Restore synchronized operation!
The trains leaving Halle traverse the mining area of Eastern Germany, passing the smoking stacks of power stations that are still coal-fired. For Prof. Dr. Michael Schulz the mighty cooling towers are a familiar sight from his office at Indalyz Monitoring & Prognostics (IM&P) GmbH.7 Sep 2018
He and his team are working on the region’s structural strategy and its energy transformation, in which artificial intelligence plays an important part. “Our goal is to develop an energy generation plant that operates autonomously, identifies damage and can run in critical situations, excluding the possibility of standstills or further damage. In short, we want an operating plant that makes its own decisions on how to react in the event of potential damage.”
Schulz says that the company monitors “everything that is powered mechanically” in gas, water and wind turbines by using artificial intelligence. He adds: “Despite sophisticated sensor technology it would be impossible for us to draw as many conclusions as we can today without AI. One person could never analyze so much data.” Schulz estimates that around 600,000 measurements per sensor, per minute are produced in a wind power station. Sound data, for example, can be collected and aggregated via a decentralized field bus or PLC in cycles of a few seconds up to several hours, depending on the protocol.
This data is then fed to an IM&P server. A genetic algorithm with weak AI establishes recursive neural networks which compare the data with images of damage or report threshold levels which have been exceeded. This is no simple task due to the volume of technical details to monitor across the entire drive train of the wind energy plant, as well as the fluctuations in wind-generated energy (for example, Central Europe typically experiences strong fluctuations in wind speeds), the connection of wind energy plant to the central grid but also the whole process control, which means that it is necessary to compensate for the sometimes strong variations in the measurement data.
Schulz adds: “Often we do not have an ideal damage hotspot. Even the sum of several small faults can cause a system shutdown, and these tiny defects are frequently overlooked by human operators. However, the dynamic of non-linear, mechanical three-body problems normally exhibits chaotic elements as well.” In effect, the movement of all degrees of freedom of a machine would have to be monitored by sensors, and for commercial reasons alone this is not an option. The important thing is the ability to detect physical signals at their precise location through measurements taken at a few points across the interaction of the individual components.
The realization of this task with AI methods demands, in particular, knowledge of the machine and its operations. Schulz is critical of the way AI is discussed in public: “The idea of running AI through a machine controlling operating data as if it were a simple matter of turning a switch on and off is just media hype. The command ‘Machine, restore synchronized operation’ is infinitely more difficult than the commands we know from our smartphones.”
Returning to the wind power plant: Neuronal networks learn from every incident of damage or from every threshold level exceeded. They work with historic data, but also reference data from other wind plant in the energy park. Currently, a maintenance team will be called out to a wind energy plant in the event of any damage or any deviation in a measurement. “That costs a lot of money. We want machines that will make decisions before a fault causes a standstill.” Schulz calculates that a three megawatt plant incurs annual operation and maintenance costs of around €100,000. Much of this is due to standstills and natural wear and tear. Two thirds of these costs, however, are due to primary damage that previously could not be identified. “Over a two-year period an operator could save €25,000 per plant. Imagine how much that would come to for a wind park – that’s not even including the standstills.”
The energy sector trusts Schulz’s algorithms. Server data from Finland, Denmark, South Africa and Turkey is processed in Halle. “The whole energy industry invests in AI, at the moment especially in weak AI, which makes it easier to monitor the process”, says Schulz, whose algorithms have proven their worth in controlling district heating networks. However, as he says: “That’s another story.”
Interested in news about exhibitors, top offers and trends in the industry?
Your web browser is outdated. Update your browser for more security, speed and optimal presentation of this page.Update Browser