The IT/OT Insider
Toggle menu

Virtual Maintenance Engineer

Build a Virtual Maintenance Engineer for Root Cause Analysis (RCA)

We are big believers in learning by doing. So the only way to really understand what AI can do for maintenance/reliability is to try it yourself. And because getting started is often the hardest part, we're giving you a problem, data, a manual and a playground to give it a go. All you need is access to any chatbot (even free versions).

← Industrial AI: The Roadmap from Chat to Virtual Operator

New Our AI Academy: first group starts February 26, 2027. Now with early bird offer for the first 10 seats. Inspire. Be amazed. Know where you stand. Know what’s next. Start fast. Deploy at scale. Discover the AI Academy →

Turbine 1 is down

Good morning! You are a reliability engineer and as you arrive at the Copper Canyon Dam you hear that Turbine 1 was shut down by operations.

A high temperature alarm on pad 3 of the turbine guide bearing went to high-high. The shift followed the shutdown procedure, but now it’s up to you to start the RCA to find out what happened so the unit can be turned back on as soon as possible.

The case: you work as reliability engineer for the Copper Canyon Dam. The challenge: Turbine 1 was stopped by the operators because of a bearing temperature alert, and you need to run the RCA process. Pictured: David and Willem visiting the Hoover Dam, which inspired this fully fictitious case.

The steps in your RCA process:

  1. Timeline

    Build a timeline of the relevant events across several data sources

  2. Data Analysis

    Combine and analyse all sorts of data: sensors, manuals, shift logs, lab results, previous RCAs and more

  3. Hypothesis

    List the possible causes that led to this incident

  4. Validate

    Test every hypothesis against the available evidence

  5. Report

    Summarise all your findings in a structured format

Doing RCAs is a bit like being Sherlock Holmes. Normally that means hours or days of digging through systems and binders. But not for you! You use your cool new RCA bot to speed up the entire process to the time it takes to get a coffee.

It starts with a prompt

"Start the RCA process for Turbine 1. CH PLC alarm TGB pad 3 high-high at 3 Nov 2026 09:00, unit unloaded and stopped by the shift per instruction OP-07."

A prompt like this is not enough. An LLM will always give you an answer. It’s up to you to ensure the answer is a good one, using these three elements:

1. Task

What

What you want the AI to do, and the role it should play.

Why

Tell the LLM the role, method and output you expect.

2. Data

What

The data the AI will work on: exports from the historian, shift logs, previous maintenance orders, manuals...

Why

Without data, no analysis.

3. Background Info

What

The situation behind the request: the asset, what happened, what has been tried.

Why

The same failure can have different causes depending on the operational context.

The demo data set: historian trends, oil analysis, reliability data, shift logs, a previous RCA, supplier documents, work orders and internal documents.

To get you started in just a couple of minutes (and show you the real power of AI), we have already created an Instruction and some Demo Data you can use.

A fully working (simulated) dam

We’ve built a complete simulated environment, and you’re free to use it. A working dam with a SCADA, shift logs, work orders, manuals, operating instructions and much more. Everything a reliability engineer would need to dig into this incident.

Perfect starting point

Start with a simple chat

A chatbot with the five starter pack files attached (instructions, documents, records, alarms and events, and a three-week trend) and the prompt to read the instructions and start the Copper Canyon Dam RCA process for Unit 1.

You have access to Claude, ChatGPT or Copilot, but you are not very familiar with AI. You want to run the RCA yourself and see what happens. This is a simplified version of our bot: you only need to upload some documents. No installation needed.

Download the starter pack (zip) →

Follow the instructions in README.pdf.

Next step

Run our actual bot

Claude Code running in the RCA bot pack folder, asked to read the library and historian READMEs, list the available documents and data, and say what is missing for an RCA.

You want to run the real RCA bot on your own computer, using Claude Code or a similar agentic tool. We’ve prepared a dataset for you. This needs a bit more knowledge and has some dependencies, such as Python.

Download the RCA bot pack (zip) →

Follow the instructions in README.pdf.

Advanced

Geek out with us

The simulated Copper Canyon Dam SCADA showing Unit 1, a 40 MW Francis turbine-generator, while generating: output, wicket gates, auxiliaries, unit control, load, the last 30 minutes of power and bearing temperatures, and the turbine bearing signals.

You’re comfortable with GitHub, Claude Code and Docker, and you want the entire environment: the dam, the SCADA, the data, the bot. The full manual is on the wiki of our GitHub.

github.com/itotinsider →

Feel free to suggest improvements!

Article · Video

What is Claude for Manufacturing? (And how to try it yourself)

We ran Anthropic’s RCA skill and our own skill on this dam, with the same prompt and the same data. Here is what we found, and how to compare them yourself.

Read the article →

⚠️ A word of caution: this is a fictitious dam, so play as much as you like. When you move on to your own plant data, use the AI tools your company provides and allows. Don’t paste company data into public services.

Running the demo is the easy part

Making it work in your organisation is the real challenge. Buying a Claude or Copilot subscription is not enough: there are technical barriers (data availability, quality and context) and organisational ones (skills, compliance and, above all, cooperation). So here’s how we’d approach it:

Do now

1. Experiment

  • Use our prompts as starting point.
  • Use your data with your company GPT (if allowed)
  • and learn where AI shines and where it fails.

Explore

2. Organize

  • Work and learn together as a maintenance community, not as isolated individuals
  • Build central, chat-based solutions that everyone can use: an RCA bot, a 5-Why bot, a training bot…
  • And remember: IT won’t make your business case. That’s on you.

Do next

3. Cooperate

  • Get to know your IT, Data and AI teams, and invite them to the shop floor
  • Start a cross-functional initiative with IT/AI and build a common roadmap
  • Scale your solutions by leveraging data platforms and Enterprise AI tools

“We’ll do it ourselves!”

Great, we love that. Here’s where to get a kickstart:

The Maintenance AI Kickstart Package

  • Assessment: where you are vs. your peers
  • On-site workshop: inspire your team and map your gaps
  • A roadmap: you can take to management, including how to get IT on board

€9,490 · fixed price (travel costs excluded)

Book your intake call with David

When it comes to AI, starting today is always the better option. So go on, download the pack and get that turbine back online.

Title slide: How to Build Your Virtual Maintenance Engineer for RCA, BEMAS, Nov 10, 2026, by David Ariens and Willem van Lammeren.

Get our full presentation

As delivered to the Belgian Maintenance Association (BEMAS) on Nov 10.

Download the slides (pdf)