In the rapidly advancing era of digitalization in industry, automation and AI play an indispensable role in increasing efficiency and optimizing industrial production processes.

A pioneering development in this context is the Blueprint for AI-based automation from Siemens, which sets new standards as a pioneering tool and method kit for tasks such as automated sorting and quality assurance. The Blueprint is the answer to the question of how Siemens will shape the future of production.

The core of the Blueprint lies in its ability not only to move objects mechanically, but also to recognize and classify them intelligently. Using advanced sensor technologies, the Blueprint identifies objects and makes real-time decisions about their destination and other characteristics such as quality or material properties. This cutting-edge combination of sensors and machine learning marks the transition from traditional automation to intelligent, data-driven decision-making.

What particularly distinguishes the Blueprint for AI-based automation is its flexibility and adaptability to a wide range of use cases. From the detection of defective parts and quality control in food production to specialized applications such as recycling – the Blueprint offers solutions for challenges in numerous industries.

The modular system is made up of three components: a lifecycle platform for AI base models to support AI engineering, a runtime and automation integration solution based on the Siemens Edge platform, i.e. a SIMATIC industrial PC from Siemens with its own operating system, and an automation solution (such as a delta picker robot), which is controlled using the SIMATIC Robot Library and can pick up objects using a gripper.

© Siemens

Foundation models are large neural networks that have been trained with huge data sets and are fundamentally changing the approach to AI projects in the industry. AI projects are traditionally based on the CRISP-DM model (“Cross-Industry Standard Process for Data Mining”), which was developed at the end of the 1990s as a universal framework for data-driven projects. The greatest effort is spent on data preparation (40-50 percent), while business and domain understanding take up around 25-40 percent. Modeling is efficient with well-prepared data (10-20 percent), while evaluation and deployment require less time (5-10 percent each). A key problem for the scaling of AI lies in the complexity of industrial applications and the frequent lack of availability of high-quality data for specialized models.

Base models offer a new approach by providing large, pre-trained AI models that are suitable for general industrial tasks and can be flexibly adapted to different applications. Through fine-tuning, these foundation models can be adapted to specific industrial use cases without having to train new models from scratch each time. This saves time, reduces costs and enables rapid scaling, as it is based on a strong model that can be adapted to different industries and machines. “Because foundation models are pre-trained on very large data sets, they can be used to solve new use cases with minimal effort,” says Claudia Holzgethan from the Distributed AI Systems research group at Siemens Austria.

Because foundation models are pre-trained on very large data sets, they can be used to solve new use cases with minimal effort.

Claudia Holzgethan, Distributed AI Systems Research Group, Siemens Austria

© Siemens

Foundation Models are characterized by “vision capabilities”, especially with regard to the identification of physical objects and defects. These capabilities can be used in applications such as robotics and in an autonomous factory. Although foundation models are pre-trained, they can still learn from data input or prompts during runtime.

Here are some examples of the applications of foundation models in industry:

PCB error detection in electronics production: PCBs (printed circuit boards) are central to electronic devices; even small manufacturing errors can cause device failures. Typical problems are defective conductor tracks, soldering errors, misaligned components or missing parts. A foundation model for the electronics industry can automatically detect defects during the assembly process.

Automated sorting in recycling: Effective recycling requires the precise sorting of materials (plastic, metal, glass, etc.) and material flows, as well as the detection of hazardous substances such as lead or mercury and contaminants such as long parts. Precise identification and separation are crucial. A visual foundation model can be efficiently adapted to the needs of the plant and the material flow and thus reliably classify different materials.

Battery disassembly in the automotive context: Lithium-ion battery packs, especially from electric vehicles, are complex and pose explosion risks. Safe and efficient dismantling is crucial. The model recognizes various components (cells, modules, etc.) and damage in order to safely support disassembly using a robotics solution.

© Siemens

Lithium-ion battery packs are complex and pose explosion risks: The AI model
recognizes various components and supports safe disassembly.

Despite the enormous potential of industrial AI, companies often face major hurdles when integrating it into existing systems. The Blueprint from Siemens solves this problem. Thanks to its standardized infrastructure, the Blueprint makes it easy to implement and scale AI solutions that are both easy to maintain and fully compatible with industrial processes. The Blueprint is thus positioning itself as a key technology for companies that want to take the leap into the future of intelligent production.

In addition to optimizing production and sorting, the foundation model for visual understanding also facilitates the provision of edge-capable AI models. Siemens Xcelerator enables centralized management of AI models and, with its user-friendly interface, enables even non-AI experts to test AI models and validate use cases with little effort and risk. “A foundation model for visual understanding is to be offered via the Siemens Xcelerator marketplace in the future, making it easier to get started with industrial-grade AI,” notes Daniel Schall, Head of the Distributed AI Systems research group at Siemens Austria.

A foundation model for visual understanding is to be offered via the Siemens Xcelerator marketplace in future, making it easier to get started with industrial-grade AI.

Daniel Schall, Head of the Distributed AI Systems research group, Siemens Austria

© Siemens

The Siemens Blueprint embodies the vision of an intelligent, flexible production environment in which AI-based base models play a key role. By constantly adapting to changing conditions, Siemens is setting new standards in efficiency and flexibility for production and sorting processes. These multimodal models, which integrate data from different sources, enable companies to optimize their processes intelligently – putting them in an ideal position to successfully meet future challenges. “An AI is only as good as its integration into the overall solution. That’s why we offer an integration framework to make optimal use of AI with automation technologies,” says Lukas Gerhold, Head of SIMATIC Application Center, Siemens Austria.

An AI is only as good as its integration into the overall solution. That is why we offer an integration framework to make optimum use of AI with automation technologies.

Lukas Gerhold, Head of SIMATIC Application Center, Siemens Austria

© Siemens