Heat pumps · Chillers · Sludge drying systems
Industrial cooling · Process

Chiller Remote Monitoring: Alarms and O&M Closure

Many chiller remote-monitoring platforms are technically working and operationally dead. The dashboard looks polished, the data updates, but nobody uses it. Six months later, faults are still reported by phone and the platform has become an electronic wall display.

The usual cause is design order: the screens were designed before anyone defined where the data comes from, who receives an alarm, and who closes the work order. A useful platform is designed backward from a single question: how many steps does it take to move a fault from occurrence to verified closure?

1. The data chain has four links

Field layer. Chillers, pumps and cooling towers expose data through Modbus RTU, Modbus TCP, BACnet or a proprietary protocol. The common problem is a closed interface or an incomplete register map. The supply contract should require a complete communication map and protocol document.

Gateway layer. The gateway converts protocols, buffers data and replays it after a network outage. Three selection criteria matter: supported protocols, local storage depth and remote configuration. Specify at least 7 days of buffering so a one-week outage does not create a data gap.

Transport layer. Use wired Ethernet where available; it is stable and avoids data charges. For dispersed sites, use 4G or 5G and check coverage and tariff. Encrypt the link; do not expose a plain Modbus port to the internet.

Platform layer. Storage, rules, alarms, reports and interfaces. Check multi-tenant support and whether the platform can expose data to the owner’s existing systems.

Layer Equipment Key specification Common failure
Field Chiller controller Open protocol, complete map Proprietary protocol or missing addresses
Gateway Edge gateway 3+ protocols, 7-day buffer Data loss during outage
Transport Wired or 4G Encrypted, stable Plain port exposed
Platform Cloud server Multi-tenant, open API Cannot interface to owner systems

2. Sampling frequency: more is not better

Chiller variables change over minutes, not milliseconds. One-minute sampling adds storage and cost with little diagnostic value.

Take 100 chillers, 30 points each, sampled every 5 minutes:

Daily records = 100 × 30 × 288 ≈ 864,000. At 50 bytes per record, that is about 43 MB/day, or 15.7 GB/year. At one-minute sampling, the annual volume rises to about 78 GB, roughly five times the storage.

Interval Annual data Best use
1 min 78 GB Short-term diagnostics
5 min 15.7 GB Routine monitoring
15 min 5.2 GB Energy reporting only

Use tiered sampling: routine analog values every 5 minutes; digital status and alarms on change; temporary one-minute sampling on a single chiller during diagnosis. Change-of-state reporting is often overlooked. Polling a digital point every 5 minutes can miss a short fault between polls; reporting on change captures every transition and usually produces less data, not more.

3. Alarms: classify, debounce, suppress, escalate

Three levels. Emergency (shutdown, over-temperature, loss of flow), important (efficiency 15% below baseline, more than 20 starts/day) and advisory (parameter drift, filter differential pressure). Only emergency alarms should trigger SMS and phone calls; important alarms push to the app; advisory alarms go into the daily report.

Debounce. A value near a threshold can trigger repeatedly. Require three consecutive samples outside the limit before alarming. For pressure and temperature, a 15-minute window is often appropriate.

Suppression. One root cause can generate a cascade: the condenser pump stops, flow is lost, the chiller trips, and supply temperature rises. Those are one event, not three. Define parent-child relationships so child alarms are logged but not pushed while the parent is active.

Escalation. An emergency alarm unacknowledged for 15 minutes should escalate to the maintenance supervisor; 30 minutes to the project manager. The escalation path belongs in the platform, not in someone’s memory.

Level Example Notification Response
Emergency Chiller trip, supply 3 K over limit, no flow SMS plus phone Within 15 min
Important Efficiency 15% below baseline, 20 starts/day App push Within 4 h
Advisory Parameter drift, high filter ΔP Daily report Next inspection

4. Close the loop from alarm to work order

Automatic dispatch. An emergency alarm creates a work order for the responsible engineer, with unit data, alarm history and the last three months of trends. The engineer can decide which spare parts to bring before traveling, which reduces repeat visits.

Structured records. The work order must capture the action, parts replaced and before/after readings. A field that says only “handled” is useless three months later.

Verification. After the work is closed, the platform compares 24 hours of key readings before and after the event. The work order should not be closed until the alarm has not recurred. A manual “complete” button hides unresolved problems.

Knowledge base. Classify recurring alarms by chiller model and symptom. New engineers can then find the previous resolution instead of starting from zero.

Step Platform function Requirement
Alarm Rules, debounce, suppression False-alarm rate below 10%
Dispatch Site and skill matching Work order within 15 min
Repair Mobile entry, parts record Mandatory fields
Verify Before/after comparison Automatic recurrence check
Archive Searchable fault library By model and symptom

5. Build or buy

A self-built platform typically costs 300,000–800,000 CNY initially, gives full data ownership and customization, but requires an internal information team. A SaaS platform charges by point or chiller per year, is faster to deploy, but has limited customization and requires a clear data-ownership clause. The usual break point is around 500 chillers. Below that, SaaS is often more practical; above it, self-build may be justified.

Write three data clauses into the contract: who owns the data, whether the vendor may use it for other purposes, and how it can be exported on exit. Without those, migration becomes expensive.

6. Three common post-go-live problems

Frequent offline. Check 4G signal strength, gateway supply and heartbeat timeout. A gateway sharing a supply with chillers can reboot when a compressor starts.

Data drift. A 2–3 K difference from a calibrated thermometer usually means a sensor is surface-mounted rather than in a thermowell. Fit a thermowell and heat-transfer paste.

Implausible efficiency. A COP above 6 usually means flow is a design value or the temperature difference is measured across the chiller rather than the load. Check the calculation basis before blaming the chiller.

6. Four reports that make the platform useful

Daily exception report. Only exceptions: alarms, units that did not run, efficiency outside the band, and sites offline for more than 30 minutes. A daily report that lists every value is ignored; one that lists only what changed gets read.

Monthly efficiency report. Plant kW per metric ton of cooling, chiller kW per metric ton, pump kW per metric ton and tower kW per metric ton, compared with the previous month and the same month last year. This separates a real efficiency loss from a change in weather or load.

Alarm-response report. Count alarms by level, median time to acknowledge, median time to close, and the percentage closed within the target. This is the report that shows whether the platform is actually being used.

Work-order recurrence report. Group work orders by chiller, fault code and component. A condenser-water pump that trips every three weeks is a design or maintenance issue, not an operator error. Recurrence is the most reliable input for the next maintenance budget.

Common questions

How do we prevent a platform from becoming unused?

Design from alarm closure, not from the dashboard. Define who receives each alarm, how quickly they must respond, and who verifies the repair. A platform that sends emergency alarms to someone who does not check their phone will fail regardless of its graphics.

What sampling interval is appropriate?

Five minutes for routine analog values, change-of-state for digital points, and one-minute temporary sampling during diagnosis. The storage difference between one and five minutes is about fivefold, while the diagnostic benefit is limited outside a fault investigation.

Is 4G data traffic expensive?

For 100 chillers at 5-minute sampling, the annual volume is about 15.7 GB, or roughly 20 GB encrypted. The traffic cost is minor compared with gateways and platform licensing.

How do we reduce alarm floods?

Use three levels, debounce over three samples, suppress child alarms under a parent, and escalate unacknowledged emergencies. Done properly, false alarms usually fall below 10%.

Can old chillers without communications be connected?

Yes, with a gateway and hardwired sensors: power meters, temperature sensors and auxiliary contacts. Accuracy is lower than native communication but sufficient for monitoring and energy reporting. Check for a reserved port before adding external hardware.

Is vendor-hosted data secure?

Check encryption in transit, storage isolation and full export on exit. For sensitive sites, keep a local copy of critical data and use the cloud for presentation and analysis.

What alarm levels?

Three: advisory, warning and critical. Each needs an owner and a response time. A dashboard without closure is decoration.

How much data buffer?

At least 7 days at the gateway so a network outage does not erase the trend. Five-minute data is enough for most chiller diagnostics.

Related reading

LONGCOM can design the monitoring point list, gateway specification, alarm rules and O&M workflow for a chiller plant. For a site review, contact 400-878-0280 or 13002001280.

Related reading

Chiller plant sensors · Chiller plant to BAS · Chiller plant group control