How Reduct automatically reduces steam demand

Verco’s machine learning based platform, Reduct, is currently being used on a series of trials looking at real world optimisation examples. This is the third in a series of bulletins showing live examples out in the field where Reduct has alerted on anomalies that would otherwise have gone unseen resulting in increased costs.

For this project we setup an automatic data stream into Reduct from the boiler controller system and Reduct learned the site’s operational practices from retrospective data. Within days, Reduct identified abnormalities and these continued over subsequent weeks.

Intelligent data capture

We setup a real time data feed from the boiler control system into Reduct. With just a single piece of low-cost additional hardware, this allowed all of the boiler parameters to be collected in real time and ingested into Reduct. Prior to this data feed setup, the boiler control system was measuring and utilising all these data points for operational control, but none of this was previously logged or analysed for performance improvement purposes.

Pattern recognition and machine learning

Once the real time feed was setup, we also did an export of the previous month’s data, which was extracted after the data collection hardware was first mobilised. This 30-day period allowed Reduct to pre-learn the expected trends for each of the boiler parameters which enables relevant anomalies to be detected from day 1 of system handover.

Within just a few days, Reduct had found its first anomaly relating to the boiler system. In this case, this wasn’t a boiler performance issue but a steam demand anomaly, relating to unexpected demands on the boiler from the downstream process equipment.

Over subsequent weeks, a series of anomalies were detected on the boiler’s measured steam output and these were found to highlight the following anomalies as shown in the screenshot below:

  • Unexpected unusual low point in steam demand
  • Later than normal start up after a weekend shutdown
  • Missed weekend shutdown
  • A further missed weekend shutdown

You can see from this summary that not all of these result in negative performance, in fact some of these anomalies result in a decrease in performance, but this decrease is not normally seen and is therefore flagged as unusual. In these instances, it may be that instances of reduced consumption could be replicated to drive further savings. In all of these examples, a simple non-AI high alarm threshold would not have identified these anomalies.

The value of monitoring control system parameters

Many businesses now have basic sub-metering in place, but it’s still rare to have access to consumption and performance trends taken directly from utilities control systems. These measurement points are generally sampled as part of the control and operation of utilities plant, but rarely are these parameters logged, stored and made accessible for alerting and retrospective analysis. These parameters certainly aren’t often fed into a pattern-recognition system for automatic anomaly detection, which presents a huge opportunity to identify savings from existing datasets.

Watch the Reduct walkthrough –  See a demo of Reduct in action, the dashboards and reporting outputs.  Get a feel for what Reduct can do for you!

Want to get involved?

Join the next wave of our Early Adopters Programme.

Our latest news & insights