Clasificación Automatizada de Facies
Labels rock type from well logs across hundreds of wells consistently, which a team of interpreters working separately cannot do.
Fills the rock between wells with geologically plausible detail instead of a smooth interpolation that no outcrop ever looked like.
Conventional geostatistics produces models that honour the wells and are far too smooth everywhere else. Flow behaviour is controlled by the connectivity of that in-between rock, which the smooth model gets wrong.
A generative model trained on analogue outcrop and high-resolution field datasets produces multiple realisations conditioned on well data and seismic attributes. Realisations are screened for geological plausibility by a geologist before any of them reach a simulator.
Plausible is not correct. The realisations are consistent with the data and the analogue, and nothing in the method makes them true. They are useful for spanning uncertainty and dangerous if a single realisation is promoted to "the model", which is the most common way this technique is misused in the industry.
Cada página de capacidad declara una limitación real. Si un proveedor no puede decirle dónde termina su producto, no le están diciendo la verdad sobre dónde empieza.
Labels rock type from well logs across hundreds of wells consistently, which a team of interpreters working separately cannot do.
Finds the bad intervals in the log archive and reconstructs curves that were never run, so the field has a complete set to work from.
Picks faults and horizons across a full seismic volume in hours, leaving the interpreter to judge rather than to click.
Searches for where the next wells should go, with drilling feasibility and surface constraints inside the search rather than applied afterwards.
A learned stand-in for the full reservoir simulator that answers in seconds instead of hours, so an engineer can test fifty scenarios instead of three.