I studied mathematics. What really fascinates me is how numbers mirror the world, like how physical products or machines can be represented in a data space through numbers.
It’s quite amazing. The numbers stand for what was recorded by microphones, cameras and controllers in the real world. If you know how to read the numbers, you can get detailed information on any one of the different components in the factory – starting from an individual drive all the way to the entire system.
That data serves as the foundation for AI applications, which is what I focus on in my work. A subset of AI is machine learning, or ML, where systems have the ability to automatically learn from experience (which is data) without being explicitly programmed.
Of course, people still play a role, like coming up with the initial design and telling the ML system what it needs to do. (This is why one type of ML is also referred to as supervised ML.) At Siemens, we also integrate our deep expertise in industrial settings – which is something that sets us apart from the competition.
Big savings with AI and ML
Our goal is to be the trusted partner and end-to-end provider of industrial-grade AI and ML. Proof that we’re achieving that can be found in real-world factories where our ML solutions are making a big difference. An example is in one of Siemens’ own electronics manufacturing plants. An AI system there predicts the probability of pseudo-errors for in-circuit testing. With this information, we know if we need to do costly retesting or if the printed circuit boards are good to be shipped.
This has cut efforts associated with pseudo errors (like manual inspection and rework) by up to 50 percent, which amounts to six-digit euro savings every year.
Too good to be true?
As well as ML does work, there is a downside: when training ML models, we typically rely on static historical extracts of generally dynamic data. However, once models are implemented, the data may change rapidly and even in an unknown and undetected manner. So over time, the performance and reliability of ML models may decrease. That means the data no longer mirrors the world as it is. Not a good thing if you’re relying on ML for sensitive processes, like end-of-line quality testing.
We at Siemens develop and provide software, services and products to keep ML consistently dependable. An example is through scalable systems that indirectly assesses model validity independent of the specific model architecture, implementation environment or feature space. You can find out more about our work in this white paper on reliable ML operations in industrial environments.
It’s all about keeping AI and ML reliable. With a system like ours in place, you can trust what the models are saying; you can be certain that they are an accurate reflection of what is happening. Services like these are one way that Siemens builds trust in AI. And for me, there’s nothing more satisfying than knowing that numbers and data in the virtual world are a true reflection of what’s going on in the physical world.
Find out more under: https://www.siemens.com/digital-enterprise-services