A standby generator spends most of its life doing nothing, and the failures that matter happen between visits: a charger that quietly stops, a battery that cranks more slowly each month, a set that starts but never takes load. Telematics is worth paying for only when it answers four questions faster than a site visit can: will the set start, did it start, is it carrying load, and can it keep carrying it. This page sets out which measurements earn their wiring, how to set alarm limits that people act on, how much data to keep, and how to prove the monitoring system itself at commissioning.
Start with four questions, not a dashboard
A monitoring portal can show forty parameters and still leave the duty engineer unable to answer the only question that matters at 02:00: if the mains fails now, will the building stay up? Avoid that by designing the system around decisions, not around what the controller happens to publish.
The first question is availability before the event. A set is available when the battery is charged, fuel is present, coolant and oil are in range, the block heater has done its job and the controller has no latched fault. Most of that can be measured at rest, and most failures announce themselves as a trend rather than a single out-of-range value.
The second question is starting. A start command, a crank attempt and a rise in oil pressure and frequency are separate events, and the sequence between them is the diagnostic. A set that cranks for eight seconds and catches is healthy; a set that cranks for eight seconds, pauses and tries again is telling you the battery, the fuel system or the ambient is changing.
The third question is whether the set is carrying load, not merely running. Run status comes from the controller; load comes from current transformers on the alternator output and the changeover position. A set can run for an hour in a test with the ATS still on mains.
The fourth question is whether it can keep carrying load. That is a thermal and fuel question, answered by coolant temperature, oil pressure, exhaust temperature, room temperature and fuel consumption over time. It is also the question that a one-minute snapshot cannot answer, which is why trends and event logs matter more than a live gauge.
| Decision | What answers it | What the answer changes |
|---|---|---|
| Will the set start if the mains fails? | Battery voltage and charger current, fuel level, coolant and oil condition, block heater, latched faults | Whether to send a technician before the next outage, or to change a charger or battery now |
| Did it start and transfer? | Start command, crank attempts, oil pressure rise, frequency, ATS position and transfer time | Whether the failure is in the set, the starting system or the changeover |
| Is it carrying the load? | Three-phase current, kW, kVA, power factor, ATS position, breaker status | Whether a test was real, whether the set is lightly loaded, whether one phase is overloaded |
| Can it keep carrying it? | Coolant and oil temperature, oil pressure, exhaust temperature, room temperature, fuel consumption | Whether to derate, improve ventilation, clean the cooling system or add fuel |
The measurements that earn their wiring
Not every point on a controller is worth a monthly data plan. The useful set is smaller than most quotations suggest: electrical output, engine condition, starting system, changeover, and the room the set lives in.
On the electrical side, three-phase current is the measurement a single total kW reading hides. An unbalanced load pushes current into the neutral and heats one winding harder than the others, so a set that looks comfortable at 60 % total load can still have a phase at its limit. Voltage, frequency, kW, kVA and power factor complete the picture; a set rated at 0.8 power factor can run out of kW before it runs out of current.
On the engine side, oil pressure and coolant temperature are the two signals that turn a run into evidence. A switch that closes at a threshold tells you that the engine did not fail; an analogue signal tells you that the pressure is drifting down over six months. Oil temperature is optional on many sets but useful on a heavily loaded unit in a hot room.
The starting system deserves more attention than it gets. Battery voltage at rest is a poor indicator, because a battery can sit at a healthy float voltage and still collapse under crank. The valuable record is voltage during the crank, with charger current and crank attempts. A failed charger shows as a slow decline over days, the kind of trend a monthly visit misses and telematics catches.
The changeover is part of the generator system even when it is a separate panel. Mains present, generator available, ATS position and transfer time belong in the same record as the set’s status. The UPS covers the second-level gap between mains loss and the generator carrying load, so transfer time proves the UPS had enough ride-through; if the ATS takes longer than the UPS can bridge, the load drops even though the generator started.
The room is the last group and the one most often omitted. A set rated at a 40 °C reference ambient has little margin above it; a blocked louvre or failed extract fan appears first as rising room temperature, then as rising coolant temperature, then as a derated or tripped set. A room sensor and a fan status contact are cheap compared with the failure they prevent.
| Measurement | Why it matters | Sampling note |
|---|---|---|
| Three-phase current and neutral current | Shows real load, phase imbalance and overload that a total kW reading hides | One-minute averages while running; event record at start and at any step |
| kW, kVA and power factor | Separates engine load from alternator load and shows whether the set is near its kW or kVA limit | Same interval as current; keep the start sequence at higher resolution |
| Voltage and frequency | Proves the alternator and governor held the bus through the transfer and the load step | One-second or event-based during start and transfer |
| Battery voltage during crank | Detects a weak battery or a failed charger before the next outage | Event record for every start attempt; daily at-rest value as a trend |
| Charger current | Shows whether the battery is being maintained, not merely connected | Five to fifteen minutes at rest |
| Oil pressure and coolant temperature | Turns a run into evidence of engine health and cooling performance | One minute while running; alarm on rate of change as well as absolute value |
| Fuel level and fuel consumption | Shows autonomy and reconciles burn against the expected curve | Fifteen minutes at rest; integrate consumption over each run |
| ATS position and transfer time | Proves the changeover completed and measures the gap the UPS had to cover | Event record for every transfer, with timestamps from the same clock |
| Room temperature and ventilation status | Detects blocked louvres, failed fans and ambient above the 40 °C reference | Five to fifteen minutes; alarm on rising trend |
| Controller fault and run status | Distinguishes a set that is available from one that is latched out | Event-based, with the controller’s own fault code preserved |
Alarm limits: tiers, delays and deadbands
The most common reason a monitoring system is ignored is too many alarms, not missing data. A single threshold on a noisy signal produces notifications the site learns to filter, and the one real event is lost. The fix is three tiers, a delay on every threshold, and a named owner for each tier.
The first tier is advisory: a service is due, a battery is ageing, fuel is below the planning reserve. It belongs in a weekly report, not a 03:00 message. The second tier is a warning: coolant temperature above the normal band, oil pressure below the running band, or a failed start followed by a retry. It needs a response during the working day, on the same screen as the trend. The third tier is critical: the set failed to start, tripped on shutdown protection, the ATS failed to transfer, or the room is above the derating point. That tier goes to a person who can act, with an escalation path.
Every threshold needs three numbers: the set point, the delay before it is announced, and the deadband before it clears. A coolant temperature warning set a few degrees below the shutdown point, with a 30-second delay, avoids announcing a transient during a load step; the exact figures come from the engine maker’s data and the project specification. A low oil pressure alarm needs a delay long enough to ride through the first seconds of a start, and a deadband wide enough that it does not chatter at the boundary.
Two alarms deserve special treatment. Failure to start is the only alarm that directly answers the question the system exists for; it should be latched, timestamped with the crank count and routed to the critical tier. Communications loss matters because a silent system looks like a healthy site on a dashboard showing only the last value. The gateway should raise a local alarm and the platform a silence alarm after an agreed period.
| Tier | Example | Expected response |
|---|---|---|
| Advisory | Service interval approaching, battery age limit reached, fuel below planning reserve | Plan the work in the next maintenance window; no call-out |
| Warning | Coolant temperature above the normal band, oil pressure below the running band, one failed start followed by a successful retry | Investigate during the working day; check the trend and the last test record |
| Critical | Failed to start, shutdown protection trip, ATS failed to transfer, room above the derating point | Immediate response and escalation; treat the site as unavailable until proved otherwise |
| Silence | No data from the gateway beyond the agreed interval | Treat as unknown, not healthy; check power, SIM and local buffer |
| Test | Scheduled start, load test or manual exercise | Suppress routine notifications but keep the event record and the load readings |
Sampling, communications and what “real time” should mean
“Real time” is a marketing word until it is tied to a decision. A start sequence needs one-second resolution because the crank, the oil pressure rise and the transfer happen within seconds. A room temperature trend needs minutes, not seconds. A fuel level needs a slow sample and a filter, because the gauge moves with the tank and with the set’s own fuel return.
The practical architecture is event-driven at the controller and periodic at the gateway. The controller records start, stop, transfer and every protection trip with its own timestamp. The gateway polls analogue values more slowly, stores them locally and forwards them in batches. That local store makes the system useful on an unreliable cellular link: when it returns, data arrives with its original timestamps rather than a gap.
Communications should be outbound from the site. The gateway connects to the platform, authenticates and publishes; there is no inbound port into the generator room and no route from the internet to the controller. On a building network, the gateway belongs on a separate VLAN with a firewall rule permitting only its outbound connection. Remote control is a different decision from monitoring, and should not be enabled merely because the gateway supports it.
Time synchronisation is part of the measurement chain. If the controller, ATS and UPS keep different clocks, transfer time computed from their logs is wrong, and a correct sequence can look as if the ATS moved before the set was ready. A local time source, or a network time service where available, keeps the event log usable; the platform should display local time and keep the UTC offset.
| Data class | Resolution | Planning retention | What it is for |
|---|---|---|---|
| Start, stop and transfer events | Event, one-second timestamps | Life of the asset | Proving the sequence and diagnosing a failed transfer |
| Protection trips and controller faults | Event with fault code | Life of the asset | Warranty discussion, root cause and repeat-failure analysis |
| Electrical and thermal trends | One-minute averages | 30 to 90 days at full resolution | Load profile, phase imbalance, cooling performance and alarm investigation |
| Long-term trends | 15-minute or hourly aggregates | Two to five years | Battery ageing, fuel burn, derating and maintenance planning |
| Annual load test record | Test schedule, readings and fuel used | Life of the asset | Evidence that the set carried 50 % to 100 % of rating and held it |
Retention: keep the evidence, not every sample
Retention is usually argued as a storage cost, but the real question is what the site will need to prove. A failed transfer is proved by the event log, not a year of one-second samples. A cooling problem is proved by the relationship between load, room temperature and coolant temperature over a few weeks. A fuel dispute is proved by the consumption record and tank readings, not the raw gauge every minute.
The annual load test deserves the most care. The maintenance basis is a load test once a year at 50 % to 100 % of rating, and the monitoring system should capture the same test: load steps, coolant and oil temperature, fuel consumed and any protection that operated. A test recorded in the same system as the operating trends can be compared directly with last year’s; a handwritten sheet cannot.
Fuel is the second record that pays for itself. The planning figures are 0.24 to 0.29 litres per delivered kWh, with a 500 kVA set using about 101 litres an hour at full load, 80 litres at 75 % and 60 litres at 50 %. A 1000 kVA set at roughly 200 litres an hour uses about 4800 litres over 24 hours. Those figures turn a fuel level into an autonomy estimate and a consumption trend into an early warning. At a PUE of 1.4, a 700 kW IT load implies about 980 kW at the facility, and the generator is sized on that figure.
The low-load band is the other trend worth retaining. Operation in the 10 % to 20 % band is hard on a diesel engine, and the maintenance schedule shortens the oil interval when a set spends time there. A monitoring system that records load duration by band turns that rule into a number the planner can act on, instead of an assumption that the set “mostly runs at full load”.
Commissioning: prove the monitoring system too
A monitoring system is a measurement instrument, and it needs the same acceptance discipline as the generator. The point-to-point check compares every channel with a local gauge or a loop calibrator: current against a clamp meter, coolant temperature against a contact thermometer, fuel level against a dip, room temperature against a handheld sensor. A channel that is wired to the wrong phase or scaled by ten will produce a plausible trend and a wrong decision.
The alarm test is next. Each tier should be injected at the sensor, not simulated only in the platform, so the whole path from controller to gateway to notification is proved. The communications-loss test disconnects the link and confirms that the local buffer keeps recording and the silence alarm fires. The time-sync test compares controller, ATS and platform clocks. The final check is a real start and transfer compared against readings taken by hand.
| Check | Method | What it proves |
|---|---|---|
| Point-to-point scaling | Compare each analogue channel with a calibrated local instrument | The number on the dashboard is the number at the sensor |
| Phase and polarity | Load one phase at a time and confirm the correct channel moves | Phase imbalance and power calculations are not misleading |
| Alarm injection | Drive each tier at the sensor and follow the notification to its owner | The whole path works, not just the platform rule |
| Communications loss | Disconnect the link, wait beyond the silence period, reconnect | Local buffering, timestamp recovery and the silence alarm work |
| Time synchronisation | Compare controller, ATS and platform timestamps before and after a transfer | Event order and transfer time are trustworthy |
| Recorded start and transfer | Run a real start with the ATS and compare with manual readings | The monitoring record matches the plant behaviour |
Common mistakes and a procurement checklist
The first mistake is monitoring the controller’s display rather than the plant. It can report “running” while the ATS is still on mains, or a healthy battery while the charger has failed. The second is a dashboard with no owner: an alarm in a shared inbox that is nobody’s job is worse than no alarm, because it creates false coverage. The third is a threshold without a delay, which turns normal transients into incidents and trains the site to ignore the system.
The fourth is retention that stops at the platform’s default. If the event log disappears after 30 days, the site cannot reconstruct a failure from last quarter or compare the annual test with the previous year. The fifth is a cloud-only design with no local buffer: a cellular outage becomes a data gap exactly when the site is in trouble. The sixth is treating telematics as a substitute for maintenance; it makes the annual test, monthly inspection and fuel checks more useful, but does not replace them.
A short procurement checklist keeps the conversation on decisions: the controller model and protocol, the points list with scaling and units, the alarm schedule with tier, delay and owner, the local buffer and silence alarm, the retention policy by data class, the time source, the communications route, the cyber boundary, and the commissioning tests with pass conditions. A supplier who answers those in writing is selling a monitoring system; one who answers with a dashboard screenshot is selling a display.
Can remote monitoring replace monthly inspections and the annual load test?
No. Telematics changes the interval between observations, not the need for them. The maintenance basis is a load test once a year at 50 % to 100 % of rating, and a monthly run at 50 % or more where the site can take it. Monitoring makes those events visible but cannot put real heat into the engine or prove that the set carries its rating.
Which single measurement is most useful?
Battery voltage during the crank, with the failed-start record. It most directly answers “will it start”, and a monthly visit is most likely to miss it, because a battery can look healthy at rest and collapse under load. Three-phase current is next, because it shows whether the set is carrying load and whether one phase is overloaded.
How often should a standby generator send data?
Events as they happen, electrical and thermal values about every minute while running, and slower samples at rest. The exact intervals are set by the controller’s capability, the communications cost and the project’s reporting requirement; the start and transfer sequence needs second-level resolution, a room temperature trend does not.
How long should monitoring data be retained?
Keep event and protection records for the life of the asset, full-resolution trends for a few months, aggregates for several years, and the annual load test record while the set is in service. Those periods are a planning starting point; the owner’s evidence requirements and the project specification decide the policy.
What happens when the cellular link fails?
A well-designed gateway keeps recording locally and forwards the buffered data with its original timestamps when the link returns. The platform should raise a silence alarm after an agreed period, because a site that has stopped reporting is not a healthy site. A cloud-only design with no local buffer turns an outage into a blind spot.
Can I start and stop the generator remotely?
Technically yes on many controllers, but remote control is a separate decision from remote monitoring. Starting a set remotely bypasses the local checks a competent person makes before a run, and stopping it remotely can leave a building without power. If it is required, it needs a safety case, permissions, local interlocks and a clear rule about who may use it; most standby sites are better served by monitoring only.
Send the controller model and the points list
Tell us the controller and ATS type, the fuel tank arrangement and what the site needs to prove to its own management. The reply sets out the measurement list, the alarm schedule, the retention plan and the commissioning checks, in writing.
Written and maintained by the KVA Grid sales engineering desk. The planning figures on this page are the ones published elsewhere on this site: the annual load test at 50 % to 100 % of rating and the shortened oil interval for 10 % to 20 % light-load running from maintenance scheduling and fault finding; the fuel figures of 0.24 to 0.29 litres per delivered kWh, with 101, 80 and 60 litres an hour for a 500 kVA set at full, 75 % and 50 % load, from fuel consumption; the UPS bridging the seconds between mains loss and the generator carrying load, the PUE 1.4 example and the 1000 kVA case at roughly 200 litres an hour and about 4800 litres for 24 hours from data center backup power; the 40 °C reference ambient and derating above it from hot and humid climates; the protection and controller behaviour from control panel and protections; and the transfer sequence from ATS and AMF. Sampling intervals, alarm set points, delays, deadbands and retention periods are planning values; the set manufacturer’s data, the controller’s capability and the project specification govern the system that is built. Nothing on this page is a quotation for a particular project. Reviewed 2 November 2026.
