Today
Good accuracy can hide a bad blind spot
- Healthy transformer records dominate the historical dataset.
- Rare Poor and Very Poor assets have limited representation.
- Overall accuracy can look strong while high-risk assets are under-detected.
Find high-risk transformersbefore overall accuracy hides them.
INSUS predicts health index from oil quality, dissolved gas analysis and paper-condition data your utility already collects, with class rebalancing so Poor and Very Poor units are not buried under overall accuracy. Asset teams keep the decision.
01
Health index from oil, DGA and paper data you already collect.
02
Trained to find rare high-risk classes, not just overall accuracy.
03
Advisory. Asset teams decide maintenance and capital action.
The problem
The highest-risk transformersare often the hardest to detect.
Healthy transformers dominate historical test datasets, while Poor or Very Poor units are rare. A model can post strong overall accuracy while missing the assets a utility most needs to identify. Oil, DGA and paper parameters may also be read separately or through rule-based scoring, so the joint pattern that marks a high-risk unit is easy to miss.
01Oil quality
02Dissolved gas analysis
03Paper condition
- 01Oil quality
- 02Dissolved gas analysis
- 03Paper condition
The shift
From overall accuracyto high-risk detection.
Asset and maintenance teams stay responsible for every maintenance and capital decision. What changes is whether rare high-risk classes get enough weight in the model.
With INSUS
Train for the assets that matter
- Class rebalancing gives rare high-risk conditions greater weight in training.
- Oil quality, DGA and paper-condition parameters are evaluated jointly.
- Asset teams get a structured fleet view for maintenance and capital planning.
How it works
From standard test datato a more reliable health index.
This fits Predictive Maintenance as asset-health decision support beside existing test and asset-management systems. Access is read-only. It does not replace testing or automatically trigger maintenance.
Connect
Bring transformer test data together
Connect oil-quality results, dissolved gas analysis, paper-condition indicators and historical health assessments from existing asset and test-history systems, read-only.
Understand
Learn from the fleet's history
Learn the historical relationship between available test parameters and confirmed health-index classifications across this utility's transformer population.
Predict
Classify transformer health
Predict the health-index category from Very Good through Very Poor, using a class-rebalanced model designed to reduce overlooking of rare high-risk conditions.
Advise
Focus engineering attention
Surface Poor and Very Poor transformers for maintenance and capital-planning review, giving asset teams a consistent fleet-level view of where further attention may be required.
Verify
Learn from what happens next
Compare predictions with subsequent confirmed health assessments and relevant failure or outage events so the model can be refined for this utility's fleet.
Getting started
Start with one fleet segmentand the tests you already run.
For utilities that already collect oil, DGA and paper-condition results. Delivered through Predictive Maintenance as transformer asset-health decision support. Expert Services can help with data assessment and validation. Advisory only. No new sensors.
- What we need. Existing oil-quality results, dissolved gas analysis, paper-condition indicators and confirmed health-index classifications. Failure or outage history strengthens validation.
- What you see first. One fleet segment such as a voltage class or region: predictions run alongside the existing health-index workflow, with high-risk detection checked against confirmed assessments.
- Where the limits are. Performance depends on the quality and class distribution of this utility's own test data. A model trained on one fleet must be calibrated before broad use. It does not replace testing or automatically trigger maintenance.
Related for utilities
Other utilitiesstarting points
Transformer health index prediction sits beside electricity theft detection and NIS2 cybersecurity compliance on the same network.
- 01Electricity Theft DetectionFlag likely theft from consumption patterns and prioritise inspections toward accounts that warrant a visit.
- 02NIS2 Cybersecurity ComplianceMap NIS2 obligations, surface documentation gaps and prepare incident-reporting readiness for compliance review.
- 03Utilities industryThe wider picture for grid operators: asset health, revenue protection and cybersecurity obligations.
Questions utility asset teams ask us
High-risk transformers are rare in test history, so overall accuracy can hide weak detection. This use case predicts health-index categories with attention to Poor and Very Poor classes for asset-team review.
Existing oil-quality results, DGA, paper-condition indicators and confirmed health-index classifications. No new sensors when the utility already runs that testing program.
No. It uses the existing testing program. Asset teams review the predicted health index before deciding what action is appropriate.
Class-rebalancing methods give rare health classes greater representation during training so overall accuracy is not the only success measure.
No. External peer-reviewed results on other fleets are not INSUS performance guarantees. Any model must be calibrated and validated on your own test history.
Through Predictive Maintenance as transformer asset-health decision support. Expert Services can support data assessment, integration and validation with your asset team.
Performance depends on the quality, history and class distribution of your own transformer test data. Predictions are fleet-specific and do not replace engineering judgement or capital decisions.
Utilities · Transformer asset health
Find the transformers that need attention earlier.
Start with one fleet segment and the test history behind it. The useful first answer is whether a class-rebalanced model improves high-risk detection on your own records.