Andrew Normand reveals how plant operators can save a small fortune by deploying a sophisticated AI engine
The past months have been rough for the power sector, with the collapse in industrial electricity demand, lower investment and cost pressures that only seem to deepen for a prolonged period. According to IEA (International Energy Agency), post-2020, the drop in global electricity demand was around 2% and GDP was 4.4% down, the largest decline in more than 50 years.
The pressure to optimise plant operations has been like never before for all players in the power ecosystem. Whether you are a conventional or renewable producer, ensuring business continuity, productivity and safety with an eye on the long-term target of improving operational efficiency and automation are mission-critical today.
In such scenarios, it is important that equipment and resources are always performing at their optimum capacity, protecting operators from costly downtimes, extensive repairs and efficiency losses. That is what artificial intelligence (AI) can help to achieve.
How Can AI Help?
AI can help optimise complex power plant networks or small-scale cooling systems. The efficiency of these systems depends not only on the equipment but also on a range of broader factors, including upstream/downstream environmental conditions such as sudden, unexpected deviations in weather patterns and the impacts of auxiliary systems and operating conditions.
The problem is that it’s often challenging to create predefined rules that will deal with the complexity of these interacting variables. Plant operators, therefore, have to settle for identifying more pronounced symptoms with limited signals to avoid false alarms while still catching an issue before a major failure. However, the damage has already been done, and it doesn’t even identify the upstream/downstream operations that may have caused the issue. This can lead to numerous problems, including system failure, loss of efficiency, and reduced availability leading to lost productivity – which could potentially cost plant operators millions of pounds in lost revenue.
Step forward AI. When adequately targeted with purpose-built applications – for plant monitoring purposes, for example – AI can review entire systems and strip away the noise to provide engineers with far greater accuracy of understanding and much more confidence in the warnings. This leads to much more targeted investigations, giving engineers the tools to understand quickly and more accurately what is happening with their equipment.
How Does It Work In Practice?
To give an example, UptimeAI’s AI Plant Expert solution (a plant operation application especially for the power industry) leverages an AI engine that continuously learns from historic and ongoing data and identifies how each of the parameters involved in the system change in relation to each other. From this data, it can continually read current new data and predict an expected value based on other parameters. It then compares this predicted value against the actual data and determines any discrepancies, creating an anomaly score that indicates the overall health of the system.
Essentially, the technology can analyse huge amounts of data, allowing plant operators to see the system as a whole and identify the impacts of different parts of the system (e.g. the pieces of equipment within it), the external forces (e.g. environmental factors), and how all of these elements interlink with each other. On top of this, there are no rules to define and manage as the AI engine can develop its own understanding of what is significant and what is due to external influence. It’s also capable of continuously learning from new experiences and can recognise new types of events – effectively learning in the same way in which a human engineer learns.
There are numerous benefits of using an AI engine such as this. Not only can it identify and diagnose problems (such as equipment faults and inefficiencies within the system), it can also make recommendations on how to resolve them and prevent them from re-occurring in the future. It can also simultaneously predict how well each piece of equipment in the system will work and flag up reliability, efficiency and product quality problems before they happen – potentially saving companies millions of pounds in lost revenue.
Has This Been Done Before?
UptimeAI was recently asked to check the efficiency of a condensing steam turbine in a 120MW power plant to determine any lost efficiency and improvements that could be made. Even relatively small losses of inefficiency in such a system can make a significant difference to output and profitability. A 0.007 bar of excess backpressure on a base-loaded 500MW plant condenser correlates to a loss of 1.7MW of power and nearly US$500,000 per year lost revenue.
The AI Plant Expert software was fed with historical data for the entire turbine, condenser and the cooling water circuit as one system. The turbine system under review was heavily influenced by a large variance in cooling water temperature due to seasonal and daily environmental changes. Teasing out the effects of large seasonal variations and operating changes to see the underlying causes is beyond human analysis.
Feeding historical data through the AI application revealed a continued upward trend, including three targeted alarms over 36 months, indicating points of significant change.
The alarms were generated, taking into account the impacts of turbine exhaust, seasonality and load fluctuations. By analysing where the anomalies were most significant, the AI diagnosis tool could make predictions on the likely failure mechanisms and prescriptive recommendations using the application’s built-in engineering knowledge. This was able to diagnose specific cooling water issues that were affecting the efficiency of the condenser and hence the turbine.
What Are The Benefits?
The UptimeAI application was able to identify a total improvement opportunity of 0.016 bar of condenser vacuum. This was made up of cooling water tower/return inefficiency (circa 25%), cooling water discharge pump low pressure (circa 24%), condenser fouling (circa 32%) and air ingress (circa 19%). This equated to a 2.2% improvement opportunity in backpressure worth an estimated £140,000 per year in increased efficiency, generating an ROI of 250%. The project was completed in three weeks and the insights were generated in a matter of hours once the data was ingested into the application.
In this example, the system identified problems that weren’t previously detected because the alarms for these systems were necessarily set sufficiently high to prevent continuous false alarms. Without the UptimeAI application, these problems would only have been determined with a lot of investigation effort and, even then, only in a crude high-level manner.
For high-accuracy equipment monitoring, the critical differences that indicate reliability and inefficiencies are often hidden within the operational and environmental fluctuations. Only by monitoring pieces of equipment within the context of their entire system can these signals be detected. The latest improvements in AI technology have been developed specifically to address exactly this problem.
Andrew Normand is UptimeAI partnership lead for Encora Energy
