Power plant inspection: how physics-based AI can can help operators

The pressures on critical infrastructure are changing across the world. Demand is both rising and subject to enormous fluctuations. Supply models are completely different. Assets are degrading unpredictably. Expertise is being lost to retirement. Andy Morris explains how combining the power of physics led AI with operational data is adding real-time intelligence to support a more targeted inspection, repair and replacement cycle

The impact of the recent heat wave on the national grid underlines the challenges facing critical infrastructure operators. Electricity prices rose sharply across European markets as the heatwave gripped, pushing demand higher.  Neso – which manages the energy systems in England, Scotland and Wales – warned at one point that an extra 1,900MW of power would be required to avoid falling short of the electricity required to power homes and businesses within its normal safety margins. Similar spikes in demand have also been linked to extreme cold weather events, as seen in North America this year. With extreme heat also blamed for a string of power plant outages, including one in France affecting 68,000 homes, the unpredictability of weather patterns is clearly adding further stress to the power network.

While such warnings are usually associated with cold weather, the spike in demand created by a heat wave is just one problem facing operators. The increase in data centres and Electric Vehicle charging, is rapidly increasing demand. Yet with a lack of spare capacity, critical infrastructure is under unprecedented pressure and unplanned downtime is not an option. 

Yet for operators, traditional models of maintenance and replacements are also being challenged both by the changing model of power generation and the rapid loss of expertise linked to an ageing and retiring workforce. The operating conditions associated with repeated cycling between power sources, including solar, wind and gas, is leading to accelerated asset degradation and potential failure. The repeated start and stop rotation required to deliver grid stability is fast-tracking asset stress, fatigue and creep, challenging traditional patterns of condition monitoring, replacement and repair.

Unknown asset performance

While asset degradation raises obvious risks to safety and operational efficiency, the biggest challenge for operators is the unknown rates of degradation. For decades, the traditional periodic condition monitoring cycle has worked. Small data subsets used for simulations have provided an accurate picture of performance in plants operating at a steady state with minimal cycling. This model is now under enormous stress. 

Simulations cannot be run across the extended e operational datasets required to accurately represent flexibly operating assets and provide vital insight into degradation. Furthermore, the workforce has aged alongside the infrastructure, with industry facing massive loss of expertise and knowledge as individuals retire. Not only are patterns of asset life changing but without skilled individuals to highlight potential issues, the industry faces a serious problem. The lack of real-time visibility into asset status throughout critical national infrastructure is becoming a very real concern.

While periodic physical condition monitoring remains important, a new approach to condition monitoring is providing essential real-time visibility into asset condition. Physics-based, AI-led condition monitoring combines thermodynamics, material science and historical infrastructure operational data with machine learning to understand how industrial assets are now behaving in a world of frequent grid cycling.

Real time condition monitoring

Real-time insight into asset integrity, stress, creep, fatigue and life consumption provides operators with vital information to support repair and replacement strategies. Accurate information about asset life allows operators to optimise the eight-week window typically taken for a major inspection, repair and replacement outage. It also avoids needless repairs and replacements, minimising the whole life cost of assets and ensures manual inspection activity is focused on identified areas of greatest need.  Furthermore, by combining this real-time conditioning monitoring information across more than one power plant, operators can quickly gain insight into the new trends in asset degradation across the entire operating environment.

This information will provide vital support to engineers throughout the operating cycle and give clear direction when the periodic inspection process is undertaken. In an industry wrestling with an ageing engineering resource, the traditional opportunities to learn how to recognise asset degradation such as metal fatigue from older, more experienced colleagues may be harder to achieve. 

Physics-led AI provides a great opportunity to capture decades of engineering expertise. It supports younger, less experienced staff to focus their inspection activity and understand what they are seeing.  And it augments decision making by providing vital clarification into plant performance in real-time. 

Conclusion

The periodic condition monitoring inspection cycle remains core to power plant operations globally but with fast evolving pressures on critical infrastructure operators need to quickly understand trends in degradation. Enhancing this incremental visual process with real-time, engineering grade information, will allow operators to understand, optimise and, critically, support the next generation of power plant engineering talent.

Andy Morris is Co-Founder, MatAlytics.