1.1 System Architecture

The system architecture defines the deployment boundaries and trust zones that separate field devices, communications infrastructure, platform services, and external integrations. Understanding these boundaries is essential for assigning responsibility, designing security controls, and planning maintenance access. The architecture supports both cloud-hosted and local data center deployments, with the edge layer providing autonomous operation during connectivity loss.

The deployment boundary diagram below illustrates the four primary zones: the Field/Greenhouse Site Zone containing all physical sensing and edge control equipment; the Communication Bearer Zone providing protocol translation and backhaul; the Platform Zone hosting data governance, analytics, and alarm services; and the External Systems Zone for third-party integrations including farm management, messaging, and weather services.

System Deployment Boundary Diagram

Figure 1.1: System Deployment Boundary Diagram — Four-Zone Architecture with Trust Boundaries

Module Relations & Data Flows

Sensor nodes publish measurements to edge devices via LoRa radio or RS-485 wired bus. Edge gateways normalize incoming data to MQTT or HTTPS and forward it to the platform. Control commands flow from the platform to the edge controller (or local logic) and then to actuators such as valves, fans, and screens. Data governance runs at both the edge (fast range and rate-of-change checks) and the platform (deep checks using historical context and cross-sensor comparisons).

Core vs. Optional vs. Supporting Components

CategoryComponentsJustification
CoreSensors, edge node/RTU, gateway/controller, data ingestion, time-series storage, QC flags, alarms, dashboardRequired for any functional deployment; without these, the system cannot deliver its primary value
OptionalML pest models, video AI analytics, satellite connectivity, advanced optimization, digital twinAdds value for specific use cases; can be added after core is stable and proven
SupportingPower distribution, UPS for cabinets, lightning/grounding, cabinet/IT room, fire linkage, physical securityInfrastructure dependencies; failure here causes core system failure even if core components are correct

1.2 Components and Functions

The component inventory spans six categories from physical sensors to platform software. Each component has defined responsibilities, inputs, outputs, key performance indicators, and typical mismatch risks. The hierarchical component diagram below provides a visual overview of all components and their relationships within the system architecture.

System Components Hierarchical Tree Diagram

Figure 1.2: System Components Hierarchical Tree — All Components with Typical Quantities and Key KPIs

ComponentResponsibilityKey KPIs (Engineering)Typical Mismatch Risk
Weather Station (T/RH/Wind/Rain/Pressure)On-site microclimate baseline measurementTemp ±0.3°C; RH ±2%RH; Wind ±0.3 m/s; Rain ±3%Wrong siting causes biased wind/rain; poor grounding causes surge damage
Soil Moisture Sensor (VWC)Irrigation decision input; root zone monitoringVWC ±2–3% (after calibration); soil-type dependentAir gaps during installation → unstable readings; wrong depth → non-representative
CO₂ SensorGreenhouse control; photosynthesis optimization±(50 ppm + 3%); NDIR preferredCondensation damage; wrong placement near vents or doors
PPFD/PAR SensorLight management; DLI calculation±5%; cosine-corrected domeShading or angle error; dirty dome causes under-reading
Leaf Wetness SensorDisease risk model inputRepeatability ±5%; mounting orientation criticalWrong mounting angle → false wetness readings
Water Level/Flow MeterIrrigation verification; tank managementFlow ±1–2%; ultrasonic or electromagneticAir in pipe; insufficient straight-run upstream
Edge Node (Low Power)Sampling, buffering, local QCUptime ≥99%; sleep current <100 µABattery undersized; enclosure IP insufficient for environment
Edge GatewayProtocol translation, backhaul, store-and-forwardBacklog ≥7 days; dual uplink supportCellular instability; no antenna lightning protection
Edge Controller (PLC/RTU)Closed-loop control; safe fallbackResponse <3–10 s; fail-safe modeNo interlock → unsafe actuator actions on sensor loss
Platform IngestionAccepts and normalizes telemetryLoss <0.1%; idempotency; schema validationDuplicate processing; schema drift causes silent data corruption
Data Governance EngineQC flags, cleansing, calibration trackingFlag accuracy; auditability; traceability"Silent corruption" if not implemented — data looks fine but decisions are wrong
Alarm EngineActionable alarms with dedup and escalationDedup rate; escalation success; MTTA <10 min criticalAlarm fatigue from poor rules → critical alarms ignored
Dashboard/BIVisibility for operators and managersRole-based views; data freshness ≤2× intervalWrong aggregation or stale data leads to wrong decisions
API GatewayIntegration with external systemsAuth, rate limiting, SLA ≥99.5%Insecure exposure; no rate limiting → DoS risk

1.3 Working Principles

Startup Sequence

When the system powers on, sensors and edge nodes perform self-tests including battery voltage check, sensor presence detection, and memory integrity verification. The gateway provides time synchronization via NTP if online, or uses last-known time with drift correction if offline. Each device pulls its configuration — sampling interval, alarm thresholds, and calibration coefficients — from the platform or local cache. The first data burst after power-on is marked as "warm-up" for sensors requiring stabilization, such as electrochemical gas sensors which typically need 5–30 minutes to reach stable readings.

Normal Operation Cycle

During normal operation, sensors sample at configured intervals. Edge nodes apply fast range checks and rate-of-change (ROC) checks, tag each measurement with a QC flag (valid/suspect/invalid/missing), buffer the data with sequence numbers, and transmit to the gateway. The platform performs deeper QC using historical context and cross-sensor comparisons, stores the time series with QC tags, evaluates alarm rules, writes work orders for triggered alarms, and sends recommendations to operators. Control commands flow from the platform to the edge controller, which applies safety constraints before executing actuator outputs.

Exception Handling & Recovery

Offline Mode: When backhaul is lost, the edge gateway switches to store-and-forward mode, buffering data locally. Edge controllers use last-safe setpoints and local constraint logic to maintain safe operation without cloud connectivity. Upon backhaul recovery, the gateway replays buffered data with sequence numbers; the platform de-duplicates using idempotency keys.

Exception Chains (Failure → System Behavior)

Failure ScenarioFailure MechanismSystem BehaviorRecovery Path
Sensor Drift Chain (Soil EC) Probe fouling/salt deposition → EC reading drifts upward → fertigation reduced incorrectly → crop stress Platform drift detector flags "suspect"; triggers maintenance ticket; fertigation logic switches to conservative mode requiring manual confirmation Dispatch calibration task; recalibrate or replace probe; backfill corrected data with versioning
Communication Outage Chain (Cellular) Tower outage → no uplink → dashboards blank → operator overreacts Edge continues local logging and critical control; platform shows "site offline" with last-known values; local alarms via on-site siren for extreme thresholds Backhaul returns; gateway replays buffered data; platform de-duplicates and fills timeline gaps
Lightning Surge Chain (Gateway/RS-485) Near strike induces surge on long cables → transceiver damage → multiple sensors silent Gateway detects bus failure, isolates segment via fused RS-485 protectors, raises "segment offline" alarm Spare gateway procedure; replace SPD if triggered; verify grounding; restore within SLA
Sensor Stuck-at Value Sensor output constant for too long → QC ROC check triggers Flag suspect; exclude from control decisions; create maintenance ticket automatically Physical inspection; clean or replace sensor; verify readings against reference

Critical Design Principle: Every exception chain must have a defined system behavior that maintains safety (no unsafe actuator actions) and data integrity (no silent corruption). Design the exception handling before deployment, not after the first incident.


← Homepage Overview Chapter 2: Design Methods →