Towards a Novel Non-Intrusive Stuck-Point Indicator for Drilling Operations
Stuck-pipe events remain among the costliest failure modes in drilling operations, contributing roughly one quarter of non-productive time and producing major financial losses across the oil and gas industry. Once the pipe becomes stuck, recovery is governed by a simple but decisive question: where is the stuck point? That location defines the effective free length of the drill string and therefore constrains any remediation decision. Conventional free-point tools rely on intrusive downhole intervention and strain- or magnetic-based interpretation, introducing operational delay and measurement uncertainty. A non-intrusive, physics-based, and computationally efficient alternative is therefore needed.
This project addresses that need by formulating stuck-point localization as a vibration-based inverse problem implemented within COMSOL. The key premise is that the location of the stuck point is encoded in the vibration response of the free portion of the drill string. However, this response is strongly altered by the surrounding drilling mud through added mass, added damping, and history-dependent fluid effects, which collectively modify both transient decay and resonance behavior. Therefore, successful identification requires a forward model that preserves these fluid–structure interaction mechanisms while remaining efficient for repeated evaluation. The problem is addressed through three sequential phases that transform high-fidelity multiphysics modeling into a practical inference tool.
The first phase developed an accelerated reduced-order model (ROM) for fluid-loaded drill-string vibrations fully within COMSOL. High-fidelity multiphysics simulations identified the mechanisms that must be retained in a reduced representation: mud shifts resonant frequencies through added mass, increases attenuation through added damping, and introduces memory effects through unsteady fluid forces. Based on this physics, the ROM was formulated as a one-dimensional beam model enriched with internal dynamic states that capture frequency-dependent added mass, damping, and memory effects without resolving the full flow at each evaluation. The resulting ROM acts as a fast forward solver that generates the vibration response associated with a given excitation input, while also enabling parameter identification by using input–output signals and frequency-response functions. A key innovation was the use of COMSOL Application Builder to construct a system-identification module that estimates ROM parameters within the same multiphysics environment.
The second phase verified the ROM against high-fidelity finite-element results. After calibration, the model was tested across multiple excitation cases and effective pipe lengths and was shown to reproduce the same transient decay, spectral content, frequency shifts, and damping trends observed in full fluid–structure interaction simulations. Time-domain comparisons showed near-perfect agreement between reduced-order and high-fidelity responses, confirming that the ROM preserves the coupled dynamics governing the inverse problem, particularly added mass and added damping.
The third phase used the validated ROM to perform stuck-point identification through surrogate inference models trained on ROM-generated data. Because the ROM is both physically grounded and efficient, it was used to generate datasets spanning different pipe lengths and excitation conditions. Neural-network and Gaussian-process surrogates were then trained to map measured vibration signals directly to stuck-point location. This yields a near real-time prediction capability that retains the physical consistency of the underlying multiphysics model. The overall contribution is an end-to-end COMSOL workflow integrating multiphysics modeling, system identification, model reduction, verification, and scalable inference for non-intrusive stuck-point localization.
