Cold Room Compressor Health and Door Monitor for Small Cold Stores

The problem small cold stores have with compressors

Walk-in cold rooms hold potatoes and fruit in cold stores, milk and yoghurt in dairies, meat and vegetables in restaurants, and medicines in pharmacy warehouses. Owners usually learn something is wrong when the room is already warm: a compressor has failed, a coil has iced over, or a door was left open overnight. By then the stock may be lost and the repair is an emergency call-out at the worst time. Refrigeration technicians say failing compressors give warnings (longer run times, short cycling, higher starting current) days before they stop, but nobody is watching those signals in a small business.

What already exists and the gap

• Monnit sells wireless temperature, door and current sensors for commercial refrigeration with alerts for rising temperature, open doors and abnormal motor activity.
• Swift Sensors offers continuous refrigeration monitoring for walk-ins, freezers and cold storage rooms, with alerts for failing compressors and open doors.
• SensorPush is a low-cost temperature and humidity logger that many businesses place in fridges and walk-in coolers, with no equipment health signals.

These products are strong on temperature alarms. The gap is a low-cost retrofit that learns each room's own normal compressor behaviour, explains anomalies in terms a technician understands, and is priced for a small cold store or dairy in markets where these products are imported at high cost. Unlike a transport logger for vaccines, this is about the machine that makes the cold, not just the temperature of the goods.

What you build

• Room node: an ESP32 with temperature probes for room air and coil, a reed switch on the door, and a split-core current transformer on the compressor supply.
• Edge processing: cycle extraction on the device so only cycle summaries and events are sent.
• Backend: MQTT ingestion, storage, an anomaly service and alerting.
• Dashboard: per-room compliance, duty cycle, door events and anomaly explanations.

Users: the owner, who wants alerts and reports, and the technician, who wants cycle charts to diagnose faults.

The technically interesting part

Compressor behaviour depends on outdoor temperature, how much stock was loaded, door openings and defrost cycles, so a fixed threshold on run time produces constant false alarms in summer. The hard part is building a per-room baseline that accounts for these factors, for example a regression of run fraction on outdoor temperature and door-open minutes, and then flagging residuals that persist. Short cycling and starting current need their own detectors. Because real failures are rare during a project, you must induce faults safely on a test unit (partly block the condenser airflow, let the coil ice) to show that your detectors fire. Evaluators will want to see detection time, false alarm rate and explanations, not only a model score.

How to evaluate it

• Cycle extraction accuracy against a manually labelled current trace.
• Induced faults: for each fault type, time from fault start to alert, over repeated trials.
• False alarms per room per week on normal operation, including hot days and heavy loading days.
• Temperature measurement accuracy against a calibrated reference probe.
• A review session with a refrigeration technician rating whether each alert and its explanation made sense.

Suggested plan

1. Month 1: visit two cold rooms, talk to a technician, install a simple logger to start collecting data early.
2. Month 2: node hardware, cycle extraction on the device.
3. Month 3: backend, storage, dashboard basics and door reporting.
4. Month 4: baseline models and anomaly detectors on collected data.
5. Month 5: test fridge fault experiments.
6. Month 6 to 7: field evaluation, technician review and report.

Path to a product

The first customer is a dairy, restaurant group or potato cold store owner who has lost a load. Sell the node with installation and a monthly monitoring fee per room. Refrigeration service companies are the best channel: they can offer monitoring with their maintenance contracts and turn emergency calls into planned visits. Pharmacy distributors with cold rooms are a later market with stricter compliance needs.

Scope warning

Do not open or modify the refrigerant circuit; everything is clip-on or taped on. Do not build a full refrigeration controller that switches the compressor. Do not claim to predict failures you have never observed; say which faults you tested and how.

Learn what you need on Sefism

• Machine learning roadmap : regression baselines and anomaly detection on real sensor data.
• Data analytics roadmap : turning cycles and door events into reports an owner reads.
• System design cheat sheet : deciding what to compute on the device and what on the server.
• DuckDB Documentation : fast local analysis of weeks of cycle data while you develop models.
• PostgreSQL Documentation : storing cycles and events for the dashboard.

Get help with this idea

Getting real data early is the most important step, so start logging in the first month. If you want a second opinion on scope, hardware choices or how to set up the evaluation before you commit, bring your plan and your bill of materials and we will pressure-test it together. New here? Create a free account first.

• Discuss this idea with a mentor
• Send a query to Tauseef
• Book a 1:1 session