From Data to Decisions: Why Our Transformer Agent Looks at History, Not Just Numbers

Yashaswini Kareppanavar — Sat, 07 Nov 2026 17:55:26 +0000

From Data to Decisions: Why Our Transformer Agent Looks at History, Not Just Numbers
Yashaswini Kareppanavar
Yashaswini Kareppanavar
Student at KLE Technological University - Hubballi (India)

November 7, 2026
From Data to Decisions: Why Our Transformer Agent Looks at History, Not Just Numbers

AI is quickly becoming part of modern engineering. It now helps manufacturing, healthcare, transportation and energy teams process information, spot patterns and make complex decisions.

The power sector is a natural fit.

Transformers are monitored through oil analysis, electrical testing, inspections and maintenance work. That produces a lot of valuable data, but it usually sits across separate reports and records. Understanding how one transformer has changed over time can mean digging through several documents.

That’s why we built Transformer Health Historian – DGA Trend Agent: a memory-driven AI agent for transformer monitoring.

Chatbot vs. AI Agent

A chatbot answers questions. An AI agent works towards a goal: it retrieves information, uses tools, processes data and takes several steps before it responds.

A chatbot can tell you about transformer maintenance in general. A transformer health agent should be able to answer this:

“What happened to transformer T-3 during its last three DGA tests, and was any maintenance done before the latest change?”

To answer that, the agent needs the right records, a way to compare them and a clear way to present the findings.

That’s where memory comes in.

What “Memory” Means Here

For us, memory isn’t just remembering past chat messages. It’s an organised store of each transformer’s history:

🔖 Transformer identification details

🧪 Previous DGA test results

📈 Gas concentration trends

⚡ Loading events

🛠 Maintenance activities

🔍 Inspection observations

⚠️ Historical abnormal conditions

Whenever a new analysis is needed, the agent pulls this history back up.

A Real Example: Same Reading, Two Answers

T-3’s latest DGA report shows acetylene at 4 ppm.

❌ A basic system checks it against a reference limit: “The reported value is within the selected reference limit.” That’s correct, but it says nothing about change.

✅ A memory-enabled agent looks back, finds acetylene was previously 0 and spots a maintenance record from last month:

“The acetylene concentration in T-3 increased from 0 to 4 ppm after the previous month’s tap-changer maintenance. This change should be reviewed alongside other gas concentrations and operating information.”

The AI isn’t any smarter in the second case. It simply has access to the history.

How It Works

1️⃣ Collect: DGA reports, loading records and maintenance details, entered manually or imported from files

2️⃣ Organise: records are grouped by transformer ID with dates, units and sources, so histories never mix

3️⃣ Remember: history is stored so it can be retrieved anytime

4️⃣ Ask: an engineer asks a question, like “Show me the hydrogen trend for T-3”

5️⃣ Retrieve: the agent identifies the transformer and pulls its readings

6️⃣ Analyse: it compares measurements, calculates changes and flags what needs attention

7️⃣ Explain: it answers in plain language and keeps recorded facts, calculated trends and interpretations separate

Why Context Matters

A rise in a dissolved gas might point to a developing internal problem. Or it might not. An experienced engineer would ask:

Was the transformer under unusually high load?

Was maintenance done recently?

Has this change appeared before?

Are other gases rising too?

That’s why our historian stores more than gas levels. It links measurements to operating and maintenance records.

But a connection between two events doesn’t prove one caused the other, and the agent says so clearly.

Just Ask

Engineers don’t need another complicated dashboard. They can simply ask:

💬 “Compare the last three DGA reports for T-1.”

💬 “Which gas has shown the largest increase?”

💬 “What maintenance was performed before the latest test?”

💬 “Summarise the available history of transformer T-3.”

This doesn’t replace technical reports. It makes the information inside them easier to reach.

Good Data In, Good Answers Out

An agent is only as good as its records. Wrong transformer IDs, missing dates or mixed units lead to misleading comparisons.

So the system checks that readings are numeric, units are consistent and every test belongs to the right transformer. It flags missing history instead of inventing values. If it only has two reports, it won’t pretend to know a transformer’s long-term behaviour.

Can It Predict Failures?

Not yet, and we won’t claim it can.

Reliable failure prediction needs lots of high-quality data, documented fault events, proper diagnostic models and careful validation. Our focus is historical trend analysis and context-aware decision support. Anomaly detection and predictive models can come later.

Who Can Benefit

⚡ Electricity distribution companies managing records across many substations

🔧 Substation maintenance teams who need past results during inspections

🏭 Industrial facilities running transformers in plants and large installations

👷 Electrical maintenance contractors keeping organised histories for the units they service

Challenges and What’s Next

🔐 Data security, because transformer records can be operationally sensitive

🧾 Traceability, so engineers can see which records back each conclusion

🧠 Explainability, so engineers know why a change was flagged

Next up: automated DGA report extraction, trend visualisation, anomaly detection, maintenance reminders and integration with asset management systems. Each will be tested carefully before it’s used in real decisions.

Final Thoughts

The real value of an AI agent isn’t just that it can generate answers. It’s the information it can retrieve, the context it understands and how reliable its reasoning is.

An intelligent system shouldn’t treat a transformer reading as an isolated number. It should understand the history behind it.

AI #AIAgents #PowerSystems #ElectricalEngineering #PredictiveMaintenance #DGA #Transformers

Watch the 4-minute demo: https://www.youtube.com/watch?v=PcSKZOFjHrc

Explore the code on GitHub: https://github.com/ayush-an-grampurohit/transformer-health-historian

DEV Community: Yashaswini Kareppanavar

From Data to Decisions: Why Our Transformer Agent Looks at History, Not Just Numbers

AI #AIAgents #PowerSystems #ElectricalEngineering #PredictiveMaintenance #DGA #Transformers