A predictive alert doesn't predict the future — it catches a slow-moving physical change (rising coolant temperature, falling oil pressure, drifting steam pressure) early enough that a technician can intervene before that change becomes a shutdown. That's the entire mechanism. No AI magic, no guesswork: just a sensor reading a value more often than a human ever could, compared against a threshold set from the equipment's own service manual.
Quick take: Most unplanned DG and boiler failures don't happen instantly — they build over hours or days through a detectable chain of physical symptoms. IoT monitoring shortens the gap between "symptom starts" and "someone finds out" from days (the next manual round) to minutes (an automatic alert), which is where the actual downtime saving comes from.
The failure chain: what actually happens before a DG trips
Take one of the most common causes of unplanned generator shutdown: coolant system degradation. It rarely fails in one step. A typical chain looks like this:
- Hour 0: A coolant hose develops a slow leak or the radiator fan belt starts slipping. Coolant temperature begins trending upward — still within the "normal" band on a gauge a human would glance at.
- Hour 2-6: Temperature crosses the manufacturer's warning threshold (typically 95-100°C for most diesel gensets). No visible symptom yet — the engine still runs and sounds normal.
- Hour 6-12: Temperature approaches the shutdown threshold (105-110°C). The engine's own protection logic will cut power automatically if it gets here, but by then the generator is unavailable exactly when it's needed.
- Hour 12+: Without intervention, sustained overheating risks head gasket damage, warped cylinder heads, or seized components — turning a ₹2,000 hose replacement into a ₹150,000+ repair.
A manual round (typically once per shift, sometimes once per day for standby generators) might catch this at hour 8 or not until the shutdown itself. An IoT sensor polling every 10-30 seconds catches the threshold breach at hour 2-6 — while the fix is still a hose, not an engine rebuild.
The same pattern on a boiler: steam pressure and water level
Boilers fail through comparable chains: a feed-water pump degrading pushes water level down gradually before a low-water trip; a fouled burner nozzle causes flue gas temperature to climb before combustion efficiency collapses; a fuel-supply restriction shows up as a slow steam pressure decline hours before a plant-wide steam shortage. In every case, the physical signal exists well before the failure — the question is only whether anything is reading it often enough to notice.
How a threshold-based alert is actually set up
There's no machine-learning model guessing at your equipment's health. The process is direct:
- Pull the manufacturer's normal operating range for each parameter (coolant temp, oil pressure, steam pressure, battery voltage, etc.) from the equipment's own documentation.
- Set a warning threshold below the manufacturer's shutdown/trip point — this is the "something's starting to drift" alert.
- Set a critical threshold at or just below the actual trip point — this is the "act now" alert, routed to WhatsApp/SMS with higher urgency.
- Route each alert to the right person — a warning might go to the site technician; a critical alert might also go to the facility manager and EddyBits' own monitoring desk.
What predictive alerts don't do
To be direct about the limits: threshold alerts catch gradual, sensor-visible degradation. They don't predict sudden mechanical failures with no leading indicator (a bearing that seizes with no prior vibration change, for instance) and they don't replace physical maintenance — an alert tells you something is wrong, a technician still has to go fix it. The value is entirely in the lead time between "problem starts" and "someone finds out," not in eliminating maintenance work.
The real number: how much lead time matters
| Detection method | Typical time to notice a developing fault | Typical intervention cost at that point |
|---|---|---|
| Manual round (1-2x/day) | 4-24 hours after onset | Often after threshold already breached — repair-level cost |
| IoT monitoring (continuous) | Minutes after onset | Usually still preventive-maintenance-level cost |
The gap between those two rows — not any predictive algorithm — is what IoT monitoring is actually selling.
Frequently Asked Questions
Does predictive alerting use AI or machine learning?
No — EddyBits' alerting is threshold-based, comparing live sensor readings against manufacturer-specified normal operating ranges. This is more transparent and auditable than a black-box model, and it's what makes the alert logic explainable during a compliance review.
How fast can an alert reach a technician?
Threshold breaches are pushed as WhatsApp and SMS alerts within seconds of the reading crossing the configured limit, so the delay is effectively the sensor's polling interval, not a batch-processing delay.
Can alert thresholds be customized per site or per device?
Yes. Thresholds are configured per parameter per device during commissioning, since normal operating ranges vary by equipment model, load profile, and site conditions.
Will predictive alerts eliminate the need for scheduled maintenance?
No. They shorten the time between a developing fault and human awareness of it, but a technician still needs to inspect and repair the equipment. Predictive alerting and scheduled maintenance are complementary, not substitutes.