RONIN AI is a transformer health index software. Start from the test data you already have: download the guided template, fill in your DGA and oil results, upload. It returns a CIGRE TB 761 health index per asset in seconds, then a whole fleet ordered by failure risk. This page is the deep dive: how it works, what you get, and how the number is built.
No sensors. No integration project. No IT rollout. Single asset or batch · 500 kVA to 800 MVA.
RONIN AI turns diagnostic data you already own into a maintenance decision. It reads the DGA and oil results you already have, scores each transformer on a health index built on CIGRE TB 761, interprets the dissolved gases with the Duval Triangle, and ranks the whole fleet worst-first. The score comes from an interpretable machine-learning model anchored to the CIGRE standards, with the clause shown next to every result. Nothing to install on the transformer, results in seconds.
Simple on the surface for the technician who is not a DGA specialist. Deep on a click for the engineer who wants to check the work. Every step is one RONIN takes for you.
Start from the test data you already have. Download the guided template, fill in your DGA and oil results from any laboratory, and upload. No sensor, no integration, no IT ticket.
✓ Any lab, guided templateA single 0 to 100 score with its band, the Duval interpretation and a recommended action. One colour, one verdict, one next step.
⏱ ~6s per assetThe whole population settles worst-first in the Prediction List, so the budget conversation starts with the right asset, not the oldest one.
⏱ ~30s · up to 500 unitsA per-asset report with the health index, Duval and action, ready to send to a board or a field team. PDF for sharing, Excel for your own reporting.
↓ PDF / ExcelEverything RONIN does today, in production. No roadmap items, no "coming soon". If it is on this page, it runs.
Score one transformer or a whole batch from the same guided template. A health index per unit, either way.
A per-asset report with the 0 to 100 index, the Duval Triangle for the DGA and a plain interpretation of the fault.
A recommended action tied to each result and its standard clause, so the next step is clear without an expert in the room.
Track an asset across testing dates and catch degradation before the next scheduled assessment, not after a failure.
Organise assets into fleets and rank the whole population by severity in the Prediction List, worst first.
Share assets and their indices as A4 or A3 PDF for the board, or export to Excel for your own reporting.
Green is healthy, red is at risk, and every asset carries its health index, date, type and area. Colour is a signal, never decoration, so the list reads even in greyscale. This is the actual RONIN interface, reconstructed here.
Every transformer you have scored, ranked so the at-risk units surface first.
A health index gauge, the Duval Triangle that places the fault, the sub-scores that built the index, the gases that drove it, and the recommended action with its standard clause. One page a technician can act on and an engineer can audit.
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.
Each parameter in a health assessment costs something: a sensor to fit, a lab test to run. RONIN's model distilled the 14 that carry the signal, out of 60 or more, so you pay to collect only what moves the score. Fewer inputs, one seventh of the usual data, the same dependable index.
Confidence is always shown, never silent. Fewer inputs means a clearly flagged data gap, not a false certainty; where the essential DGA is not there, it stops rather than guess. RONIN compensates for missing parameters; it does not mask poor-quality data. Good data in remains a precondition.
RONIN scores the oil and gas, so it does not care who built the transformer or where the data came from. If you have a lab report, RONIN can read it.
RONIN's score comes from an interpretable machine-learning model anchored to CIGRE TB 630 and TB 761, with the standard clause shown next to every result. The trust comes from interpretability and standards, not from hiding the AI or pretending there is none.
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.
Developed in alignment with the standard for AI governance and transparency, and governed against the EU AI Act obligations for AI used on critical infrastructure.
The 80% is the R² (coefficient of determination) of RONIN's health-index model, a regression metric, not a per-diagnosis error rate. It is high for a model trained on real, messy, cross-utility data. Forcing it toward 85% or more would mean overfitting the test set, a number that looks better on a slide and performs worse on your fleet. The goal is generalisation across diverse transformer populations, not a score tuned to one dataset.
The reason to trust it is not the number. It is that the model is interpretable and tied to published standards: every score carries its sub-scores, its Duval interpretation and its clause.
Colour always comes with a label and the value, so a report stays readable in greyscale and nothing is judged by hue alone.
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.
Each customer runs on a separate domain, accessible only to the owner. Asset data is non-sensitive by design.
Encryption at the highest standard, everything over HTTPS. The ML service is never exposed to the internet.
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.
Our transformer expert, grounded in the standards and RONIN’s library, answers right here, in your language. This public preview reads the library, not your fleet.
HAKUTAKU (白澤) is the beast of East Asian myth that knows every ailment and names it - fitting for one that knows transformer faults and names them.
The full expert agent that reasons over your own transformers inside the RONIN platform, not just the library. In development now.
Register once. The guided demo runs your first health index on preloaded assets in minutes, then on your own data when you are ready. No sales calls.
Free account · guided demo on sample data · 500 kVA to 800 MVA
