A chiller compressor does not fail without leaving a trail — the electrical current it draws, the refrigerant pressure it maintains, and the delta-T it produces across the evaporator all shift subtly as a fault develops, long before the compressor trips, short-cycles, or stops producing adequate cooling. These three signatures, current, pressure, and delta-T, are exactly what an AI failure prediction model watches continuously, learning what normal looks like for a specific chiller and flagging the moment its behavior starts drifting away from that baseline. For a piece of equipment as central and as expensive to replace on an emergency basis as a chiller, catching that drift weeks ahead rather than discovering a failure during peak cooling demand is the difference between a planned repair and a costly outage. Teams wanting to see this running on their own chiller data can start with a conversation with iFactory's support team about what signature-based chiller monitoring looks like in practice.
A Chiller Compressor Fault Shows Up in Its Current Signature Weeks Before It Shows Up in the Building
Compressor current, refrigerant pressure, and delta-T signatures reveal a developing fault mode weeks before capacity loss, short-cycling, or an outright trip.
Why These Three Signatures, Specifically
Compressor current reflects the mechanical load the compressor is working against, and a gradual rise or an unusual pattern in that current draw often signals developing bearing wear or valve issues before they become audible or visible. Refrigerant pressure across the high and low sides of the circuit reveals whether the refrigerant charge or a component like an expansion valve is behaving normally, and delta-T across the evaporator ties both of those signals back to actual cooling performance, making the combination of all three far more diagnostic than any single measurement watched in isolation.
Common Chiller Fault Modes and Their Signatures
Different developing faults produce different combinations of signature drift, and recognizing the pattern is what lets a model point toward a likely cause rather than a generic "something is wrong" alert.
Compressor Bearing Wear
Typically shows as a gradual, sustained rise in current draw at constant load, often the earliest detectable signature before any audible or vibration symptom becomes apparent.
Refrigerant Loss
Produces a characteristic decline in low-side pressure alongside a widening delta-T deviation as the system struggles to maintain cooling capacity with a diminishing charge.
Fouled Condenser or Evaporator
Shows up as elevated head pressure combined with reduced delta-T, reflecting reduced heat transfer efficiency well before it becomes a capacity complaint.
See Your Chiller's Current, Pressure, and Delta-T Signatures Together
Book a 30-minute walkthrough of how iFactory's AI models track chiller signatures against a learned baseline for early fault detection.
Detection Approaches Compared
How chiller condition is monitored determines how early a fault mode is actually caught relative to when it would otherwise become noticeable.
| Approach | Signals Monitored | Typical Detection Point |
|---|---|---|
| Manual Log Review | Periodic spot readings recorded by technicians | Only if the anomaly happens to be active at check time |
| Threshold Alarms | Single-point alarms on fixed limits | After the value has already crossed an unsafe threshold |
| AI Signature Analysis | Continuous current, pressure, and delta-T trend together | Weeks before a threshold would be crossed |
Building a Chiller-Specific Baseline the Right Way
A model is only as useful as the baseline it compares against, and building that baseline correctly for a specific chiller is what makes the difference between a system that catches real faults and one that either misses them or floods a team with false positives.
Capture a Full Operating Season
A baseline built from data spanning a full cooling season captures the normal range of behavior across varying load and outdoor conditions, not just a narrow snapshot.
Normalize Against Load and Weather
Comparing current signature at a given load and outdoor temperature against the expected pattern for those same conditions is what let the model in the scenario above separate a real leak from normal variation.
Refine as More Data Accumulates
The model's accuracy improves as it accumulates more operating history specific to that chiller, reducing false positives over successive seasons.
A Composite Scenario: The Delta-T Drift That Predicted a Refrigerant Leak
A commercial building's chiller had been operating within its normal alarm thresholds for months, with no fault codes or trips recorded, when its AI monitoring model flagged a gradual, combined pattern of declining low-side pressure and widening delta-T deviation that had been building slowly enough to stay under any fixed alarm threshold. Facilities staff scheduled an inspection based on the flagged pattern even though the chiller showed no operational symptoms at the time.
The inspection found a slow refrigerant leak at a fitting that had likely been present for weeks, gradually reducing charge in a way that had not yet become severe enough to trip a low-pressure alarm or noticeably affect cooling capacity. Repairing the leak and recharging the system before the loss progressed further avoided what would eventually have become a capacity shortfall during a high-demand period, and the cost of the planned repair was a small fraction of what an emergency mid-summer refrigerant system repair would have run.
Mistakes That Undermine Chiller Fault Detection
Relying Only on Fixed Threshold Alarms
A slow drift that never crosses a hard alarm limit, as in the scenario above, goes completely undetected by threshold-based monitoring alone.
Watching a Single Signal Instead of the Combination
Pressure alone or delta-T alone might not clearly indicate a problem, but the combined pattern, as seen in the scenario above, is what makes the fault mode identifiable.
Dismissing a Flag Because the Chiller "Seems Fine"
Equipment operating within normal alarm thresholds can still be developing a real fault, exactly the situation the AI model caught in the scenario above well before any symptom appeared.
Deferring Inspection of a Flagged Pattern Until the Next Scheduled Visit
Waiting for a routine visit to investigate a flagged anomaly gives up much of the lead-time advantage that prompted the flag in the first place.
Is Your Chiller Monitoring Catching Slow Drifts, Not Just Alarms
Current, pressure, and delta-T are tracked together, not as isolated readings
The combination is what revealed the refrigerant leak in the scenario above, while any single signal alone likely would have stayed ambiguous.
Flagged anomalies are investigated even when no alarm has triggered
A pattern flagged by trend analysis, well before it reaches a hard threshold, needs its own investigation trigger separate from the alarm system.
Chiller data is captured continuously across the full operating season
Continuous capture is what let the slow leak in the scenario above be caught over weeks rather than only at the next scheduled inspection.
Frequently Asked Questions
Why is compressor current signature useful for predicting chiller failure?
Compressor current reflects the actual mechanical load the compressor is working against, and developing issues like bearing wear or valve problems typically increase that load gradually before producing any audible or operational symptom, which is why a rising current trend at otherwise constant operating conditions is one of the earliest reliable indicators of a developing mechanical fault.
Can a slow refrigerant leak really stay undetected by standard alarm systems for weeks?
Yes, a slow leak reduces refrigerant charge gradually, and standard low-pressure alarms are typically set to trigger only once charge has dropped enough to risk equipment damage or significant capacity loss, meaning a slow leak like the one in the scenario above can persist for an extended period below that threshold while still degrading system performance and worsening steadily toward eventual failure.
How does an AI model distinguish a real fault from normal seasonal variation in chiller signatures?
The model accounts for outdoor conditions, load, and other operating context when establishing what normal looks like, comparing current behavior against the expected pattern for those specific conditions rather than a flat baseline, which is what allows it to flag a genuine deviation, like the pattern in the scenario above, rather than mistaking seasonal cooling demand changes for a fault.
What should a facilities team do when a chiller signature anomaly is flagged?
Scheduling a targeted inspection focused on the specific signature that deviated, rather than waiting for the next routine visit, is what allowed the leak in the scenario above to be found and repaired before it worsened into an emergency situation. Prompt investigation of a flagged pattern is what preserves the lead-time advantage the detection was designed to provide. Book a demo to see how iFactory prioritizes and routes flagged chiller anomalies to facilities teams.
Is chiller signature monitoring worth the investment for a single central plant chiller?
Given how costly and disruptive an emergency chiller failure is, particularly during peak cooling demand, monitoring even a single central plant chiller often justifies its cost through avoided emergency repair expense and downtime alone, exactly the kind of outcome achieved in the scenario above where a planned repair replaced what would likely have been a mid-summer emergency. Facilities teams evaluating this for a single critical chiller can reach iFactory support for guidance.
Catch the Fault Mode Before Peak Cooling Season Finds It For You
iFactory tracks compressor current, refrigerant pressure, and delta-T together against a learned baseline, flagging chiller faults weeks before they become an emergency. Book a walkthrough to see it running on live chiller data.
