Just another AI day
Lots of artificial intelligence (AI): We’re accompanying our Service Manager Aina through her workday. Even though we’re not usually aware of it, artificial intelligence is already helping us by making our lives more comfortable, safe, and sustainable – a research overview.
Good morning, it’s time for the first cup of coffee of the day. While Aina enjoys her breakfast and her mind starts to focus (seven o’clock isn’t her best time) artificial intelligence is already on the job at the power plant delivering electricity to the coffeemaker.
Gas turbines at the power plant generate electric current. Certain settings, like the one that distributes natural gas to different zones, require continuous readjustment during operation to keep turbine emissions low. Because of the complex relationships and interactions between the many different parameter settings, it’s extremely difficult for humans – even experienced experts – to determine the ideal configuration for specific weather conditions. On the other hand, artificial intelligence – in this case, the control policy comprising a neural network (NN) – is extremely successful at it. With the help of reinforcement learning – a method that Aina also used, for example, when she learned to ride a bike – AI learned an optimal control strategy for the turbines based on past operational data. Gradually each turbine finds its optimal setting, and the benefit for gas turbines is clear: In actual practice, nitrous oxide emissions can be reduced by 10 to 20 percent. For many power plant operators, this is reason enough to invest in AI, especially since existing turbines can be retrofitted.(example project)
Aina works in the city center, but she lives on the outskirts. Because she has an appointment, she takes the car instead of riding her bike. On the way she encounters the usual morning traffic ☹ – but thanks to her navigation system, she still makes good time and only has to stop at one light.
Today navigation systems in cars are standard, including those that factor in the current traffic situation when planning the route. Artificial intelligence provides even more convenience in the form of a neural network that has learned to predict the switching behavior of a traffic light whose timing is also dependent on the time of day, traffic conditions, and vehicle density on the road.
As a result, the vehicle “knows” when it has to stop at a light and how many seconds it will take the light to change to green, so it can automatically decide whether it’s worthwhile to shut off the engine while waiting. When equipped with a GPS system, the vehicle can recommend an optimal speed for hitting all the green lights, or it can suggest faster routes more reliably than ever before. The system has already been successfully piloted in Wolfsburg, Düsseldorf, Hamburg, Berlin, and Vienna. Siemens and Volkswagen are collaborating on its ongoing development. (Learn more about AI for better mobility)
Today Aina has a doctor’s appointment before work. A week ago during a routine exam, her primary physician noticed irregularities in her blood panel that can often be the first sign of a serious illness. Very scary! That’s why she’s now going to see Dr. AIpro, a specialist who can evaluate Aina’s condition more precisely using an AI-supported diagnostic system. His findings: Everything’s okay, there’s no need to worry. Aina is so happy and relieved that she wishes she could hug the doctor and his AI.
In medical diagnostics, physicians often have to analyze image material like blood samples, histological sections, and CT and MRT images. In these cases, neural networks can offer especially valuable support provided that they’ve previously learned to recognize what’s normal and abnormal based on several thousand sets of training data. Patients benefit and diagnoses are more precise because Dr. AIpro can concentrate on all the potentially problematic areas of the blood panel. But it isn’t enough for an AI system to supply a diagnosis without also communicating how the diagnosis was reached. Doctors can’t prescribe treatments based on results from AI without being able to trace their logic. We can and must expect AI’s decisions to be traceable, not just in medicine but wherever decisions can have serious consequences (also see the EU ethics guidelines). Experts call this explainable AI, or XAI.
This means that humans have the ultimate decision-making responsibility, while AI systems act as subordinate assistants that handle the tedious data-related tasks. To optimize collaboration between humans and AI systems, system developers are focusing their efforts on providing what are known as digital companions. These are systems that have been designed to work extremely well with humans.
Aina finally arrives at the office – and not a moment too soon, because there’s already an important warning from the AI service system flashing at the top of her mailbox. As a Service Manager, Aina is responsible for the smooth operation of pumps used for oil extraction. Today one of the pumps appears to have a major problem and is on the brink of failure. This is a scenario that Aina must avoid, whatever it takes, and so she has a stressful day ahead of her. The exact cause of the problem has to be found, decisions made, technicians given detailed instructions, and management notified. But by the end of the day, she’s managed to do it all. The pump was repaired with no downtime required.
The failure of large technological systems like pumps, production machines, or trains is often extremely expensive for operators, so it’s in their best interest to prevent failures at any cost. What works in their favor is that, theoretically, measurable anomalies are already apparent hours or even several days beforehand. Unfortunately, even with the right sensors and optimally defined limit values, conventional monitoring systems can’t detect all failures in advance – because they can be detected only if all the sensors are working together properly, and also because the data patterns that identify them differ from device to device. But with the appropriate sensors – for example, to measure pressure and temperature – and artificial intelligence to correctly interpret the various machine states, these patterns can be recognized before there’s an actual failure. Experts call this predictive maintenance.
When a problem is detected, the person in charge has to find a solution. This usually requires expert knowledge to determine, for example, how quickly to react or what components need to be replaced. Experience with similar cases is helpful, as is knowledge of where and how quickly replacement components can be obtained. All the most important information is generally stored in various distributed data sources, but finding it can be extremely time-consuming. Here again, artificial intelligence can offer valuable support, provided it has learned how different data sources are related in terms of content. The most well-known AI-compatible technique for displaying knowledge and relationships is the Industrial Knowledge Graph. Using AI, Service Managers can avoid a lot of tedious searching (a use case from Mobility).
It’s finally quitting time! That was enough stress for one day. Aina is happy to get home shortly after six o’clock. Now she’s looking forward to a refreshing shower, followed by a glass of wine in front of the TV. She doesn’t need artificial intelligence to do that, does she? Wrong!
While Aina relaxes, AI at the water utility is on the job protecting the drinking water supply. Water is a vital necessity, and yet a large portion of our drinking water trickles into the ground through leaks. In regions where water is scarce, water theft is also a frequent problem. And there are other risks, like impurities. Artificial intelligence can assist by monitoring systems and pipes and immediately issuing a warning if irregularities are detected.
Subscribe to our Newsletter
Stay up to date at all times: everything you need to know about electrification, automation, and digitalization.