Tersa · Energy and power generation

How Tersa used AI to cut soot blowing by 35%

At Tersa's waste-to-energy plant, an AI model built into the control system now decides when the boiler needs cleaning. The steam it saves goes to the turbine instead: 400 MWh more electricity a year.

  • 35%fewer soot blowing cycles in the boiler's third duct
  • 400 MWhmore electricity a year, since May 2023

At a glance

  • Customer

    Tersa, waste-to-energy plant serving the Barcelona metropolitan area

  • Services

    AI, Data

  • Starting point

    Data-driven production: live process data, but boiler cleaning on a fixed 12-hour cycle

  • Where they wanted to go

    Smart production: cleaning triggered by the process itself, only when it is needed

  • Result

    35% fewer soot blowing cycles in the boiler's third duct and about 400 MWh more electricity a year, since May 2023

  • Delivered by

    AG Solution, a Quariq company

Where they were

A boiler cleaned by the clock

Tersa turns waste from the Barcelona metropolitan area into electricity. The waste is burned, the heat raises steam in the boiler, and the steam drives a turbine.

As it passes through the boiler, the flue gas leaves soot and ash on the heat exchanger surfaces, and fouled surfaces transfer less heat. To keep them clean, the plant ran soot blowing in the boiler's third duct every 12 hours. Each cycle takes steam that would otherwise reach the turbine, so every cleaning costs electricity. And the timer took no account of how dirty the duct actually was.

Tersa already had live process data and a diagnostics team that closely monitored the boiler. What it lacked was a way to turn that data into the cleaning decision.

"AI is a hot topic, but how could we seamlessly integrate it into our daily operations? That was our challenge."

Alex Sas, Maintenance Manager, Tersa

Where they wanted to go

Clean only when the boiler needs it

Tersa wanted the boiler to decide for itself: soot blowing when the duct needs it, triggered automatically, with no risk to boiler performance. Every cleaning avoided leaves more steam for the turbine and more electricity from the same waste.

In factory-of-the-future terms, that is the move from data-driven production, where people use live data to run the plant, to smart production, where the process anticipates what it needs and acts on it. Tersa chose to make that move on one routine, in one duct.

The challenge

Why fewer cleanings is harder than it sounds

Clean too little, and the surfaces foul; heat transfer drops, and the boiler loses performance. Clean too often, and the plant spends steam for nothing. The right moment also shifts as the composition of the waste changes, so any fixed interval is wrong in one direction or the other.

Two conditions followed. The model had to predict well enough for the plant to let it act on the boiler. And it had to work inside the control system the operators already use, not in a separate tool someone has to remember to check.

What we did together

One duct, one decision, built into the DCS

Rather than an AI programme for the whole plant, Tersa and our team started with a single routine whose cost was easy to measure: the number of cleanings, and the steam each one takes from the turbine.

Before anything went live, we trained an artificial neural network on the plant's historical process data and tested whether it could reliably predict when the duct needed cleaning. The model went live only once that held.

It now reads the boiler's operating parameters in real time from the plant's process database, and integrates the results into the DCS. The model triggers soot blowing only when the duct needs it, automatically and without operator action. A routine that used to run on a timer is now automated and monitored. It went live in May 2023.

"By using real-time data and algorithms, we pinpointed the exact cleaning times, saving energy and improving efficiency. AI wasn't just about automation; it was about gaining deeper insights into our production processes."

Xavier Morera, Online Diagnostic Technician, Tersa

Results

35% fewer cleanings, 400 MWh more electricity a year

Since the model went live, soot blowing in the boiler's third duct has fallen by 35% against the fixed 12-hour schedule. The steam used for cleaning has fallen by the same 35%, and that steam now reaches the turbine: about 400 MWh of additional electricity a year.

What other manufacturers can take from this

Start with the routine nobody questions

Most plants run something like Tersa's 12-hour cycle: a cleaning, purging, or flushing routine set conservatively years ago and repeated on the clock, regardless of the equipment's condition. Clean-in-place cycles in food and beverage and purge sequences in chemicals are typical examples. Each one consumes steam, water, chemicals or production time.

These routines make strong first AI use cases. The cost of each cycle is easy to measure, years of history usually sit in the plant's historian already, and the decision is narrow enough for operators to trust. Tersa's case also shows that smart production doesn't have to arrive plant-wide. Tersa reached it for one routine, on data it already had.

Transformation in practice

Automation

How Tersa migrated its waste-to-energy plant from Siemens S7 to PCS7

Read the story

AI

How a European chemical manufacturer predicts reactor scaling up to six months ahead

Read the story

View all stories

Let's talk

Talk with our industrial AI team

Tell us which routine runs on a timer at your plant. We'll look at the data behind it and tell you whether a model like Tersa's could take over the decision.

Talk with our industrial AI team