Accurate forecasts of operating figures are a boon to any industrial enterprise, enabling them to make precise production estimates and avoid extra costs. Siemens has long used artificial intelligence (AI) to create forecasts and provide targeted support to its departments. In the past, however, the development process ate up a lot of time. To boost the efficiency and quality of these kinds of AI models, Siemens IT developed a platform based on Microsoft Azure that functions as a customizable template for the entire company. Developing and deploying AI models used to take months. With the platform [ai:attack], it can now be done in just a couple of weeks.
The challenge: Establish a standard for AI models
AI-driven digital solutions can do so much: They easily read out metadata in PDF documents and use it to make forecasts along time series. They identify seasonalities, such as increased demand for products around Christmastime, thus simplifying production and sales planning. Using automated image analysis, they detect any product damage, making returns management easier. AI automatically extracts all relevant data from incoming invoices and feeds it into the company’s finance systems, speeding up processing. In short, AI makes processes more efficient and relieves employees of monotonous, time-consuming tasks. But optimizing the way these kinds of solutions are developed hinges on the underlying process being as straightforward as possible. Ideally, not a lot of time should go by between idea and solution rollout.
Siemens IT was already using various AI solutions and had already established defined processes for developing them, but they wanted to become even better at this. Previously, when departments had come up with ideas for solutions that could help them speed up business processes, they first had to explain these ideas in a way the data analysts could understand, and convey their specialized domain knowledge. The departments, in turn, didn’t understand how AI works, and the data analysts didn’t have the right, user-friendly tools to realize the departments’ ideas. There was no standardized language, no standardized code, and there were no centralized, reusable components with which to build AI models. Dr. Ioannis Petrakis, IT Principal Key Expert Data Analytics & AI at Siemens IT, understood this challenge very well: “With every new AI model, we were always reinventing the wheel. We repeated many things in the development process that really wouldn’t have been necessary—configurations, for instance. We didn’t use components. Instead, we wrote thousands of lines of code from scratch every time.”
It often took several months to go from idea to finished AI model—much too long when your aim is to provide important solutions quickly. Petrakis and his team wanted to develop a platform that would offer data analysts a wide array of standardized AI services. They also wanted to create an environment in which experts in, say, the finance or logistics department could try their hand at data analytics and AI to build up their knowledge and experience and gain a better understanding of artificial intelligence. In 2021, the Siemens initiative [ai:attack]—a platform designed for developing AI models and fully based on Microsoft Azure—went live.
The solution: An AI platform with reusable components
[ai:attack] speeds up the deployment of AI models enormously: it used to take several months for AI solutions to go from idea to rollout. Today, ideas are translated into finished solutions in just a few weeks. At the core of [ai:attack] is Azure Machine Learning, a service that facilitates a standardized, uniform process from the moment an idea is born until a new AI model goes live.“The automated training of machine learning models in Azure was particularly important to us: this makes it possible to create AI models automatically. As a result, we are a lot faster and have cut our development time down from several months to a few weeks.”
Dr. Ioannis Petrakis, IT Principal Key Expert Data Analytics & AI, Siemens IT
Siemens is able to use the various Azure functions when developing its models, enabling the company to develop a variety of models without having to write new code for each one. “Demand is particularly high for AI models that automatically extract data from documents and images,” Petrakis says. Azure Cognitive Services provide the components for this within the platform. Text analytics enables data in PDF documents—such as order forms—to be read out automatically.
And Computer Vision lets AI models identify, say, damaged devices in images. In addition, the user-friendly work environment of Azure Machine Learning Studio helps data analysts keep tabs on all the model development workflows.
“We rely a great deal on reusable assets when developing AI models. The components we use have been well tested and are available to our data analysts for quick adaptation. And thanks to Microsoft Azure, everyone can directly access an infrastructure that lets them jump right into their development project.”
Dr. Ioannis Petrakis, IT Principal Key Expert Data Analytics & AI, Siemens IT
Thanks to this modular development approach, the days of writing thousands of lines of code for each new model are over. This increases both the efficiency of the process and the quality of the code. “It’s not like we develop a model and then it just remains productive,” Petrakis says. “Code has to be maintained and enhanced. The uniform standard we developed for all code in Siemens AI models saves us a lot of work.” Before, Siemens was in need of three data analysts to maintain the code for three projects. Today, this task can be done by a single individual.
“We can now deploy compute clusters in Azure in just a few minutes. That wasn’t possible before.”
Dr. Ioannis Petrakis, IT Principal Key Expert Data Analytics & AI, Siemens IT
[ai:attack] serves both as an enabler for employees in the departments and as a tool for the data analysts. Some 500 employees have already received training in using the platform and in the basics of data analytics, with Siemens IT placing a strong emphasis on IT security and compliance. “Of course we educate our employees about using data and AI responsibly. But we also integrated these standards into our platform,” Petrakis says. “The platform itself automatically prevents actions and code that deviate from our standards. We also set great store by the principle of responsible AI, which means we train our AI models exclusively with non-discriminatory data that assesses scenarios strictly without prejudice.”
“It used to take us months to complete development of an AI model. Thanks to Microsoft Azure and the AI platform [ai:attack] based on it, today it takes us no more than a couple of weeks. Ideas for projects often come straight from the various departments. The AI models for implementing them are then immediately available in real time.”
Dr. Ioannis Petrakis, IT Principal Key Expert Data Analytics & AI, Siemens IT
Follow Microsoft