Plant Library
How RadMah AI produces physically-valid SCADA telemetry across a broad range of industries: layered fidelity tiers, a deep library of pre-built plant templates for well-studied verticals, and a composition path for everything else.
Automatic fidelity selection
The runtime picks the highest-fidelity path it has evidence for. Customers never pick the tier — the platform chooses based on the natural-language description, matched template, and available coverage.
| Path | Data grade | What it does |
|---|---|---|
| Pre-built plant template | Training-grade | Full mechanistic physics for well-studied verticals. Purpose-built templates with calibrated dynamics, realistic disturbances, protocol bindings, and cyber-effect overlays. |
| Composed plant | Composition-grade | The platform composes a plant from a validated library of unit-operation building blocks (control, electrical, fluid, material) connected in a directed graph. Every run is validated against the same quality gate. |
| Statistical fallback | Augmentation-grade | Used for protocol-centric signals where physics is a nuisance parameter, so the requested operating profile governs mean, variance, and temporal structure. |
Pre-built plant templates
Each pre-built template ships a full domain model: named signals with physical bounds, realistic dynamics, an observation layer that emits true/measured/reported triples, domain invariants, and a parameter schema the platform fills in from the customer’s description.
| Vertical | Domain coverage |
|---|---|
| Water / wastewater | Treatment process dynamics, pH chemistry, biological kinetics, realistic diurnal demand patterns. |
| Power & substation | Transformer thermal behaviour, generator inertia, battery state-of-charge, tap-changer action, protection coordination. |
| Oil & gas midstream | Pipeline pressure behaviour, compressor stability, separator operation, multi-component composition tracking. |
| Chemical process | Reactor dynamics with temperature-dependent kinetics, mass and energy balance, jacket cooling, conversion and selectivity tracking. |
| Buildings / HVAC | Zone thermal comfort, chiller and boiler part-load behaviour, damper dynamics, occupancy-driven setpoint shaping. |
| Manufacturing | Discrete-event behaviour, cycle-time tracking, conveyor dynamics, buffer management, operational KPIs. |
| Mining | Ventilation pressure, flotation cell dynamics, thickener level, ore throughput. |
| Food processing | Thermal pasteurisation, fermentation dynamics, batch cooking, cold-chain tracking. |
| Logistics | Conveyor networks, buffer occupancy, discrete item tracking. |
| Transport (rail, port, maritime) | Hoist kinematics, belt mechanics, rail block occupancy, signalling state, marine propulsion behaviour. |
| Hydro & dams | Reservoir dynamics, spillway behaviour, penstock valve control, flood-control timing. |
Industries outside this list are supported through the composition path — the platform generates a plant structure from a library of validated building blocks rather than picking a ready-made template. Either path passes the same validation gate.
Composition library
A library of validated unit-operation building blocks across four families, each with its own internal dynamics and integration policy. Blocks are connected through typed ports; the platform builds the graph and dispatches the integration in the correct order.
- Control: PID, setpoint trackers, alarm panels, interlocks, sequencers.
- Electrical: breakers, relays, buses, transformers, loads, metering panels.
- Fluid: pumps, valves, tanks, pipes, flow meters, pressure transmitters.
- Material: conveyors, buffers, mixers, reactors, hoists.
Determinism & the seal
Every run is reproducible bit-for-bit given the tuple (description, seed, parameter set, engine version). The dispatch path, plant fabrication output, and simulation state all feed into the evidence seal — two runs with the same seal hash produce identical artifacts.