Simulation & Modeling
Digital Twin Engine
A first-principles simulation framework that composes individual physics models into process-level digital twins. Wire equipment models into a directed graph, propagate state in real time, and run what-if scenarios — all on the edge.
How the Engine Works
Four stages from raw sensor data to system-level simulation.
Sensor Ingestion
Raw telemetry from Modbus, OPC UA, BACnet, and MQTT is normalized into a unified tag namespace. Each sensor maps to a model input port.
Model Composition
Individual physics models are wired into a directed acyclic graph (DAG). Outputs of one model become inputs to the next — just like real process flow.
State Propagation
On every tick, sensor values flow through the graph. Each model evaluates its equations and passes calculated states downstream. Sub-100ms end-to-end.
Scenario Execution
Fork the live graph state, inject a perturbation (pump trip, setpoint change, load spike), and propagate forward to see system-wide impact — without touching real equipment.
Built-In Model Library
Nine first-principles equipment models ready to compose into any process.
| Equipment Model | Physics Domain | Inputs | Outputs | Edge Latency |
|---|---|---|---|---|
| Centrifugal Pump | Hydraulics | P_in, P_out, Flow, Power, Speed | η, BEP%, NPSH_m, SE | <8ms |
| Positive Displacement Blower | Pneumatics | P_in, P_out, Airflow, Power | η, Surge_margin, T_out | <6ms |
| Centrifugal Chiller | Thermodynamics | T_chws, T_chwr, T_cws, Power | COP, kW/ton, Capacity% | <10ms |
| Diesel Generator | Electromechanical | Fuel_rate, Power_out, Freq, T_exhaust | η_thermal, kWh/gal, Load% | <5ms |
| Battery Bank | Electrochemical | V_bus, I_charge, I_discharge, T_cell | SOC, SOH, C-rate, Capacity | <4ms |
| Cooling Tower | Psychrometrics | T_hw, T_cw, T_wb, Airflow | Approach, Range, η_evap | <7ms |
| Gravity Clarifier | Sedimentation | Flow, TSS_in, Blanket_depth | SOR, TSS_out, Sludge_rate | <5ms |
| Media Filter | Filtration | Flow, dP, TSS_in, Run_time | TSS_out, Backwash_ETA, Capacity% | <4ms |
| UV Disinfection | Photochemistry | Flow, UVT, Lamp_hrs, Power | Dose_mJ, Log_inact, Lamp_life% | <3ms |
Engine Capabilities
The simulation framework that powers every TwinEdge digital twin.
Process-Level Composition
Wire individual equipment models into full process trains. A pump connects to a pipe connects to a clarifier — just like the real system. Change one and see the cascade.
What-If Scenario API
Fork the live process state, inject a perturbation, and simulate forward. Test pump trips, setpoint changes, storm events, and equipment failures without touching real equipment.
Directed Acyclic Graph (DAG)
Models are nodes. Data flows are edges. The engine resolves execution order, handles parallel branches, and propagates state changes in topological order.
Sub-100ms Propagation
Full graph evaluation — sensor ingestion through model composition to output — completes in under 100ms on edge hardware. Fast enough for closed-loop control.
Edge + Cloud Split Execution
Time-critical models run on the edge for real-time response. Fleet aggregation, historical trending, and model retraining run in the cloud. Each layer does what it does best.
Model Lifecycle Management
Create models from the built-in library or custom equations. Validate against historical data. Deploy to edge. Monitor drift. Retrain when conditions change.
Physics Analytics vs Digital Twin Engine
Physics Analytics monitors individual equipment. The Digital Twin Engine simulates entire systems.
| Physics Analytics | Digital Twin Engine | |
|---|---|---|
| Scope | Single equipment asset | Entire process train or facility |
| Example | Pump curve + efficiency at one operating point | What happens to the whole plant when that pump trips |
| Output | Real-time KPIs (η, NPSH, BEP%) | System state propagation + scenario comparison |
| Input | 3-5 sensors per asset | All sensors across the process, composed through a model graph |
| Use case | Monitor and trend individual equipment health | Simulate operational decisions before executing them |
Technical Specifications
Performance characteristics on TwinEdge edge hardware (ARM64, 4GB RAM).
Graph evaluation latency
Full DAG propagation on edge hardware
Max models per edge node
Depends on model complexity and hardware
Scenario throughput
Parallel what-if evaluations
Supported model types
Plus custom equation builder
State snapshot interval
Configurable per process
Scenario horizon
Forward simulation window
Graph serialization
Version-controlled, portable
Model hot-swap
Replace models without restarting the graph
Simulate Before You Operate
Test operational decisions on the digital twin before executing them on real equipment. See the system-wide impact of every change — in under 100 milliseconds.