This is the engineering summary of an open reference architecture. The paper, its object model as JSON and the model register are free to reuse under CC BY 4.0: https://muhammadumar89.github.io/codeninja-research/factory-fire-monitoring-saudi-arabia/. The operator is described by class, never by name.
An operator of industrial cities in Saudi Arabia has a factory estate in the thousands, a hundred of them classed high risk. Fire alarm panels, fire pumps, fire water tanks and energy meters each hold a fragment of fire readiness, and none of it reaches the operator until a periodic round or an incident. The requirement is live, read-only visibility of all four across the highest-risk factories, published into the operator's own IoT platform, with no server or GPU asked for and no write path into certified life-safety equipment.
Here is how that turns into a system.
1. Read, never write
Panel general alarm status comes from dry contacts. Pump run, fault and fail-to-start come from controller PLC inputs and relays. Tank level comes from submersible transmitters, energy from CT meters. Every interface is a read into certified equipment; the design never issues a command to a panel, pump or valve. Monitoring beats control where life safety is.
2. Join first, model second
Fourteen typed objects: high-risk factory, fire alarm control panel, fire pump, fire water tank, tank level transmitter, energy meter, LoRaWAN gateway, monitored point, safety-critical alert, audit log record, periodic monitoring round, preventive maintenance visit, technical personnel and monitoring officer.
The monitored point is the focal object: it binds each reading to its source device, sampling cadence and MQTT topic. From one alert a traversal reaches the point that raised it, the device behind the point, the gateway that carried it, the factory, the officer who acknowledged it and the maintenance visit it triggered. The object model ships as JSON (ontology/objects.json, format hyper-ontology/1).
3. One central tier, arrived at by subtraction
| Tier | Runs | Why |
|---|---|---|
| Field | LoRaWAN gateways with store-and-forward buffers, no compute | The requirement asks for no edge server; keeping compute out of hazardous areas is satisfied by not putting it there |
| Platform | Threshold rules on the safety-critical path, plus one forecaster on CPU | The operator's existing IoT platform, hosted in Saudi Arabia, is the store and broker of record |
The one model is IBM's Granite Tiny Time Mixers (TTM-R2), about 0.85 million parameters and 0.003 GB at FP32, Apache-2.0. A mixer forecaster holds no key-value cache, so the usual memory arithmetic is one line.
4. Design for dust, sun and hose washing
Enclosures carry IP ratings per IEC 60529, NEMA types and SABER conformity, with UPS and clean shutdown. Gateway count and placement are justified by an RF survey, and every device's birth and death is logged, because a silent gateway is a safety gap.
5. Keep a person on every alert
A monitoring officer acknowledges or escalates every safety-critical alert, and the audit log records every state transition and configuration change. Rounds and maintenance visits stay human; the system makes them visible.
6. What it costs
Equipping a factory costs about 2,600 to 3,200 US dollars in list prices, the range being the gateway class. Across the hundred high-risk factories the estate is 262,000 to 321,000 dollars, and 328,000 to 439,000 with support and power over three years. There is nothing to rent: the devices sit at the factories in every option, and ingesting their messages on a hyperscaler would add at most about 1,200 dollars over three years. Every price is cited in the paper's Appendix A.
Full design, figures and the object model: the paper.
Designed on Praxis, CodeNinja's platform for designing physical AI systems. The object model imports into Hyper Ontology, which turns it into a living system. Load it yourself with the open hyper-ontology loader.
Top comments (0)