A well-to-surface digital twin for cyclic steam and rod pump operations.
Heavy oil wells lose value in the gap between the steam cycle and the lift system. WellTwin AI couples a reduced-order reservoir-heating model to a rod-pump model so an operator can test a steam design and a pump set-point together, before committing a cycle. This build is a demonstration on deterministic synthetic data, framed around the Baghewala heavy-oil context of Oil India Limited.
Field overview · synthetic
Deterministic mock telemetry, refreshed from a fixed seed
CSS and SRP are optimised separately, so the well is optimised by nobody.
In cyclic steam stimulation the reservoir is heated in campaigns; in artificial lift the pump runs continuously against a moving inflow. Decisions on each side are usually taken by different teams on different timescales, using spreadsheets and last-cycle intuition.
Steam is spent, not designed
Cycle volume, injection rate and soak duration are often carried forward from the previous cycle. The last quarter of injected steam frequently buys very little incremental oil, which shows up as a rising steam-oil ratio.
Pumps fight a moving target
Inflow after a steam cycle rises then declines. A fixed pump speed over-pumps late in the cycle: fillage collapses, rod loading climbs and failures follow, and the loss is mis-read as reservoir depletion.
No safe place to test a change
There is no cheap way to ask what happens if steam is cut 20% and soak extended two days. Every experiment is a real cycle on a real well, so practice ossifies.
One twin, from reservoir heat to surface separator.
The prototype keeps a single connected model of the well: heat delivered by the steam cycle, mobility gained during soak, inflow to the pump, and the lift envelope of the rod string. Change any input and the whole chain responds.
Steam design
Cycle tonnage, injection rate, steam temperature and soak days.
Reservoir response
Heat penetration, viscosity reduction, peak rate and decline shape.
Lift envelope
Pump displacement vs inflow, fillage, rod loading, failure risk.
Decision output
Scenario comparison, guarded recommendation, expected uplift.
Schematic field view
Click any well inside the demo to open its twin
Scenario comparison · BGW-08
Baseline practice vs twin-optimised set-point
Illustrative output of a simplified surrogate model on synthetic data. Not a field prediction; independent engineering validation and field calibration are required before deployment.
Eighteen working screens, built like a control room.
Signing in opens the operational product: situational awareness, the twin, the simulator, and the assurance layer that keeps a model honest.
Command Center
Field KPIs, schematic map, live alert feed and today's recommendations.
Digital Twin
Well-to-surface schematic with modelled vs measured channels for BGW-08.
What-If Simulator
Move steam, soak and pump inputs; compare up to three saved scenarios.
CSS Operations
Injection / soak / production phase board with steam accounting.
SRP Optimisation
Pump envelope, fillage map and rod-loading guardrails.
Predictions
P10/P50/P90 rate forecast and 30-day failure risk by well.
Optimisation Centre
Field-wide ranked opportunity list under a steam constraint.
Operator Recommendation
Accept, defer or reject with rationale, guardrails and an audit trail.
Alerts & Risk
Severity triage, acknowledgement and a likelihood/impact matrix.
Maintenance
Predictive work orders with crew, hours and deferral risk.
CSS Cycle History
Cycle-by-cycle steam, oil, SOR and outcome versus plan.
Reports
Preview and download a plain-text cycle, health or business-case report.
Data & Model Health
Tag coverage, freshness, staleness and drift by model.
Model Explainability
Feature contributions behind each flagged diagnosis.
Judge Demo
A guided five-minute path through the full story.
Walk the full operator journey.
Sign in with the demo account and follow the guided judge path from a field alert to an accepted, guard-railed recommendation on BGW-08.