Sector · Oil and gas
Physical AI for oil and gas
Wells, terminals and plants hold safety, inventory and process data in systems that were never joined, often on sites where the data may not leave the country. These are the questions engineers ask first, answered from 4 published CodeNinja Atoms reference architectures.
- What does a physical AI system for oil and gas look like?
- Which AI models can an oil and gas operator run on its own hardware?
- How much compute and hardware does AI in oil and gas need?
- Is it cheaper to own AI hardware or rent cloud GPUs in oil and gas?
- What ontology or object model does an oil and gas AI system need?
- Who approves the decisions an AI system makes in oil and gas?
What does a physical AI system for oil and gas look like?
A complete physical AI design for oil and gas names what to sense, which existing systems to join, the object model that joins them, the models and hardware, the three-year cost and the person who approves every action. CodeNinja Atoms has published 4 such reference architectures for oil and gas, each free to reuse under CC BY 4.0.
| Design | Country | What it does |
|---|---|---|
| Sovereign HSE Watch | Pakistan | An open reference architecture for predicting HSE incidents at an oil and gas operator in Pakistan, on the operator's own hardware, with no data leaving the country and no third-party AI service in the serving path. |
| Loop Integrity Watch | Pakistan | field hardware that rebuilds the signal path on three Modbus loops, and a governed inventory ontology that catches a flatlined or swapped radar gauge reading before the central inventory picture misleads anyone. |
| Corridor Guard | Pakistan | a validation-first study that digitises an operator's scanned pipeline records, verifies the corridor in the field, calibrates a direct current interference model in validated software and hands the re-runnable model over on the operator's own hardware, so every safe distance ships with its evidence. |
| Pipeline Integrity Watch | United States | An open reference architecture for system design in physical AI. A single on-premises model of a natural gas transmission estate in the United States: passive taps watch the OT network at every station and in the gas control centers, open-weight models watch compressor health and leak indicators at the station, and a control room assistant explains each alert and drafts a response for a pipeline controller who decides every action. The security directive evidence generates itself from the monitoring. |
Which AI models can an oil and gas operator run on its own hardware?
Each published oil and gas design names its models and why, and every model is open-weight or no model is used at all, so the operator can run it on hardware it owns.
| Design | Models |
|---|---|
| Sovereign HSE Watch | 6 self-hosted open-weight models, including GLM 5.3 (reasoning), Chronos-2 (forecasting), RF-DETR (vision), BGE-M3 (retrieval in English and Urdu) and PaddleOCR-VL 1.6 (scans) |
| Loop Integrity Watch | None: every integrity check (stuck value, swap, distortion) is deterministic |
| Corridor Guard | |
| Pipeline Integrity Watch | Mistral Large 3 675B (Apache-2.0) for the control room assistant, the compliance agent and the work surface; Chronos-2 (Apache-2.0) for telemetry forecasting with residual anomaly detection; RF-DETR Apache sizes for station camera scene detection; BGE-M3 (MIT) for retrieval |
| Design | Choice | What was picked | Why |
|---|---|---|---|
| Sovereign HSE Watch | Frontier reasoning model | GLM 5.3 open weights at FP8, self-hosted | Strongest open agentic model; the bespoke license exempts purely internal use from the model-as-a-service security-review trigger |
| Sovereign HSE Watch | Detector | RF-DETR, Apache-2.0 Nano to Large checkpoints, BF16 | The practical sovereign answer to the AGPL gate; Plus XL and 2XL excluded from the serving path |
| Sovereign HSE Watch | Forecaster | Chronos-2, Apache-2.0, about 0.48 GB at 32 bit | Zero-shot multivariate forecasting with no field-of-use restriction; weights may be held, fine tuned and redistributed |
| Sovereign HSE Watch | Embeddings | BGE-M3, MIT, FP16 | Dense plus sparse plus multi-vector retrieval in one pass with an 8,192-token window |
| Sovereign HSE Watch | Document parsing | PaddleOCR-VL 1.6, Apache-2.0, about 0.9B parameters, BF16 | Strong on degraded multilingual scans; no user or revenue threshold; fine tuning permitted |
| Sovereign HSE Watch | Tracking | Roboflow trackers, Apache-2.0 | Stable identity across frames without reintroducing copyleft after an Apache detector |
| Sovereign HSE Watch | Frontier node class | One node of 8 x 141 GB HBM GPUs (H200 class) | 1,128 GB holds the 904 GB FP8 footprint with KV cache headroom |
| Sovereign HSE Watch | Edge compute class | The operator's NPU/GPU accelerators, sized from measured stream and decode load | Sizing is stated as a requirement and verified at the phase-one survey |
| Sovereign HSE Watch | Cameras reused subject to ONVIF reuse gates | The existing Vision AI IP camera estate, stream evidence, with gaps priced as new | Reuse on measured density, angle and class purchases |
| Sovereign HSE Watch | Time synchronization holdover, driving linuxptp and chrony | OCP Time Card GNSS grandmaster with cameras and process tags | One defensible chronology across detectors, |
| Sovereign HSE Watch | Edge orchestration disconnected install | Red Hat OpenShift AI self-managed, model serving, registry, pipelines and | Documented disconnected procedure for workbenches |
| Sovereign HSE Watch | Serving runtimes node | KServe at the edge, vLLM on the central | Model serving matched to each tier's load |
| Loop Integrity Watch | Model register | Zero entries; all integrity logic is deterministic thresholds and rules | The failure modes are physical and protocol-level, so no learned model is needed, no weights are held and no license question arises |
| Loop Integrity Watch | Hardware class | Flameproof enclosure discipline | The junction boxes sit in the classified area, so the enclosure class is chosen against the zone at each exact position, not by part number |
| Loop Integrity Watch | Hardware class | Area classification as the gating document | The zone, gas group and temperature class are read from the operator's current classification drawing before any hardware is specified, and the position list is countersigned |
| Loop Integrity Watch | Hardware class | Reading an Ex certification marking | The Eex'n' marking is a Zone 2 restricted-breathing class, so a Zone 1 position would force a different, heavier enclosure and must be caught before ordering |
| Loop Integrity Watch | Hardware class | Equipment protection levels drive the purchase | Protection levels, not catalog convenience, decide the explosion proof junction boxes, 24VDC power supplies, breakers and M20 Exd glands |
| Loop Integrity Watch | Sensing | radar tank gauges on all 13 tanks | These are the field devices whose readings must arrive continuous, undistorted and unswapped; the design improves their signal path without modifying them |
| Loop Integrity Watch | Sensing | 24VDC field power supplies | They power the boosters and the loop electronics, and their health is a named failure mode with its own degraded status |
| Loop Integrity Watch | Sensing | 3-core 2.5 sq.mm CU/PVC/SWA/PVC loop cabling | The loop cabling is the medium the distortion travels in, so its schedule is recorded in the as-built against each loop |
| Loop Integrity Watch | Sensing | Modbus RTU loop polling via the boosters | The polling cycle is the pace of truth for every check, and the boosters are what make the three loops readable end to end |
| Loop Integrity Watch | Pattern | System of Context, primary | The central inventory picture is an ontology projection over the gauging host, never a shadow copy, so there is one place where tank truth lives |
| Loop Integrity Watch | Ground | The operator's own site application server, on-premises | The two containers run inside the operational technology boundary on the operator's own hardware, consistent with the Pakistani Cloud First posture, so nothing about tank truth depends on a link leaving the site |
| Corridor Guard | Model | PaddleOCR-VL 1.6, a vision-language document parser of about 0.9 billion parameters, bf16 precision, Apache-2.0 | Digitises the scanned package under a license the operator holds outright, running as released with human reconciliation before any value enters the register |
| Corridor Guard | Model | CDEGS or COMSOL, industry-recognized validated electromagnetic and grounding software at the site tier | The governing clauses require validated software and score unsupported modelling lower, so the interference physics never leaves mandated tools |
| Corridor Guard | Hardware class | One server-class host of the 48 GB PCIe inference GPU class inside the operator's estate | Holds the parser weights under 2 GB and leaves the balance of the card for concurrent solver runs and parser sessions |
| Corridor Guard | Sizing rule | Weights against usable memory first, solver memory with headroom second; context length never binds | The parser reads single scanned pages, so the kv cache ceiling is never the constraint the card is sized on |
| Corridor Guard | Sensing | Pipe-to-soil DC potential at corridor test points, in volts against the copper-copper sulfate electrode with the protective criterion at -0.85 V, plus four-pin soil resistivity by corridor section | The physical world is sensed by electrical measurement, not imaging: no camera duty exists anywhere in the scope |
| Corridor Guard | Pattern | A provenance-carrying validation register, marking every attribute measured, inherited, assumed or unknown | Data errors are the contractor's cost, so every correction has to be traceable and cost-defensible |
| Corridor Guard | Pattern | A re-runnable model handed over on the operator's own hardware, with no consultant-side copy surviving handover | The compiled records and software are the procuring agency's absolute property, and a federal public sector corporation sits under the national cloud-first policy |
| Corridor Guard | Ground | One site-tier host on existing power and cooling inside the operator's facility | No field compute, no new network and no data crossing the boundary, so the stack runs where the data already sits |
| Pipeline Integrity Watch | Frontier work surface model | Mistral Large 3 675B (Instruct), FP8, about 682 GB of weights | Apache-2.0 with no field-of-use restriction or Entity-List exposure; carries the assistant, the compliance agent and the work surface on one control-center node |
| Pipeline Integrity Watch | Telemetry forecaster | Chronos-2, about 0.48 GB at 32 bit | Zero-shot covariate forecasting with residual anomaly detection; small enough to run offline on the station edge boxes |
| Pipeline Integrity Watch | Camera detector | RF-DETR Apache sizes at BF16, about 61 to 68 MB for Nano to Large, about 254 MB for 2XL | Apache-2.0 checkpoints only with the version pinned; fine-tunable on site footage and exportable to the edge NPU |
| Pipeline Integrity Watch | Retrieval embedding | BGE-M3, MIT, about 1.1 GB at fp16 | Hybrid dense and sparse retrieval in one pass keeps tag names, error codes and part numbers findable |
| Pipeline Integrity Watch | Edge compute class | Volume-production modules in 7W to 60W envelopes | Chosen by thermal envelope for station cabinets; runs the station models offline and needs no export license for a domestic install |
| Pipeline Integrity Watch | Hazardous-area ground | Edge Compute Enclosures In Hazardous Areas | Station electronics sit inside classified areas, so the enclosure class is a design input, not an afterthought |
| Pipeline Integrity Watch | One-way egress | One Way Path Out Of An OT Zone and Data Diode One Way Transfer | Telemetry and evidence leave the OT zone with no return path to misconfigure |
| Pipeline Integrity Watch | Placement rule | When Inference Belongs At The Edge | Detection survives backhaul loss at the station; reasoning stays where the context and the people live |
| Pipeline Integrity Watch | Sensing | Passive OT network taps at every station and both control centers, fixed methane point sensors at the highest risk stations, vibration sensors on compressor frames and bearing housings, discharge temperature probes, existing line pressure and flow transmitters, PTP grandmaster clocks | The physical coverage every model in this table reads |
| Pipeline Integrity Watch | Standing patterns | System of Context as the primary architecture, zone-and-conduit boundary, edge-first detection, shadow mode before any controller-facing alert | Every row above was tested against these patterns before it was committed |
| Pipeline Integrity Watch | Ground it runs on | Operator-owned control-center GPU node and station cabinets, K3s, Triton Inference Server, vLLM, Harbor registry | Open weights held outright on premises, inside the operator's boundary, with updates through an approved change path |
How much compute and hardware does AI in oil and gas need?
The compute follows from the models: the published oil and gas designs size it as follows, from no new hardware to a full GPU node.
| Design | Part | The design |
|---|---|---|
| Sovereign HSE Watch | Frontier compute | One node of eight 141 GB HBM-class GPUs holds GLM 5.3 at FP8 (753 GB of weights, 904 GB with headroom) |
| Sovereign HSE Watch | The hard dependency | 141 GB-class accelerators need a US export licence for Pakistan (Country Group D:4); the rollout's first gate confirms installed hardware first |
| Loop Integrity Watch | Compute | Two containers on the operator's own site application server, inside its OT boundary |
| Loop Integrity Watch | Field scope | 13 fuel tanks on 3 Modbus RTU loops; 3 booster installations engineered from a measured signal survey inside explosion proof junction boxes |
| Corridor Guard | Compute | One server-class host of the 48 GB PCIe inference GPU class inside the operator's estate; no field compute, no cloud, nothing crosses the boundary |
| Pipeline Integrity Watch | Compute | At each station, volume production edge modules in 7W to 60W envelopes inside hazardous area enclosures, detecting through backhaul loss; centrally, one node of eight 141 GB HBM-class GPUs in the gas control center |
Is it cheaper to own AI hardware or rent cloud GPUs in oil and gas?
Each oil and gas design prices three years of ownership in its Appendix A, with every price cited, against renting the same capacity from a cloud region at its deepest three-year commitment where hardware is bought.
| Design | Line | Three years |
|---|---|---|
| Sovereign HSE Watch | Three-year cost, owned | About 670,000 US dollars with support and power at Pakistan's industrial tariff |
| Sovereign HSE Watch | Three-year cost, rented | 1.1 to 2.8 million US dollars for the same GPUs in the nearest cloud region; no hyperscaler runs a region in Pakistan |
| Loop Integrity Watch | Three-year cost | No compute to price; the field equipment is priced by OEM quotation against the survey (Appendix A) |
| Corridor Guard | Three-year cost | About 19,500 US dollars to own the host over three years (18,700 to 20,200), about four fifths of the deepest three-year AWS commitment for one L40S card, which could not keep the data in Pakistan in any case (Appendix A) |
| Pipeline Integrity Watch | Three-year cost | About 465,000 US dollars to own the control center node for three years (432,000 to 485,000), about three quarters of the 625,000 the deepest AWS commitment in US East costs and well under the 1.66 million on demand; the station modules, enclosures, cameras, diodes and sensors are priced by quotation (Appendix A) |
What ontology or object model does an oil and gas AI system need?
The object model is the part that makes the system an ontology rather than a pipeline: typed objects for the things in the oil and gas operation, with properties, status values and typed links. Every published design ships its object model as hyper-ontology/1 JSON that loads into Hyper Ontology.
| Design | Size | Objects | Download |
|---|---|---|---|
| Sovereign HSE Watch | 12 objects, 14 links | Operating Facility / Site, HSE Equipment, HSE Incident, Near Miss Report, Corrective Action, Inspection / Audit Record, HSE Document, Sensor Reading, Anomaly Event, Agent Recommendation, HSE Person, HSE Role | objects.json |
| Loop Integrity Watch | 15 objects, 15 links | Fuel Tank, Radar Tank Gauge (RTG), RTG Communication Loop, Modbus Booster (Repeater), Explosion Proof Junction Box, Gauge Reading, Data Quality Event, Central Inventory Picture, Loop Wiring As-Built, Service Order, Warranty and Support Record, OEM Authorization Letter, HSE Work Permit, Instrumentation Technician, Location Engineer | objects.json |
| Corridor Guard | 15 objects, 9 links | 8-inch Line, 10-inch Loop Line, Original Anode Bed, TEG Unit, Anode 1 (MP-3.40), Anode 2 (MP-6.97), Insulating Joints PFIJ, Electrode Station, HVDC Transmission System, CP Test Point, Cathodic Protection System, Soil Investigation Report, Detail of Available Data for Assessment, DC Interference Assessment, Mitigation Measure | objects.json |
| Pipeline Integrity Watch | 14 objects, 13 links | Compressor Station, Meter Station, Pipeline Segment, Compressor Unit, OT Device, Pipeline Controller, Alarm, Leak or Abnormal Operation Event, Remote Access Session, Work Order, Security Directive Control, Historian, CMMS, Hazardous Area Classification | objects.json |
Who approves the decisions an AI system makes in oil and gas?
In every published oil and gas design a named person makes the decision that changes the physical world; the system prepares it.
| Design | Human control |
|---|---|
| Sovereign HSE Watch | Every recommendation is approved or rejected by a named person; nothing executes on equipment |
| Loop Integrity Watch | Every data quality event is acknowledged and resolved by a named person; the location engineer signs acceptance |
| Corridor Guard | A named corrosion control engineer closes every discrepancy and accepts every assessment; no value enters the register without provenance and no safe distance ships without its evidence |
| Pipeline Integrity Watch | The system never writes to a field device and shutdown stays with the certified control paths; a hardware one-way diode is the only way out of each station OT zone; every decision-influencing model runs in shadow mode before any controller-facing alert, and a named controller confirms or clears every leak event |
Every answer on this page is drawn from the papers linked in it: each paper's At a glance table, model register and cost appendix. Full text for agents: llms-full.txt. Designed on Praxis; object models load into Hyper Ontology.
Want one of these designs for your own operation? Talk to us.