Sector · Heavy industry and construction
Physical AI for heavy industry and construction
Mills, factories and building sites produce counts, alarms and progress data in many systems, and each one sees only part of the operation. These are the questions engineers ask first, answered from 5 published CodeNinja Atoms reference architectures.
- What does a physical AI system for heavy industry and construction look like?
- Which AI models can a heavy industry and construction operator run on its own hardware?
- How much compute and hardware does AI in heavy industry and construction need?
- Is it cheaper to own AI hardware or rent cloud GPUs in heavy industry and construction?
- What ontology or object model does a heavy industry and construction AI system need?
- Who approves the decisions an AI system makes in heavy industry and construction?
What does a physical AI system for heavy industry and construction look like?
A complete physical AI design for heavy industry and construction 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 5 such reference architectures for heavy industry and construction, each free to reuse under CC BY 4.0.
| Design | Country | What it does |
|---|---|---|
| Structure Phase Watch | Saudi Arabia | one live model of a construction site's structure phase that forecasts schedule slips three days out and shows safety breaches as they happen, on the contractor's own hardware inside the Kingdom. |
| Steel Count Ledger | Pakistan | cameras and a GPU industrial PC at every casting strand and cooling bed count billets, ingots, rebars and girders as they are made, publish the counts through a one-way link into one record the authority owns, and let revenue officers reconcile counted against declared production. |
| Factory Fire Watch | Saudi Arabia | live, read-only visibility of fire alarm panels, fire pumps, fire water tanks and energy meters across an operator's highest-risk factories, read through contacts, PLC inputs and LoRaWAN into an IoT platform hosted in Saudi Arabia, with never a write path into certified life-safety equipment. |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Saudi Arabia | An open reference architecture for system design in physical AI, built for a cement company in Saudi Arabia that runs several integrated plants, each with a preheater and precalciner kiln line of several thousand tonnes of clinker a day, where owning the inference node that runs it costs about 570,000 US dollars over three years. When a new alternative fuel lot drifts the burning zone over a shift, the design joins the streaming tags, the fuel lot, the raw meal, the laboratory results, the shell scan and the shift log into one argument, so the control room operator sees the likely cause, the recovery moves that worked in matching past upsets and the setpoint adjustments with their expected effect on thermal energy per tonne, and every confirmed outcome is kept as the memory the next shift reads. The models run entirely on the company's own hardware in the country under the national energy efficiency program's reporting, the environmental permit's emission limits, the national operational technology cybersecurity controls and the company's own management of change procedure. |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Pakistan | An open industrial AI reference architecture for system design in physical AI, built for a fertilizer manufacturer in Pakistan whose ammonia and urea complex, designed to run steadily between annual turnarounds, has been stopped and restarted repeatedly in 2026 by gas supply suspensions, where owning the inference node that runs it costs about 572,000 US dollars over three years. When the gas comes back after a two week suspension, the design joins the historian, the work orders, the inspection and corrosion records, the vibration readings and the shift log into one story per piece of equipment, so the restart agent drafts the plan from the last three restarts, the synthesis compressor's seal is flagged for inspection before the start, and the shift superintendent confirms every step. The models run entirely on the operator's own hardware inside the country, behind a hardware data diode that makes the read of the control system physically one way. |
Which AI models can a heavy industry and construction operator run on its own hardware?
Each published heavy industry and construction 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 |
|---|---|
| Structure Phase Watch | 5 self-hosted open models: GLM 5.3 (reasoning), Chronos-2 (forecasting), BGE-M3 (multilingual retrieval), RF-DETR (vision), Roboflow trackers |
| Steel Count Ledger | 3 self-hosted open models: RF-DETR (detection, fine-tuned per product type and mill), Roboflow trackers (identity across camera fields), GLM 5.3 (compliance work surface) |
| Factory Fire Watch | One: IBM Granite Tiny Time Mixers (TTM-R2), about 0.85 million parameters, Apache-2.0, on CPU inside the platform; threshold evaluation on the safety path is rule logic |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | GLM 5.3 (GLM-5.3 bespoke licence, 753B at FP8) for cause drafting, handover drafting, ontology maintenance and the command centre's cited answers; Chronos-2 (Apache-2.0) forecasting the stable band on the streaming tags to raise the drift trigger; BGE-M3 (MIT) embedding past upsets, shift logs and maintenance history in Arabic and English; no third party model service, nothing trained into weights |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | GLM 5.3 (GLM-5.3 bespoke licence, 753B at FP8) for the restart, cycle damage and turnaround drafting agents and the cited conversational interface; Chronos-2 (Apache-2.0) forecasting historian tags and thermal cycle stress residuals on the site server; BGE-M3 (MIT) embedding work orders, findings, readings and shift log entries for cited retrieval; vision models on Frigate NVR and ONNX Runtime for steam plume, flange leak, operator, scaffold and fire or smoke; no third party model service, nothing trained into weights |
| Design | Choice | What was picked | Why |
|---|---|---|---|
| Structure Phase Watch | Detection and | RF-DETR fine tuned on site footage | Apache-2.0 from package to checkpoints; |
| Structure Phase Watch | segmentation | the JV owns and hardens weights a | packaged safety product would rent |
| Structure Phase Watch | Tracking | Roboflow trackers, ByteTrack class | Apache-2.0 motion only identity without the AGPL exposure of the BoxMOT collection |
| Structure Phase Watch | Forecasting | Chronos-2 universal forecasting | Apache-2.0 with no field of use restriction; weights may be held, fine tuned and redistributed on site |
| Structure Phase Watch | Embedding | BGE-M3 hybrid multilingual embedder | MIT with no field of use restriction; multilingual retrieval across a workforce tagged by language |
| Structure Phase Watch | Frontier | One node of eight 141 GB HBM GPUs (H200 | 1,128 GB holds the 904 GB of FP8 weights, |
| Structure Phase Watch | hardware | class) | KV cache and activations with headroom for sessions |
| Structure Phase Watch | Edge compute powered enclosures | Jetson Orin industrial edge class in solar coverage at gates and laydown areas | Detection survives patchy private LTE |
| Structure Phase Watch | Site inference | Eight L40S class PCIe GPUs, 48 GB class | Forecasting, embedding and breach |
| Structure Phase Watch | server | aggregation on JV hardware in the site data | room |
| Structure Phase Watch | Camera class Profile S/T; thermal at 9 Hz or less | Fixed visible cameras, Frigate ready ONVIF a reuse survey holds; the 9 Hz ceiling keeps | The existing 38 fixed cameras serve where US origin thermal units outside the 6A003.b.4.b export license to Saudi Arabia |
| Structure Phase Watch | Sizing rules gain, with a cleaning interval as a design parameter | Enclosure thermal budget including solar accuracy problems unless the thermal budget is engineered first | Dust, glare and 50 C summer ambient fail as |
| Structure Phase Watch | Sensing haulage GPS for 40 tracked trucks plus proposed trackers for 20 mixers, weather station, biometric turnstiles, batch plant SCADA | Crane anti-collision logs, crawler telematics, crews actually did, joined through the adapters | Ground truth of what cranes, trucks and |
| Structure Phase Watch | Pattern recommendation with a named approver PA R T I I I · C H A P T E R 9 Shadow Mode Comes Before Any Flag Is Trusted Three phases with counted items and hard gates prove the slip forecast and the breach detection against the record before anyone acts on a flag. Chapter 8 fixed the models, their licenses and the hardware classes they run on. This chapter sets out how the design earns the right to be trusted: three phases with counted items, workstreams and hard exit gates, so the slip forecast and the breach detection are proven against the record before anyone acts on a flag, and the work can be stopped cheaply if a gate does not hold. | Adapter tier, one object model, never directly, and every re-sequencing | Every source enters through an adapter, output stays a recommendation a person approves |
| Steel Count Ledger | Detection model | RF-DETR, Nano to Large checkpoints, fine-tuned per product type and per mill | Apache-2.0 from package to checkpoints; the operator owns and hardens the weights that count its product |
| Steel Count Ledger | Tracking | Roboflow trackers, OC-SORT and ByteTrack class | Apache-2.0; holds identity across frames so one billet crossing two camera fields counts once |
| Steel Count Ledger | Work surface model | GLM 5.3, 753B parameter mixture of experts, FP8 | Frontier open weights for the operator's compliance agents; 904 GB against one node of 1,128 GB; bespoke license allows commercial use, fine-tuning and redistribution |
| Steel Count Ledger | Edge compute | GPU-accelerated industrial PC class, sealed, wide temperature, cooled enclosure | Sized from stream decode plus inference with headroom, hardened for mill heat, dust and electromagnetic interference |
| Steel Count Ledger | Frontier node | One node of eight 141 GB HBM GPUs, H200 class, in a dedicated data center | The class is export controlled for Pakistan, so the node runs where an export license naming the end user holds, checked at the phase one licensing checkpoint |
| Steel Count Ledger | Cameras | IP66/67 HDR industrial camera class, chosen from the pixel-density table per installation point; thermal, ECCN 6A003, only where hot billets blind visible cameras | Survives the mill environment while meeting country legality per model |
| Steel Count Ledger | Enclosures and power | Closed-loop cooled enclosures, vibration-damping mounts, surge protection, UPS with clean shutdown under NUT monitoring | A failing cooler is caught before the industrial PC cooks |
| Steel Count Ledger | Site networking | Hardened PoE industrial switches, VPN tunnel, IEC 62443 zoning | The camera zone stays segmented from mill control |
| Steel Count Ledger | Time synchronization | PTP grandmaster with GNSS and OCXO | Count events reconcile across systems only when every clock agrees |
| Factory Fire Watch | Model | IBM Granite Tiny Time Mixers TTM-R2, about 0.85M parameters, FP32, about 0.003 GB | CPU-only forecaster inside the operator's platform hosted in Saudi Arabia; Apache-2.0 with no field of use restriction leaves the artifact owned outright |
| Factory Fire Watch | Hardware class | Rugged enclosure, mounting, power and UPS with NUT-class clean shutdown; IP ratings per IEC 60529; NEMA enclosure types; SABER conformity | Outdoor exposure to dust, direct sun and hose washing, with certification lead time built into the bill of materials |
| Factory Fire Watch | Hardware class | Fanless hardened industrial switches; copper-versus-fibre run selection | Fibre between buildings and near the high-current systems being metered; copper for short powered drops |
| Factory Fire Watch | Sensing | FACP general alarm status contacts; fire pump controller PLC IOs and relays | Read-only interface into certified life-safety equipment, never a write path |
| Factory Fire Watch | Sensing | Submersible tank level transmitters; CT energy meters; LoRaWAN field sensors with store-and-forward gateways | The four monitored point families: panel status, pump status, water level and energy |
| Factory Fire Watch | Compute | None purchased; CPU inside the operator's existing IoT platform; field gateways carry no model runtime | The requirement asks for no server or GPU, and compute that is not needed at the extremity is not deployed there |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Reasoning model | GLM 5.3, 753 billion parameters filed, FP8 (753 GB of weights), GLM-5.3 bespoke licence, served on vLLM on one node in the company's own data centre in Saudi Arabia | Named in the requirement; the licence allows commercial use, fine tuning and redistribution, and its model as a service trigger stays inapplicable to internal advisory use, so the company holds the weights; carries cause drafting, shift handover drafting, ontology maintenance and the command centre's cited answers |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Forecast model | Chronos-2, about 0.24 GB at 16 bit (0.48 GB at 32 bit), Apache-2.0, on the plant GPU server | Zero shot multivariate forecasting with covariates on the streaming tags; forecasts the stable band and its residuals are thresholded to raise the drift trigger; Apache-2.0 imposes nothing beyond attribution and no fine tuning is planned |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Embedding model | BGE-M3, about 1.1 GB at FP16 (2.27 GB at float32, 0.6 GB at 8 bit), MIT, on the plant GPU server | One pass retrieves dense, sparse and multi vector across Arabic and English shift logs and maintenance part numbers; embeds past upsets, shift logs and maintenance history for the reasoning tier; MIT licence held outright |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Frontier hardware | 141 GB HBM GPU class, eight accelerators per node, one node (1,128 GB) | Holds GLM 5.3's 753 GB of FP8 weights plus the 1.2 planning factor (904 GB) on one node with 375 GB beside the weights for KV cache and activations; held at one node, with two nodes of the 80 GB class as the fallback shape if the export licence position changes it |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Sizing rule | GLM 5.3 node count by GPU class | Sizes the frontier node from the filed parameter count before any hardware order |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Plant GPU server class | Plant GPU server, one or two PCIe cards, no high bandwidth memory node | Chronos-2 and BGE-M3 fit on a single card with room to spare, so no further frontier hardware is bought for them; the paper does not size this server |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Edge compute class | Industrial octa core SoC edge unit, 6 TOPS NPU, sealed fanless enclosure, served by ONNX Runtime | Holds roughly 33 C of margin at the 52 C record ambient before solar gain; restarts and resynchronises against the plant time source after a power loss |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Boundary equipment | Lidi with Eurydice one-way transfer stack (data diode class) | Reads leave the control zone for analysis and no return path exists to block; fails closed: the adapters buffer and the agents pause while the kiln's control path is untouched |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Serving runtime | vLLM for GLM 5.3 and ONNX Runtime at the edge | Serves the FP8 weights within the KV cache ceiling the sizing reserves; the edge unit serves its models under ONNX Runtime |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Registry | Harbor, air gapped offline artifact and model registry | Containers and models are staged |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Sensing | Streaming DCS tag stream of hundreds of variables, clinker quality data, fuel and alternative fuel feed records, the shell scan profile, shift log and production figures | Existing instrumentation only, read as it exists; no new field devices and no camera system |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Pattern: memory | Memory loops over confirmed outcomes, nothing trained into weights | The requirement forbids training on the fly; every confirmed cause, accepted recovery and shift outcome is kept and read before the next draft, and a moving model would invalidate the frozen baseline |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Pattern: proof | Controlled on and off comparison of thermal energy per tonne of clinker on the same line | The sector precedent read 4.1 percent in a continuous month and 2.2 percent under a controlled comparison, so the phase one gate reads the comparison only, never the headline month |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Pattern: approvals | Shadow mode before any draft is approved, the advisory layer treated as a process change under the plant's own hazard study methods | Drafts run first in shadow, every draft names its cause and expected effect, and the operator's confirmation or rejection is the recorded outcome the memory compounds on |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Pattern: boundary | The company's own hardware in its own plant and data centre in Saudi Arabia; one way inbound, read only, no write path into the control system | The national operational technology cybersecurity controls by facility criticality level and the company's own rule that nothing about the plants leaves the country |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Reasoning model | GLM 5.3, filed at 753 billion parameters, FP8 (753 GB of weights; about 756 GB as published), GLM-5.3 bespoke licence, served on vLLM on one node of eight 141 GB class GPUs on the operator's own dedicated in country hardware in Pakistan | The model the operator's own requirement names; frontier class at or above the 300 billion threshold; carries the restart, cycle damage and turnaround drafting agents and the cited conversational interface; the licence allows commercial use, fine tuning and redistribution and its managed access trigger does not bind internal use |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Forecast model | Chronos-2, Apache-2.0, FP32, about 0.48 GB at 32 bit as published and about 0.24 GB at 16 bit, served by ONNX Runtime on the site server | Zero shot multivariate forecasting with covariates over the historian tag mirror; thresholded residuals give the cycle damage agent its cycling stress signal with no fine tuning |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Embedding model | BGE-M3, MIT, run at FP16 (about 1.1 GB; about 2.27 GB at float32 and about 0.6 GB at 8 bit), served by ONNX Runtime on the site server | Dense, sparse and multi vector retrieval in one pass keeps tag numbers and error codes findable across work orders, inspection findings, corrosion readings and shift log entries, so every answer cites its record |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Vision detector models | Unnamed in the register: detection models for steam plume, flange leak, operator, maintenance scaffold and fire or smoke, on Frigate NVR and ONNX Runtime on the site server in phase one | Fixed cameras at the reformer top deck, the synthesis compressor train and the bagging plant lane watch fixed process equipment, so no tracking or trajectory duty exists and no camera group edge unit is needed in phase one |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Frontier hardware | 141 GB HBM GPU class, eight GPUs per node, one node (1,128 GB), new dedicated in country hardware the operator owns | Holds GLM 5.3's 753 GB of FP8 weights on one node; an ECCN 3A090 and 4A090 licence determination precedes any card being quoted, run as a phase one checkpoint, with an approved in country data centre held as the fallback placement |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Frontier footprint sizing rule | GLM 5.3 node count by GPU class, with the 1.2 planning factor | 753 GB of weights times 1.2 is 904 GB against the node's 1,128 GB, leaving 375 GB beside the weights of which about 151 GB is reserved for KV cache and activations; at BF16 the weights would be roughly 1.5 TB and would not fit |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Site inference server | An air cooled server inside the plant boundary holding Chronos-2, BGE-M3 and the vision models, class named and not sized in the paper | Forecasting belongs beside the historian it reads continuously, retrieval beside the records it embeds, and vision on the site server rather than camera group edge units in phase one |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | One way crossing | A hardware data diode carrying the lidi one way file service, with store and forward on the OT side | The requirement never writes to the control system; a diode is physically one way where a firewalled read only connection is only policy one way, and any change to the boundary passes management of change with the control engineer's signature |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Serving runtimes | vLLM for GLM 5.3 on the frontier node; ONNX Runtime for the site models | Serves the FP8 weights within the KV cache ceiling the sizing reserves; one runtime for every small model on the site server |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Event backbone | Apache Kafka 4.3.x with KRaft, replicated in country on the operator's own hardware | Ordering per partition key (tag id, work order number, restart identifier); delivery at least once with idempotent writes keyed by source timestamp and identifier; buffering on the OT side of the diode |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Time synchronisation | OCP Time Card grandmaster under linuxptp with a chrony NTP fallback | Historian, vibration and shift log records join on a common clock |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Platform services | K3s for container orchestration, Harbor as the air gapped artifact and model registry, Keycloak for identity on site, Prometheus, Grafana and Loki for observability | Artifacts arrive through the controlled import path and the registry's versions are the rollback; identity is held inside the boundary; delivery lag is watched so a stalled feed is seen before the next shift's refresh |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Ontology and graph store | Hyper Ontology v0.9 (Semantics plus Kinetics), fourteen objects refreshed shift by shift | The one place the plant's story is joined, anchored on the equipment master |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Time series store | Class only: the InfluxDB 3 Core, TimescaleDB and QuestDB family, product confirmed at design | The pick turns on the historian's volume and the operator's estate |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Sensing | Fixed cameras at the reformer top deck, the synthesis compressor train and the bagging plant lane; historian tag mirror, compressor vibration, route based corrosion readings, laboratory feed, gas restoration notices, digitised shift log; a citation check on conversational answers | Twelve sources once sensing is counted: eight systems of record named by the operator plus the camera lanes and the citation check; condition collection is route based, with continuous sensing on the critical tier held open |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Pattern: memory | Memory loops over confirmed restart causes and accepted inspections, nothing trained into weights | The requirement asks that the next restart plan improve without training a model on the fly, and the weights the operator holds stay the ones it can account for |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Pattern: shadow mode | Restart drafts run beside the standing written procedure until measured agreement | The superintendent confirms every step before any draft is trusted; a procedure nobody reads is no protection at all |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Pattern: ground | New dedicated in country hardware in Pakistan, open weight licences, no third party model service; report packs for the insurer's engineer leave only through a controlled export path after a disclosure review | Nothing about the plant leaves the country; the operator owns the object model, the weights, the decision record and the boundary |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Pattern: export licence gate | The ECCN 3A090 and 4A090 licence determination as a phase one checkpoint before any card is quoted | Export controls turn on total processing performance rather than memory; if the determination blocks the class, the fallback in country route is already held and the rest of the design is unchanged |
How much compute and hardware does AI in heavy industry and construction need?
The compute follows from the models: the published heavy industry and construction designs size it as follows, from no new hardware to a full GPU node.
| Design | Part | The design |
|---|---|---|
| Structure Phase 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) |
| Structure Phase Watch | Edge | Five Jetson Orin class nodes in solar powered enclosures at the gate, laydown, crane slew zone, batch plant and haul road |
| Steel Count Ledger | 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), in a dedicated data center behind an export licence checkpoint |
| Steel Count Ledger | Edge | One sealed GPU industrial PC and IP66 HDR cameras per installation point; counting survives a slow wide-area link |
| Steel Count Ledger | The hard dependency | 141 GB-class accelerators need a US export licence for Pakistan (Country Group D:4); the node is ordered only once a licence naming the end user holds |
| Factory Fire Watch | Compute | None bought: the requirement asks for no server or GPU, and the field gateways carry no model runtime |
| Factory Fire Watch | Equipment, per factory | About 2,600 to 3,200 US dollars in list prices: a LoRaWAN gateway, two tank transmitters, two CT meters, four contact nodes, a switch, an enclosure and a UPS |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Compute | One node of eight 141 GB HBM accelerators (1,128 GB) in the company's own data centre in Saudi Arabia, holding 753 GB of FP8 weights with a 1.2 planning factor, 904 GB; a plant GPU server of one or two PCIe cards for the forecast and embedding models; a sealed fanless industrial edge unit at the plant. The node is a 4A090 class item for a D:3 and D:4 destination, so an export licence determination comes before the order, with two 80 GB class nodes as the fallback shape |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Compute | One node of eight 141 GB HBM cards (1,128 GB) on new dedicated in country hardware the operator owns, holding 753 GB of FP8 weights with a 1.2 planning factor, 904 GB; an air cooled site server inside the plant for Chronos-2, BGE-M3 and vision; a hardware data diode on the OT boundary and an OCP Time Card grandmaster for time. The node is a 4A090 class item for a destination in Country Groups D:2, D:3 and D:4, so an ECCN 3A090 and 4A090 licence determination is a phase one checkpoint before any card is quoted, with an approved in country data centre held as the fallback |
Is it cheaper to own AI hardware or rent cloud GPUs in heavy industry and construction?
Each heavy industry and construction 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 |
|---|---|---|
| Structure Phase Watch | Three-year cost, owned | About 642,000 US dollars with support and power at the Saudi industrial tariff |
| Structure Phase Watch | Three-year cost, rented | 1.41 million to 2.81 million US dollars for the same GPUs around the clock; ownership is about one half the cheapest three-year commitment |
| Structure Phase Watch | Closed model break-even | The cheapest closed model matches the owned stack at about 31 users; above that, ownership is cheaper and the gap grows with every user |
| Steel Count Ledger | Three-year cost, owned | About 4,445,000 US dollars for 300 installation points, of which the counting kits are 2,832,000 and cannot be rented; owning the frontier node costs about 686,000 against 751,000 on AWS's deepest three-year plan |
| Steel Count Ledger | Closed model break-even | The cheapest closed model matches the owned node at about 33 users; above that, ownership is cheaper |
| Factory Fire Watch | Three-year cost, 100 factories | About 328,000 to 439,000 US dollars with support and power; the eight-factory pilot is about 21,000 to 26,000 of equipment |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Three-year cost | About 570,000 US dollars to own the frontier node for three years (a range of 542,000 to 594,000: server, support and power at the national commercial electricity tariff), against about 751,000 to rent the same eight cards on the deepest three year commitment in AWS Middle East (UAE), so ownership costs about three quarters of it, and about 2.00 million on demand, against which ownership costs a little over a quarter. Renting is a price reference only, because AWS lists no Saudi region on the day read and every rented option would carry plant data across the border (Appendix A) |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Three-year cost | About 572,000 US dollars to own the node for three years (a range of 542,000 to 610,000: server, support and power at the national industrial tariff), against about 751,000 to rent the same eight cards on the deepest three year commitment in AWS Asia Pacific (Mumbai), so ownership costs about three quarters of it, and about 2.00 million on demand, against which ownership costs a little over a quarter. Renting is a price reference only, because AWS lists no region in Pakistan and every rented option would carry plant data across the border (Appendix A) |
What ontology or object model does a heavy industry and construction 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 heavy industry and construction 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 |
|---|---|---|---|
| Structure Phase Watch | 15 objects, 12 links | Tower crane (TC-01 to TC-14), Crawler crane (6 units), Casting bed, Precast unit, Flatbed or mixer truck, Batch ticket, P6 activity (owner baseline or fragnet), Permit to work, Incident or observation form, Gate movement, Delivery window, Lift schedule entry, Pour, Worker, Work zone (Zone A to D) | objects.json |
| Steel Count Ledger | 14 objects, 12 links | Steel melting and re-rolling unit, Installation point, Industrial PC, Production count event, Product type (Billets, Ingots, Rebars, Girders), Declared production, Discrepancy case, Revenue field officer, Audit team, Steel mill staff, Tamper or offline alert, Daily uptime record, Product calibration record, Weighbridge record | objects.json |
| Factory Fire Watch | 14 objects, 12 links | High-risk factory, Fire alarm control panel, Fire pump, Fire water tank, Tank level transmitter, Energy meter (CT set), LoRaWAN gateway, Monitored point, Safety-critical alert, Audit log record, Periodic monitoring round, Preventive maintenance visit, Technical personnel, Monitoring officer | objects.json |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | 16 objects, 10 links | Kiln line, Equipment, Tag, Setpoint, Recipe, Raw meal lot, Fuel lot, Clinker sample and quality result, Upset, Operator action, Shell scan, Work order, Shift, Control room operator, Shift supervisor, Production manager | objects.json |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | 14 objects, 10 links | Process unit, Equipment, Historian tag, Trip, Restart, Thermal cycle, Work order, Inspection finding, Corrosion reading, Vibration reading, Lab result, Shift log entry, Turnaround task, Restart plan | objects.json |
Who approves the decisions an AI system makes in heavy industry and construction?
In every published heavy industry and construction design a named person makes the decision that changes the physical world; the system prepares it.
| Design | Human control |
|---|---|
| Structure Phase Watch | Every re-sequencing is a recommendation a named planner approves; every breach alert is confirmed, dismissed or escalated by the HSE officer |
| Steel Count Ledger | Every discrepancy case is judged by a named revenue field officer; the system counts and reconciles, it never assesses, and the mill never edits a count |
| Factory Fire Watch | A monitoring officer acknowledges or escalates every safety-critical alert; the design monitors and never controls a panel, pump or valve |
| A Self-Improving AI System for Cement Kiln Control in Saudi Arabia | Agents draft and people decide, by name: the control room operator confirms the upset agent's cause and recovery draft, the shift supervisor approves the setpoint agent's draft, and the production manager confirms the shift agent's handover and its energy and quality comparison; drafts run in shadow before anyone approves them, and every confirmation or rejection is kept as the memory the next shift reads |
| Industrial AI Reference Architecture for Fertilizer Plant Reliability in Pakistan | Agents draft and people decide, by name: the shift superintendent confirms every step of the restart plan, the reliability engineer approves the cycle damage draft and the inspections it brings forward, the turnaround manager approves the scope additions, and report packs for the insurer's engineer leave only through a controlled export path after a disclosure review; every status change carries a named person and a timestamp, and every confirmation is kept as memory the next restart reads, never as a weight update |
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.