AI-Augmented Digital Twin for Integrated Surface-Subsurface Fracturing Management and Real-Time Risk Mitigation

Abstract

Abstract This paper presents an integrated surface-to-subsurface digital twin platform developed for end-to-end management of hydraulic fracturing and acidizing operations. The work addresses a persistent visibility and knowledge gap in stimulation campaigns: surface equipment layout and health status are often managed separately from downhole treatment dynamics, while historical job records remain trapped in unstructured reports that are difficult to reuse for offset-well guidance. The platform combines six tightly coupled capabilities. First, a data-driven surface digital twin supports custom, true-to-scale modeling of pumps, wellheads, treating lines, and equipment rooms through free drag-and-drop placement, reducing the need for repeated field surveys before equipment mobilization. Second, sensor streams drive a live site twin in which any asset can be inspected and critical parameters such as treating pressure and slurry rate are monitored with configurable alarms. Third, artificial intelligence (AI) unstructured-data extraction has been used to harvest and standardize approximately 300 acidizing wells and 600 fracturing wells into a unified engineering schema consistent with field data templates. Fourth, natural-language-to-SQL (NL2SQL) analytics enable rapid offset-well and pad-level queries across this asset base. Fifth, the resulting knowledge supports design improvement and real-time design-versus-actual tracking with early warning when execution drifts from plan. Sixth, coupling with professional fracturing software, together with historical offset parameters, provides risk analysis and recommended operating envelopes, while surface–wellbore twin visualization unifies field and downhole situational awareness. Application to representative tight-gas fracturing wells and a deep carbonate acid-fracturing well demonstrates how structured offset comparisons and stage-level deviations can be surfaced for decision support. The novel contribution is the combination of editable surface twins, large-scale unstructured stimulation data assets with NL2SQL access, and software-assisted risk recommendation inside one operational environment, extending digital twins beyond visualization alone.