AI/ML Intelligence
The intelligence layer
isn't pattern recognition.
It's physics.
VIE's AI does not train on historical failures and look for similar patterns. It builds a virtual model of each transformer from its physical geometry and compares every measurement against that model. The distinction is fundamental: a pattern-recognition system cannot catch a defect that was present at installation. A physics-based model can.
Not a Learned Baseline.
A Physics-Based Model.
Most condition-monitoring systems learn a baseline from the asset's own operating history and then flag departures from it. That approach assumes the transformer was healthy when monitoring began. VIE does not make that assumption.
Operators routinely install sensors on transformers of unknown condition. A baseline learned from a unit carrying an existing defect would absorb that defect as normal. It would hide the problem the system was built to find.
VIE works from a different reference point. From each transformer's approximate geometry, VIE constructs a virtual model of how that specific unit should vibrate and conduct heat, given its construction, its load, and its ambient environment. Every measurement is compared to that model, not to the asset's own past. A fault present at installation is visible from day one. A fault that develops later is visible as it develops.
A clean reading from VIE is an affirmative statement that the transformer's measured behavior matches a healthy unit. It is not merely the absence of an alarm.
The Model Improves With Every Sample
VIE sensors record 3 to 4 samples per hour. Every sample refines the virtual model. The platform does not converge on the asset's past behavior. It converges on an accurate, physics-grounded understanding of what that specific transformer should do. It never absorbs an existing fault as the new normal.
The same geometry that informs the model also informs sensor placement. Each sensor is positioned to best resolve the mechanical and thermal signatures that indicate fault conditions on a transformer of that construction.
3-4 samples per hour
10+ years
Six Failure Mode Channels
VIE monitors six distinct failure mode categories simultaneously, on every transformer in the monitored fleet. These are the failure modes that precede gas formation — the ones that develop for months before dissolved gas analysis picks up anything.
Weather and Load Integration
Environmental variation is the most common source of false alarms in threshold-based monitoring systems. An ambient temperature spike causes thermal readings to rise. Online DGA concentrations shift with load and temperature. Without a model that accounts for these variables, any monitoring system will produce findings that require human investigation to resolve.
VIE fuses ambient temperature, solar load, wind speed, and humidity data with every sensor reading. The platform models the transformer's expected thermal and mechanical state under each set of conditions. The finding that surfaces is the deviation that remains after weather and load effects are removed — the signal that indicates an actual fault, not an environmental condition.
95%+
Vie's reported prediction accuracy.
From Alarm to Planning Tool
Most monitoring systems operate on a binary logic: within tolerance, or in alarm. That binary forces maintenance programs into a reactive posture — nothing happens until a threshold trips.
VIE tracks rate of degradation as a continuous measurement. The platform does not ask "is this transformer in alarm?" It asks "how fast is this transformer's condition changing, and when does that rate of change put the asset at risk?" The answer is what allows maintenance teams to plan interventions months in advance rather than respond to failures.
The platform converts monitoring from a surveillance system into a planning system. The window it provides is 3 to 6 months.
By the Numbers
Sensor Sample rate, continuous
Typical detection lead time before failure event
Prediction Accuracy
ROI in months (KPMG-validated)
Frequently Asked Questions
No. VIE does not train on a database of historical failures and then pattern-match against new data. The platform constructs a virtual model from each transformer's physical geometry and uses that model as the reference for every measurement. The approach is physics-based, not pattern-learned. This means VIE can detect a defect that was present at the time of installation — not only one that develops after monitoring begins.
No. The virtual model is constructed from geometry, not from observed operating history. It is active from the first sample. VIE refines its model continuously as more data accumulates, so detection accuracy improves over time — but no training period is required before the platform produces findings.
VIE does not use fixed thresholds. The platform fuses weather data (ambient temperature, solar load, wind speed, humidity) with each sensor reading to model the transformer's expected state under those specific conditions. The finding it surfaces is the deviation that remains after weather and load effects are accounted for. This eliminates the false alarm rate that threshold-based systems produce when environmental conditions change.
Lab DGA results feed directly into VIE's platform. The AI combines gas data with the continuous vibration, thermal, and oil quality record it has been building for that specific transformer. This integration extracts correlations no domain expert could derive from gas data alone — rising vibration signatures alongside specific gas ratios, thermal shifts that precede a lab test result by weeks. The accuracy of VIE's model compounds with every additional lab result.
The Science Has a Technical Reference
The physics behind VIE's monitoring approach is explained in full in the vibration science documentation. The demo shows it operating on live transformer data.