The vision of power plants operating autonomously is about to become reality. Already plants can operate several days unsupervised. Now digital twins, Big Data and AI are taking them to a new level with smart forecasts and instructions.
Jan Weustink has set his sights on a future where sensors, autonomous robots, digital twins, smart analyses, and AI ensure smooth and autonomous power plant operation. As an expert on simulations and digital twins, he develops strategies and technologies for Siemens Energy to make the vision of an autopilot for gas and steam power plants a reality.
An increasing number of power plant operators are already signaling an urgent need – and Arik Ott, who coordinates the Autonomous Operations portfolio at Siemens Energy, is fielding more and more requests for appropriate solutions. “The hope is to be able to meet several of the current challenges at once,” Ott explains. The growing percentage of renewable energy sources in the grid is steadily increasing the pressure on gas and steam power plants to operate more flexibly and efficiently. A massive shortage of skilled control room and maintenance personnel is also looming on the horizon.
“The hoped-for solutions are closer than many people think,” says Weustink. “Autonomy doesn’t start with the autopilot. Long before that, power plants are supported by smart analyses, smart forecasts, smart recommendations, and smart instructions. Knowledge graphs connected to software agents, which provide a machine-usable description of a plant similar to a dynamic Wikipedia and make the entire functional chain of a defect comprehensible, and AI-supported fault analyses which identify the cause in real time are both steps toward autonomy that will very soon provide power plants with much more flexibility and efficiency.”
Siemens Energy has already taken steps in this direction. Ott and his colleagues have developed a solution that relieves power plant personnel of their daily inspection rounds. They no longer have to stand by on-site ready to search for leaks, check operating values, and investigate unusual noises. This task is performed by AI-supported analysis algorithms that regularly filter out signs of irregularities from the data supplied by cameras, microphones, and other sensors mounted on the plant or installed on robots, and that request support when needed. Off-site maintenance and control room personnel can support several power plants simultaneously.
“This solution not only offers power plant operators the option to reduce their inspection round efforts,” says Ott. “We’ve also paved the way for an important step toward autonomous power plants. If virtual inspection rounds are coupled with a power plant’s I&C system, and we can also ensure that necessary materials like resins and lubricants require only occasional replenishment or monitoring, then several days of unsupervised standard operation becomes a possibility. In the European Union, this move toward autonomy is already possible for a three-day period.”
For many power plants, multi-day unsupervised operation already represents an attractive degree of autonomy. “In the case of the former base-load power plants that are now kept on hand to provide a fast startup when needed, it’s in their best interest to reduce their effort during standby mode as much as possible,” says Ott. “However, unsupervised operation is also valuable for power plants that run at full capacity over longer periods of time – including those that need to generate district heating in winter or operate desalination plants for drinking water treatment.”
Stefan Lichtenberger, who manages Siemens Energy‘s portfolio of data-based services, is already developing the next level of autonomy with his team. This involves also supporting power plant components outside the turbines with smart forecasts and recommendations. “We do this by performing operational analyses in which AI readjusts the digital twins of the components based on actual sensor data, with the result that it’s able to distinguish between normal aging and extraordinary events,” explains Lichtenberger. “This allows normal aging to be automatically taken into account in analyses, making forecasts and recommendations even more precise. Power plants’ capacity utilization can be improved even further.” A pilot of the development is currently being tested on a heat recovery steam generator.
To do this, Weustink can use the same digital twins and algorithms of simulators that Siemens Energy employs for training power plant personnel. “But unlike people, it’s hard to teach AI about an entire power plant all at once,” says Weustink. “It’s easier if you divide the power plant into sections that are controlled separately by networked AI.” To prepare this solution, he has already started using knowledge graphs to make the individual sections machine-usable. Taking these as his foundation, he wants to develop a generic solution for a central power plant component that can then be applied to the power plant as a whole.
For Jan Weustink, there’s no question that the first autonomous plants connected to the grid are coming soon. “The technologies necessary for actual applications are here,” he says.