TECH TALK
suffi cient data will be extremely challenging, and the number of false alerts will be prohibitive. The solution is to use Digital Twins to represent the physical models of the components. These follow the thermodynamics, and results clearly show the expert precisely what is wrong - this is “Explainable AI” (XAI). The challenge of defi ning equations for hundreds of thousands of components and operating conditions is handled by digital twins predicting the correct behaviour at every operating condition for many components, all following the same physical model. Identifying the point when, for example, the compressor in a unit deviates from the digital twin will not only highlight the deviation in performance but also point directly at the root cause.
Explainable AI increases the precision of early warnings with minimal human intervention. As components follow the
same physics, the digital twins - for example, a compressor - off er a good prediction for similar compressors. A library of digital twins can be applied without requiring individual training for each product. Training a reliable digital twin can require data for 6-12 months to cover all operating conditions and loads. The fi gure below shows that the digital twin (orange dots) is capable of predicting measured condenser effi ciency (blue dots) within a few per cent during a dynamic operation where the variation of effi ciency is > 20%. Widespread implementation of AFDD will not be achieved with manual confi guration due to a lack of experienced engineers, even if the savings would be much higher than the cost. In Figure 5, the blue dots represent the measured condenser effi ciency, and the orange dots represent the predicted value from the twin. Four scatters of data clearly represent diff erent operating conditions. These scatters represent summer and winter operations at 100% and 50% capacity, respectively. Figure 6 shows the prediction (orange) trained on the condenser in one unit and the measured (blue) in another. The Digital Twin is capable of representing the performance with high accuracy. It shows that the trained condenser is 1-2% less effi cient than the tested one, which is well within the acceptable tolerances.
Conclusion Growing climate awareness and requirements on
sustainability reporting, combined with the increased availability of sensor data and digitalisation, accelerates the transition towards predictive maintenance (PdM) and energy optimisation.
Initial investments in data collection and analysis are
required. However, these costs are insignifi cant compared to the fi nancial and environmental impact of ineffi cient operations, unexpected failures, and energy waste. By embracing Explainable AI (XAI) over conventional black-
box AI, the industry can unlock new levels of early warning and precision in fault detection. Instead of relying on time- consuming on-site troubleshooting and reactive maintenance, deviations can be identifi ed early, reducing the number of failures, site visits, system downtime, and performance drift. The result is not only lower costs but also longer equipment life and reduced environmental impact, which increase property value.
The challenge is transitioning from today’s maintenance Figure 5
contracts, which have low initial costs but higher future costs in energy losses, troubleshooting, and premature failures, to a proactive, data-driven approach that ensures long-term effi ciency and reliability. Without Explainable AI, effi ciency is dependent on the individual technician/engineer evaluating data. With Explainable AI, unbiased and actionable information is available in real-time. The industry can no longer aff ord to assume that systems operate effi ciently by default. Embracing digitalisation, automation, and Explainable AI is no longer an option; it is necessary for a sustainable and cost-eff ective future.
Figure 6 30 April 2025 •
www.acr-news.com Download the ACR News app today
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40
