Probabilistic Failure Occurrence Estimation Based on Survival Analysis Modeling: First Approach

Abstract

Workover is a critical operation performed by drilling rigs, given the operational impact of resources involved in this activity. The yearly estimation of workover occurrences for drilling campaigns or mature fields is challenging for resource allocation, given the need to maintain the production of oil and gas fields. Based on these aspects, we established a group of business rules to identify failure historical with experts from a large Brazilian energy company. We presented an adapted survival analysis methodology to estimate failure occurrences. We scheduled several meetings with workover experts from a large Brazilian company to establish the business rules required to identify failure historical occurrences. To estimate future failure occurrences, it was necessary to adapt a traditional survival analysis into a well's failure survival analysis methodology that considers all premises from the business rules. The failure survival analysis methodology generates a survival curve; however, it still needs to be fulfilled, considering the rarity of failure occurrences. The next step was fitting a continuous distribution into the survival curve to complete the information. The continuous distribution represents the shape of the survival curve; thus, we performed linear regression to adjust the continuous fitted distribution in real data. We apply our methodologies to real well's data from two Brazilian fields. The failure identification methodology could identify most failures based on the business rules. We tested the failure estimation methodology and compared the total real failures with the total estimated failures. The failure methodology can be applied for failure estimation in new campaigns (using a similar field) or following years in mature fields (using the field's history itself). The results allow a future multivariate approach, introducing new features to multivariate survival analysis or a machine learning model. A multivariate failure model can generate more robust results, providing better estimation. The first novelty is the group of business rules established by workover experts to identify failure historical. The documentation and standardization of these rules compose the base for future works focused on probabilistic estimations of failures and workovers, building the premises required by any survival analysis or machine learning model. The second novelty is the methodology developed to adapt a traditional survival analysis into a failure survive analysis.