A multi-engine approach to transformer condition assessment. Combine deterministic diagnostics, probabilistic uncertainty analysis and a trained model where applicable, then connect the findings to trends, Action Plan and reports.
New release in preparation. Contact us to discuss your data and assessment workflow.
Ronin AI does more than apply a trained model to laboratory data. It selects applicable diagnostic methods, keeps uncertainty distinct from condition, and connects findings with the asset's history. Natural and synthetic esters have dedicated assessment routes rather than inheriting a mineral-oil model.
Prepare the data, interpret it for the right fluid, examine its history and review the next actions. This is the workflow of the forthcoming Ronin AI release.
Upload supported laboratory files and let the parser prepare the measurements. Review values, units, sample dates and asset identity before submitting the analysis.
Check field mappings, fluid identity and flagged or missing measurements against the source report. Ambiguous layouts need correction; scanned documents depend on OCR availability and source quality.
Read applicable deterministic diagnostics alongside the condition assessment. Mineral oil, natural esters and synthetic esters follow dedicated assessment routes, not one universal model.
Where available, probabilistic calculations show how classifications respond to measurement uncertainty. The trained health-index model is used where applicable. Neither an index nor a simulation frequency is a calibrated probability of failure.
Examine the asset history, changes in diagnostic findings and eligible gas-generation trends. A single sample cannot show how the evidence has evolved.
Check sample dates, units, fluid and assessment method before comparing results. Missing measurements are not zero, and a change in method must not be mistaken for physical deterioration.
Examine proposed actions linked to diagnostic findings and evidence gaps. Review timing and stated cost assumptions with the engineer responsible for the asset.
The Action Plan supports maintenance review. It does not issue autonomous operating instructions or optimise decisions from a claimed probability of failure. Loading, criticality and site constraints remain part of engineering judgement.
Prepare an Asset Health Report or Action Plan for review, and export selected analysis data to PDF or Excel. Keep findings connected to the available evidence and its limitations.
Asset Health Report, Action Plan and selected analysis exports. A dedicated fleet health report is not included in the forthcoming release scope.
The capabilities below describe Ronin AI in release preparation, not the earlier application. HAKUTAKU advisory and a dedicated fleet-report renderer are not included in this feature list.
Import supported laboratory files, preserve source values and review field mappings, units and asset associations before submitting analyses.
Use applicable Duval and standards-based methods for mineral oil and supported ester families. The mineral-oil ML model is not reused as an ester-trained model.
Review actions connected to findings, with dates, references and explicit cost bases where available. Recommendations support approval by the responsible engineer.
Read gas levels and eligible trends together. Missing values are not plotted as zero, and a change in assessment method is not presented as physical deterioration.
Organise assets and examine assessed condition alongside evidence coverage. Prioritisation still requires the consequences of failure and your operational context.
Share an Asset Health Report or Action Plan, and export analysis selections to PDF or Excel. A dedicated fleet health report remains outside the current release scope.
This existing illustration uses sample values to explain a condition-ordered list. It is not the new Ronin AI interface, and a colour band is not proof of safe operation.
Example assets ordered by assessed condition. Sample scores are out of 100, not failure probabilities.
The existing illustration below explains the relationship between condition, gas evidence and diagnostic interpretation. It is not a preview of the final Ronin AI report design or an operating instruction.
Illustrative mineral-oil model HI. Higher means better assessed condition, not lower failure probability. Ester assessments use their producing method's scale.
High ethylene, low acetylene: a hot-spot, not arcing. Coherent T3.
Illustrative sample, not real customer data. The pattern (recurring ethylene, a rising thermal fault caught by the trend before a sudden failure) mirrors a real Seetalabs deployment on a step-up unit, anonymised.
Collect measurements appropriate to the question and asset. Ronin AI separates the input requirements of its trained model from the applicability rules of deterministic methods. Reducing input count is not a reason to omit a necessary test.
Data coverage is not diagnostic accuracy. Ronin AI distinguishes missing measurements, method eligibility and uncertainty, and can withhold a result when the available evidence does not support it.
A laboratory report is the starting point, not the whole context. Fluid type, transformer duty, sampling conditions and data quality determine which methods apply.
Ronin AI combines deterministic diagnostic engines and probabilistic uncertainty analysis with a trained health-index model where applicable. These components answer different questions: what the evidence indicates, how stable the classification is, and how the assessed condition changes. The method used and its limitations matter as much as the number.
The international reference for the transformer health index and fleet prioritisation. RONIN's model is anchored to it, so the output maps back to a published method.
The Duval Triangle interpretation of the dissolved gases follows IEC 60599, so the fault type on the report is one an engineer can independently check.
AI governance requires a documented intended purpose, validation, oversight and lifecycle controls. Ronin AI's architecture supports that work; software features alone do not establish AI Act conformity or ISO/IEC 42001 certification.
The historical R² figure describes the fit of one trained health-index model to its evaluation targets. It is not the accuracy of the complete Ronin AI system, a probability of failure, or the chance that an individual diagnosis is correct. A higher R² does not automatically imply overfitting: that requires evidence from independent validation.
Ronin AI separates measured inputs, diagnostic rules and uncertainty assumptions. A result can be computable yet unsuitable for interpretation. Missing evidence, method disagreement and fluid-specific applicability therefore remain visible instead of being compressed into one reassuring score.
These are the mineral-oil model bands, assigned after rounding for display. Deterministic ester assessments use the producing engine's condition class and scale. Neither scale represents probability of failure.
No dark patterns and no fine print. Here is exactly where your data lives, what it is used for, and how it is kept apart.
Your data sits on a VPS in the European Union and is handled under the GDPR.
Your DGA, oil and asset data are never used to train or tune the model. Your fleet stays yours.
No behavioural tracking. Account data is minimal: name, email and an optional company. No photos.
Transformer records can contain commercially or operationally sensitive information. Access permissions, customer separation and retention requirements should be agreed for the deployment; asset data should not be assumed non-sensitive.
The application uses HTTPS for data transfer. Hosting, access controls and the boundaries of backend services should be reviewed against your organisation's security requirements.
Access is firewalled and the scoring service is reachable only through the application, not directly.
No hardware to buy, no capacity licence to negotiate before you can start, no hidden line items. Here is how it works, in the open, instead of a "contact us for pricing" wall.
A usage model: you pay for the predictions you run, one per asset assessed. No sensor, no gateway, no subscription for capacity you never use. Start with a single transformer.
The more assets you assess, the less each one costs. A large fleet scored in a batch is priced well below the same assets one at a time, so ranking a whole population stays affordable.
Running continuously across a very large fleet, or need dedicated terms? Higher-volume plans are arranged directly, sized to your fleet and cadence, not forced into a tier that does not fit.
Oil-treatment, testing and service companies can resell the health index and reports under their own brand. A software add-on to the sampling you already do, arranged on request.
Transparent by design. The free account runs a guided demo before any commitment. No sales calls, guided by AI.
HAKUTAKU is the planned expert advisory layer for RONIN. It is not included in the currently available application. Its scope and release date will be confirmed after implementation and validation.
The planned role is to help engineers examine evidence and references. Diagnostic calculations remain the responsibility of the validated engines, not generated text.
Discuss the roadmapAsset-specific advisory is a development direction, not an available feature or a promise of autonomous operation.
Discuss your transformer data and assessment workflow with us. Ronin AI is in final release preparation; existing customers can continue using their current application access.
Ronin AI: release preparation. Existing customer access remains available.
