Real-Time Probabilistic Inference of Influx Dynamics Using Coupled Thermo-Solubility and Two-Phase Flow Models

Abstract

Gas influx events during drilling remain a persistent source of operational risk because the observable signals used for detection and response are indirect, delayed, and confounded by uncertain multiphase physics. Surface pressure and flow measurements provide high-availability telemetry, yet they are filtered through annular hydraulics, compressibility, and regime-dependent slip, while the downhole state of the influx evolves through dissolution, expansion, and migration. This paper develops a real-time probabilistic digital-twin framework for kick detection and response that explicitly couples thermo-solubility, transient two-phase transport, and regime-switching closures with sequential Bayesian inference and risk-aware control. The central technical contribution is a hybrid reduced-order model whose continuous state captures pressure, holdup, dissolved-gas inventory, and thermal effects, while a discrete latent variable encodes flow-regime-dependent slip and friction structure. The model is embedded in an online estimator that assimilates surface pressure and flow data to infer posterior distributions over influx size, location, and composition-sensitive transport parameters, with quantified uncertainty and detectability metrics. A risk-sensitive decision layer then maps the inferred state to choke and circulation policies formulated as chance-constrained optimal control, with constraints representing pressure integrity margins and operational actuation limits. The resulting architecture targets field utility by enforcing computational budgets, bounded-error updates, and stability of closed-loop decisions under noisy telemetry. The framework is intended as a principled bridge between mechanistic hydraulics and data-driven signal interpretation, emphasizing deployable inference and control rather than retrospective reconstruction.