System Architecture

The Industrial PMR architecture is engineered for low latency, high resilience, and industrial-grade reliability in demanding factory environments.

End-to-End Pipeline Overview

The physical sensor interface is handled by two specialized edge hardware models:

ipmr223m-v1 (Legacy Production Node):
- Earlier, simpler controller focused on core cycle and stroke monitoring.
- Connects to machine limit switches, proximity sensors, or optical photo-eyes via optoisolated digital inputs.
- Monitors operating current via non-invasive split-core Current Transducers (CTs) to identify running vs. idling states.
- Features an onboard battery-backed Real-Time Clock (RTC) for microsecond timestamp precision and SPI flash ring-buffering.
ipmr226 (Flagship Integrated Power Analyzer & Production Node):
- New, refined controller combining cycle counting with a high-precision polyphase power analyzer.
- Direct 4-wire potential measurement (Phase $R, Y, B$ + Neutral) for true RMS line-to-neutral and line-to-line voltages.
- Computes active power (kW), reactive power (kVAR), power factor (PF), frequency (Hz), phase balance, and cumulative kWh.
Local SPI Flash Ring-Buffer:
- Automatically activates on both ipmr223m-v1 and ipmr226 whenever factory Wi-Fi or broker connectivity drops.
- Telemetry records are written to non-volatile flash memory and chronologically replayed with original RTC timestamps once the network is restored.

Telemetry transmission is optimized for minimal network overhead and high reliability:

MQTT Protocol: Operates over TLS on port 8883 (mqtts://mqtt.industrialpmr.com) using QoS 1/2 for reliable delivery.
Shard-Aware Topics:
- ipmr/{shard}/{mac}/telemetry: High-frequency sensor streams from edge units.
- ipmr/{shard}/{mac}/downlink: Client-to-node control (keep-alive pings, configuration adjustments, sampling rate commands).
- unit_to_server/#: Backward-compatible protocol pipeline for legacy node firmware.
On-Demand Downlink Keep-Alive: To protect factory cellular or Wi-Fi bandwidth when hundreds of nodes are deployed, edge nodes default to a low-frequency heartbeat. When an operator opens a machine card on the web dashboard, the dashboard issues a KEEP_ALIVE downlink pulse. The node then escalates to 1 Hz live streaming for as long as the supervisor is viewing that machine.

The cloud backend processes millions of incoming data points using a distributed, event-driven pipeline:

MQTT Middleware Dispatcher:
- High-throughput Node.js microservice running in container clusters.
- Performs validation, timestamp normalization, and dual-dispatch to InfluxDB and Google Cloud Pub/Sub.
Google Cloud Pub/Sub:
- Shards incoming records into dedicated topics: ipmr-v2-records-pipeline (cycle pulses and machine states) and ipmr-v2-electricity-pipeline (3-phase energy parameters).
Time-Series Storage (InfluxDB):
- High-density retention for second-by-second voltage waveforms, current peaks, and power factor histories.
Firebase Realtime Database (RTDB):
- Serves as the real-time operational state cache, powering instantaneous multi-user web dashboard synchronization.

Architecture: Built on modern React and TypeScript using the Chakra UI design system with full support for light and dark factory floor modes.
Reactive State Management: Implements Jotai atoms with custom reactive hooks (useDeviceTelemetry, useUnitMac) that maintain WebSocket links with automatic unmount garbage collection.
Continuous Operation: Designed for 24/7 unassisted operation on plant wallboards, with automatic memory cleanup and reconnection resilience.