DSFB: Structural Residual Semiotics Engine for Upstream and Midstream Oil and Gas Systems

Abstract—Modern oil and gas supervisory systems produce dense,high-frequency sensor residuals that are routinely discarded after prob-abilistic anomaly filters have rendered their verdict. These discardedresiduals are not noise; they carry structured information about the phys-ical regime transitions, load migrations, and boundary interactions thatcharacterize subsea hydraulic circuits, pipeline flow-pressure couplings,rotary drilling mechanics, and rotating equipment dynamics. This pa-per introduces the Drift–Slew Fusion Bootstrap (DSFB), a deterministic,read-only residual structuring framework that decomposes discardedresiduals into a signed drift component and a magnitude-rate slewcomponent, then maps the resulting trajectory through an admissibil-ity envelope into a human-readable operational grammar. DSFB nei-ther predicts failure, nor replaces Real-Time Transient Models (RTTM),Kalman filters, Model-Predictive Controllers, Statistical Process Control(SPC) charts, or any existing probabilistic or control system. It operatesstrictly as an observer: it reads signals, writes no setpoints, and issuesno alarms. Its output is a deterministic event log whose entries—driftaccumulations, slew spikes, envelope violations, boundary grazings,and recovery trajectories—are human-interpretable analog-to-symbolmappings grounded in subsurface and surface physics. Evaluation onsynthetic time series across all four oil-and-gas domains and on realsensor logs from three independent sources—Petrobras 3W (9 087 realtimesteps; 12 labelled well episodes; six fault categories), Equinor Volvewell 15/9-F-15 (5 326 depth-steps of WITSML surface-torque data; NCR= 18.9), and RPDBCS ESPset (6 032 vibration snapshots from 11ESP pump units; EnvViolation rate 20.7% vs. 20.4% true fault rate)—demonstrates measurable reduction in episode count, noise compres-sion ratios between 7:1 and 13:1 on synthetic data, 18–19:1 on thecontinuous-stream real datasets (Petrobras 3W and Equinor Volve), and1.5:1 on the ESP snapshot classification dataset (a structural propertyof the dataset rather than a framework limitation; see Section 8.4),and operator-meaningful state annotations that require no machine-learning inference and no system modification. The framework’s limita-tions—including its complete dependence on upstream sensing quality,the calibration burden of domain-specific envelope parameters, and itstotal absence of predictive capability—are stated explicitly and at length.Index Terms—residual structuring, drift decomposition, slew analysis,operational semantics, subsea hydraulics, pipeline integrity, drilling sys-tems, rotating equipment, deterministic observer, SCADA integration