Significant Improvement in Assessing Wellbore Integrity, Leveraged By Machine Learning

We are suggesting a unique hybrid wellbore health estimation solution, based on combining finite element analysis with machine learning. The access to underground energy resources is mainly provided by wellbores. It is estimated that more than 90% of the US total energy supply depends on some form of wellbore and the number of wells drilled will continue to increase. Ensuring long term integrity in these wells is a key objective in well design. However, failure in well integrity is a common issue, with around 35% of the wells presenting some sign of leakage, and which can be evidenced by sustaining casing pressure (SCP). Wellbore integrity is essential to ensure long term production, but it involves many uncertainties and rely on heavy computational simulations. Therefore, developing an effective and robust tool for well integrity analysis, assisted by machine learning (ML), can help reduce costs and enhance the life cycle of oil, gas, and geothermal wells. Finite element analysis (FEA) has proven to be an effective tool to assess wellbore integrity. FEA is able to predict fracture initiation and failure paths in the wellbore, but the simulations require extensive knowledge of the reservoir and several operational parameters. In this project, we propose the use of a ML approach to integrate the available data and eliminate the user bias on the estimation of wellbore parameters. ML is a powerful tool for predictive modelling, such as cement bond parameters, and to facilitate the acquisition of precise results in FEA. Based on the previous SCP patterns and available cement evaluation data, as training samples, the cement state can be reconstructed to allow evaluation of wellbore integrity. The main advantage of this approach is that the model only needs a small dataset and does not require analysis of the complex physics of the underlying processes to make predictions. The Phase I work plan will develop and assess the basic platform for running the forward model and using AI for inversion of the synthetic data enriched with the white noise. The proposed platform will be initially developed to work on workstations but will be moved the cloud in phase II. These proof-of-concept studies will help us prepare a foundation for marketing our technology for large-scale systems with the field data. Scientists at Penn State, with experience in developing simulation tools for the well- completion industry, will design material the forward model with input from Parthian Energy, whose experience in new AI product development includes proprietary analytical systems and instrumentation for the energy sector. The team will obtain economic validation from operators currently looking at these data for cement integrity evaluation. The commercial application of this technology will allow US operators to ensure the integrity of their wellbores and safety of their operations while maximizing injection/production rates in these wells by taking an advantage of this predictive technology. The applicants plan to use the technology as a cloud platform to license commercial service to other industries, such as unconventional oil and gas industry, enhanced geothermal systems and carbon sequestration operations via licenses to both existing and new customers.