How Onnu is using AI to get more from pyrolysis

AI is fast becoming a huge part of how our engineering team works - it takes on the heavy analysis so that our engineers can spend their time on the decisions that matter. Most people use AI through a chat window, asking general questions and getting general answers. That has its uses, but a run-of-the-mill chatbot knows nothing about your plant, the day-to-day processes, or your data. Training AI on a business's own operating data, and making the results reliable enough to act on, takes far more work.

‍

A pyrolysis plant is a data machine

A CarboFlow plant runs around the clock. Every few seconds, sensors across the plant log temperatures, pressures, feed rates, auger speeds, gas flows and much more. A single plant generates millions of data points a year, and the total grows with every new installation.

Data on that scale is valuable, but no engineer can read through all of it. The most useful patterns build slowly. A temperature might drift over several weeks, or a bearing might start to show early signs of wear over many months. Some patterns only appear when you compare one site's feedstock with another's. Finding these patterns and predicting what comes next is exactly what AI does well.

‍MainBrain

MainBrain is the name we use for Onnu's in-house AI system. It takes in all running data from connected CarboFlow plants in real time but, more importantly, is trained on the plant itself: the engineering design, the equipment specifications and the control logic.

That combination is what makes it so useful. A general AI tool can see a number change but MainBrain knows how a CarboFlow plant is designed to behave, so it can tell when something is drifting away from normal operation and suggest why.

What it delivers

- Real-time advice for operators: MainBrain continuously assesses how the plant is running and gives operators clear guidance to keep it stable and efficient.

- Predictive maintenance: By picking up early signs of wear or drift, MainBrain helps us schedule maintenance before a fault causes downtime. Maximizing up-time is central to the return on any plant.

- Better future designs: Real operating data across sites and feedstocks shows our engineers where the design can be improved. Those lessons go straight into the next generation of CarboFlow.

- More accurate performance forecasts: When a new customer brings a new feedstock, we can base biochar yield and throughput forecasts on real running data rather than estimates alone. That gives project developers more confidence in their business case from the start.

Engineering that keeps learning

AI trained on large amounts of real running data is an ideal tool for our engineers. They can query the data, uncover trends and improve future designs in minutes rather than days. That frees them up to focus on new projects.

For our customers, it means CarboFlow plants that run more reliably, perform as forecast and keep on improving. Every hour of extra up-time is another hour spent turning waste into biochar and removing carbon.