observed but are probable to occur in the future. This dynamic modeling approach was also utilized to study patterns showing unclear well behaviors, to perform sensitivity analysis and for integrity issue characterization. The library was then employed as a training library for a machine learning system and scaled up for application to a larger number of wells. Additionally, the results of transient models have been used as a basis for surrogate models, which can be used by operational personnel for quick ‘what-if’ analysis. The developed concept has several benefits: on one hand, it is fast and robust for practical implementation, whilst on the other hand, it also provides us with the necessary understanding of the ongoing well processes.
The concept has demonstrated that many different anomalies can be detected (not only integrity-related) which creates additional value for operations. This paper describes practical cases with explanations of well behavior and gives examples of the early detection of well integrity issues.
