In-plant cold chain — continuous monitoring of cold rooms and freezers

In-plant cold chain vs in-transit: what a manufacturer actually needs. Direct reading from your PLCs. WhatsApp alerts before product is compromised.

Every bakery, every dairy, every meat plant depends on holding precise temperatures inside cold rooms, freezers, cooling tunnels and pasteurizers. If a chamber climbs 4°C for 18 minutes, the batch may stay safe or may be compromised — it depends on the product, the exposure time and the process step. What does not depend on those is whether you find out: if you discover it the next day, it is already too late. Rela AI continuously reads your existing PLCs and dataloggers, detects sub-second deviations and alerts you over WhatsApp before the product is compromised.

In-plant cold chain vs in-transit

A critical distinction the industry often blurs:

flowchart LR
  Farm[Farm or primary producer] --> Truck1[Truck in-transit]
  Truck1 --> Plant[Production plant]
  Plant --> Chamber[In-plant chambers and freezers]
  Chamber --> Truck2[Truck in-transit to distribution]
  Truck2 --> Store[Store or supermarket]

  Truck1 -.->|Sensitech Tive Controlant| In_Transit[In-transit cold chain]
  Truck2 -.->|Sensitech Tive Controlant| In_Transit
  Chamber -.->|Rela AI| In_Plant[In-plant cold chain]

• In-transit (Sensitech, Tive, Controlant, Roambee). Covers the truck, the marine container, the plane. Sensors with SIM card, GPS, battery. They report to a logistics cloud. Typical buyer: pharmaceutical distribution chain, meat exporter, global retailer.
• In-plant (Rela AI). Covers the production plant: proofing chambers, ripening rooms, freezers, cooling tunnels, pasteurizers. Reads the PLCs and dataloggers already installed in the equipment. Typical buyer: food SME from 5 to 100 million euros that wants BRCGS certification and to stop running HACCP on Excel.

A modern plant needs both — but the biggest hole today in food SMEs is in-plant, because in-transit vendors have not pushed down to that layer and the OEM apps from the PLC vendors do not talk to each other.

The problem: legacy alarms and zero digital audit

Typical case — Italian dairy with 8 million euros in revenue, mozzarella di bufala and stracciatella. Five finished-product freezers, two refrigerated rooms (milk reception and ripening), a post-pasteurizer cooling tunnel, an HTST pasteurizer.

How is it monitored today?

• Five Carel pCO5 freezers. Each with its own mobile app. Alarms beep on the maintenance manager's device, who silences them with a swipe without opening.
• Two Pego Plus refrigerated rooms. Pego says "you have to look at the local display". Whenever an operator walks past, they glance at the display. If they do not pass, they do not look.
• Schneider Modicon cooling tunnel. The integrator who installed it left a touch panel with a 30-day local history. After that it overwrites.
• HTST pasteurizer. It has its own control system on a Windows industrial PC nobody wants to touch. The HACCP sheet is signed by hand by the shift supervisor with a pen.

The result:

• Deviations are detected late or never detected.
• The BRCGS audit means printing dataloggers from each piece of equipment and pasting them into Excel. Three days of QA work each time.
• When a customer complains about defective mozzarella, there is no way to reconstruct what happened to the batch in the plant.

This is what Rela solves: a single monitoring layer that reads every piece of equipment, alerts in real time, logs every reading and exports audits in one click.

What is in-plant cold chain monitoring?

Continuous sub-second monitoring of the temperatures (and humidity, pressures, hold times) of the cold-side assets in a food plant. Three technical notes:

• Continuous, not sampled. The system reads every time the PLC emits (typically between 1 and 10 seconds depending on the equipment). Not 5-minute polling.
• Sub-second, from read to evaluation. Async pipeline that evaluates against bands, filters transients and, on a real deviation, fires an alert.
• With retroactive audit. Every reading is stored. To the question "what was chamber CRO-2 reading on March 15 at 14:30", the answer is a query.

How it works with Rela

flowchart LR
  Camera[Proofing chamber] -->|Modbus TCP| VPN
  Freezer[Carel freezer] -->|MQTT or OPC UA| VPN
  Pasteur[Schneider pasteurizer] -->|Modbus| VPN
  VPN --> Rela[Rela AI Plant OS]
  Rela --> Whatsapp[Pre-recall WhatsApp alert]
  Rela --> Audit[Exportable audit]
  Rela --> HACCP[Unified HACCP CCP]

The flow:

1. Each cold asset exposes its registers via the native protocol (Modbus TCP, OPC UA, MQTT, S7 for Siemens, EtherNet/IP for Allen-Bradley).
2. A single VPN tunnel from the plant Edge Gateway to Rela. One outbound connection covers every asset.
3. Rela AI Plant OS ingests readings, normalizes by asset, evaluates against configured bands, applies the grace period and filters transients.
4. On a real deviation, three actions fire in parallel: WhatsApp notification to the responsible person, a maintenance work order if the cause is likely the equipment, a HACCP deviation if the chamber is bound to a CCP.
5. Exportable audit all the time: QA can produce the BRCGS report with one click.

Typical equipment Rela monitors

For each one, Rela supports the native protocol and the manufacturer's standard registers. The integrator does not have to rewrite anything: once the first asset of a vendor is configured, the next ones take minutes.

Reading without installing new sensors

The key economic differentiator. Sensitech, Tive and Controlant sell their own sensors. Each sensor costs between 150 and 350 euros. A plant with 50 cold chain monitoring points means an upfront investment of 7,500 to 17,500 euros in hardware alone, plus installation, plus battery maintenance, plus platform.

Rela reads your existing PLCs and dataloggers via VPN. Zero new sensors.

Upfront comparison — dairy plant with 50 cold chain points:

And critically: when you add a new piece of cold equipment tomorrow, with Rela you simply configure the asset and the registers it exposes. With Sensitech you have to buy and deploy more sensors.

Recovery monitoring — the detail that matters

When a chamber opens its door to pull product, it climbs from 4°C to 8°C in seconds. When the door closes, it drops again. How long does the drop take?

• If it is under 90 seconds, the chamber is healthy.
• If it is between 90 seconds and 4 minutes, the chamber is starting to tire its compressor.
• If it is over 4 minutes, there is a real problem: low refrigerant, dirty condenser, failing fan.

Rela automatically measures recovery_time after each detected door-opening event (typically recognized by a PLC signal or by the temperature profile signature itself). It builds a per-asset histogram. When the 90th percentile drifts away from the learned baseline, it opens a maintenance work order before the chamber fails outright.

This is what software-only cannot do. You need sub-second PLC reads to reconstruct the recovery profile. A HACCP checklist app will never catch it.

HACCP integration

Every chamber or freezer reading is potentially a CCP in the HACCP plan. Rela shares the same asset_graph between cold chain and HACCP — they are not two separate systems duct-taped together.

When a cold chain deviation is also a HACCP violation, a single unified alert reaches the alert aggregator, not three copies.

A practical example:

• Asset: freezer_PT_03 (finished-product freezer).
• Bound HACCP CCP: maximum temperature minus 18°C, biological hazard, grace 600 s.
• Cold-chain detector: rolling 5-min average maximum minus 18°C, alert to operator.
• Predictive (maintenance) detector: drift in compressor power consumption, alert to maintenance.

If all three fire within the 15-min window: a single row in the unified inbox with source_systems = ["haccp", "cold_chain", "energy"], canonical severity critical (inherited from HACCP biological), and two linked work orders (HAC-xxx for QA, MAI-yyy for maintenance).

Audit and reports

Every config change, every deviation, every corrective action, every raw reading is logged with an immutable timestamp. For BRCGS, FSSC 22000 or FSMA audits, Rela exports a PDF dossier with:

• Cold-chain asset list with identifier, location, manufacturer, model, configured band, grace period.
• Reading histogram of the audited period per asset (chart + table).
• Deviation list with timestamp, value read, band violated, batch affected, corrective action taken, owner, electronic signature.
• Calibration list of physical probes with result and technician.
• Chain of custody for batches affected by deviations, with final disposition (hold, rework, destroy, release).
• Configuration audit trail — who changed which limit when and why.

Each PDF carries a timestamp, a cryptographic hash of entries and an electronic signature from the QA owner. Acceptable to BRCGS auditors and health authorities.

Edge Gateway — rural resilience

Internet connectivity in industrial zones is not always stable. A dairy plant in rural Italy or a meat plant in a Colombian free-trade zone faces fiber drops of 30 minutes to several hours, monthly.

Rela deploys a physical Edge Gateway (Raspberry Pi 4 with HiveMQ + Node-RED) that:

• Buffers locally every reading while connectivity is down.
• Keeps evaluating the configured bands of each asset on the device.
• Syncs to cloud when the link returns, preserving temporal order.
• Notifies the operator via physical LED or local screen when there is a deviation during the outage (WhatsApp can wait; the alert cannot).

Result: zero data loss, continuous monitoring even without Internet. Critical in industrial zones in Italy, Colombia, Mexico, Peru with unstable connectivity.

Use cases

• Cold chain in bakery
• Multi-machine single tunnel VPN
• Proofing chamber fleet with Modbus
• Compressor drift — early detection

Comparison: in-transit vs Rela in-plant

Bottom line: if you ship product from plant to distribution, you need a Sensitech (or similar). If you want to stop running HACCP on Excel inside your plant, you need Rela. Most food SMEs today have the in-transit half patched and the in-plant half empty. Rela closes that gap.

How to deploy cold chain in your plant

Four sequential steps:

1. Equipment and protocol inventory. A list of chambers, freezers, tunnels and pasteurizers with vendor, model and exposed protocol. One morning of work with your integrator.
2. Edge Gateway and VPN tunnel. Raspberry Pi 4 with WireGuard. One afternoon of installation.
3. Source and band configuration. Each asset is configured via the dashboard or over WhatsApp ("chamber CRO-2 between 2 and 4°C, grace 5 min"). For 30 pieces of equipment: one day.
4. Silent learning period (1 to 2 weeks). Rela monitors silently, catches drifting sensors, misconfigured transients, final tuning. Operational notifications then go live.

Total time to production: 3 to 4 weeks.

Key benefits

• Zero new sensors to buy — Rela reads the ones you already have.
• Sub-second detection — deviations caught before product is compromised.
• Native recovery monitoring — cold equipment failures detected before full breakdown.
• One-click BRCGS-exportable audit — from paper archive to signed PDF dossier.
• Edge Gateway with offline resilience — zero data loss in unstable-connectivity zones.
• Single shared asset graph across HACCP, predictive maintenance and OEE — end of data silos.
• Native WhatsApp in Spanish, Italian and English — no new app for the operator.
• Natural-language configuration — the plant manager describes the band and Rela applies it.
• Upfront investment 95 percent lower than Sensitech / Controlant for plant coverage.
• Time to production: 3 to 4 weeks from signature to live WhatsApp alert.