From Unstructured Frac Data to Decisions: An Automated WorkFlow to Unlock Operational Excellence

Abstract A persistent challenge in leveraging data analytics in hydraulic fracturing operations is the prevalence of unstructured data formats and the lack of some data on rock mechanics and reservoir parameters. In this paper, an in-house code was developed to majorly 1) automate data extraction from 200 non-tabular documents of frac operations, 2) investigate intercorrelations among 37 key attributes relevant to frac operations, and minorly 3) impute missing data that might surpass 40% of some features. Data was collected from test- and post-job frac operations reports across 6 fields in the MENA region. The workflow started with OCR preprocessing to convert scanned documents into machine-readable text. Regular expressions and text mining were used to automatically locate relevant tables, eliminating the time needed to examine the entire document – typically 20 pages. Next, data quality rules and software testing standards were enforced to ensure the integrity of the extracted tables. Then, data mining techniques were implemented to unveil complex patterns of unsuccessful jobs. Ultimately, second-order statistics, MICE, was investigated to impute missing data. The proposed approach successfully retrieved large amount of data from the unstructured reports where it has proved to be outperforming in extracting numerical information from unstructured data, compared to an LLM (Llama2 – 7B), primarily due to LLMs hallucination with inconsistent operational terminologies across reports. Moreover, it was observed that implementing inter-feature imputation techniques yielded reasonable results when applied on features with less than 50% missing data, where a subsequent post-imputation analysis was carried out to quantitatively assess the robustness of this approach showing average RMSE of 0.13 on a sample of features. The analysis of hydraulic fracturing operations revealed that treatment success is primarily influenced by closure pressure and near-wellbore friction, with failures predominantly occurring at closure pressures of 9,000-12,000 psi and near-wellbore friction exceeding 1,500 psi. Hi-way fracturing demonstrated the highest success rate (95%) compared to conventional (78%) and foam (75%) treatments. The investigation established critical relationships between proppant selection and fracture geometry, with successful treatments showing optimal performance for 16/30 mesh proppants at fracture widths of 0.07-0.13 inches. To sum up, the data set retrieved and constructed from operational reports contributed to developing quantitative thresholds for data-driven decision-making, which would not have been achievable without data mining techniques. This paper presents a comprehensive analysis of hydraulic fracturing jobs based on a large dataset of 200 wells in the MENA region. The introduced framework of data extraction, exploration, analysis and imputation for frac operations, is publicly available as an open-source code on GitHub, to enable researchers to further investigate and derive insights from legacy and unstructured frac documents.