Recent years has been tough on professional gamers, having been challenged – and defeated – by more and more AI-based programs in their master game.
Besides the art of designing such AI-based players, computer games offer some specific characteristics that benefit the training process:
- Access to almost infinite training data, since it can be computer generated
- You can hardly break anything, since you are experimenting in a 100% artificial environment and
- A clear definition of winning, since you either win or lose the game based on the defined rules of the game designers.
When developing industrial AI or any application that relates on real-world data, the above characteristics rarely hold.
When we at Siemens Tech develop AI-based applications for real-world systems, we usually have to literally start at the game design itself.
Here is why:
The moment you work with real-world systems, the system dynamics are either hidden in the data or in best cases approximated by existing simulations or mathematical descriptions. Capturing real-world data is costly, and existing data sets are often insufficient and come with high uncertainty.
Imagine the sole challenge of collecting data from an offshore wind turbine from a limited number of sensors, that might have been originally installed for fault monitoring only.
Modeling the system dynamics, in that case the dynamic of the wind turbine in interplay with the highly stochastic wind that remains rather unknown over the entire swept rotor area, defines a highly complex task to accomplish.
At Siemens Tech, we have spent over 30+ years to develop tools and methods, combining AI and the arts of engineering, to be able to model real-world systems digitally despite these adverse conditions.
Once we have crafted such digital representation, the definition of ‘winning’ is the most complex and substantial next step.
One fundamental, inviolable element thereby is to assure save and sustainable development and operation along the complete lifecycle of our solutions at all time. Additional success criteria are developed in close cooperation with the domain and business experts.
For wind turbine operation, one evident target is the generation of a maximum power output with no compromises on lifespan and safety with the domain-specific addition, that only transparent, and interpretable solutions are accepted.
With the developed understanding of a target function, the digital ‘game’ can be setup and the training process for an AI-based solution can be started. Leveraging research competencies across various data analytics, AI and engineering fields, the development of a reliable, robust and save AI system becomes a true multiplayer effort.
And it doesn’t stop there since industrial AI solutions do not only have to safely perform in our testing environments, but in real-time operation often simultaneously across various locations around the globe. The competence of safely deploying and operating such solutions in real-time is just one competence embedded in more than 500 patents we have been able to develop with our team of 250+ engineers, innovators and research scientist located at 7 different locations around the world. Standing in line with some of the biggest IT giants when it comes to the number of patent applicants by number of patent families in AI [WIPO, 2019].
Leveraging our research in artificial intelligence in the arts of engineering, we aim to create a safe and sustainable future, of which self-optimizing wind turbines are only one example.
If you would like to learn out more about our research or our AI services, please be invited to get in touch, directly or via our AI Lab.