Machinery in the manufacturing industry is a major energy consumer. The energy consumption of these machines is currently hardly optimized because it depends on many factors. Artificial intelligence can help to develop models that automatically adjust the parameters of machines so that as little energy as possible is consumed.

For the metal processing application, the Alpen-Adria University of Klagenfurt, voestalpine High Performance Metals DIGITAL SOLUTIONS GmbH, Siemens and the Belgian KU Leuven are working together on the FFG-funded SAELING research project to reduce energy consumption in the metal processing industry. The project name SAELING stands for “SAving Energy by Learning and ImproviNG logic-based optimization models”. The aim is therefore to use machine learning methods to develop prediction models and energy- and resource-saving strategies for machines. These models are the core of the research project and address the aforementioned problem that the energy optimization of machines depends on many criteria and must take many different objectives into account at the same time. The methods of machine learning are therefore a building block for solving a so-called multi-criteria optimization problem.

Voestalpine High Performance Metals DIGITAL SOLUTIONS GmbH is a company in the voestalpine Group’s High Performance Metals Division that implements digitalization and innovation projects for the division. The division is the global market leader for tool steel and other special steels. It focuses on technologically sophisticated high-performance materials, which are manufactured at eight production sites in Europe, North and South America and distributed via a sales and service network with 130 locations in 40 countries. In the global value-added services areas of this voestalpine division, the sawing process is one of the core processes in the manufacturing environment.

© voestalpine

Using methods and algorithms for optimized decision-making, energy efficiency in the voestalpine application is to be increased by up to 30 %.

The sawing process is carried out using band saws. Their energy consumption is the focus of the research project. The band saws are used to cut large pieces of steel into smaller pieces according to customer requirements. In the future, energy optimization will be extended to other metalworking machines for grinding and milling. A typical bandsaw consumes around 8.4 MWh per year. The High Performance Metals division has more than 2,000 bandsawing, grinding and milling machines, which consume a total of more than 21 GWh per year. This is roughly equivalent to the annual energy consumption of 4,750 average Austrian households.

Focus on the energy consumption of band saws

Metal processing machines handle a stream of tasks. A workstation with band saws, for example, consists of a series of machines on which a job can be assigned to different band saws. The energy consumption of a task depends on the characteristics of a machine, for example on the wear situation, but also on the material, such as its shape, volume and quality. In addition, the energy consumption when processing a task is determined by controllable parameters such as the speed of the band saw and the material feed.

“Assuming that a sufficiently accurate physical model of the machine is available, minimizing energy consumption would be a multi-criteria optimization problem in which the tasks are assigned to the machines, the sequence of tasks on the machines is determined and the parameters of the machines are set so that the energy consumption is as low as possible,” explains project manager Richard Comploi-Taupe from the research department at Siemens Austria. “However, since humans are not able to specify sufficiently accurate physical models with reasonable effort, we want to generate machine learning methods in our research project and combine them with symbolic optimization methods to solve scheduling problems such as in our use case in the metal industry. Scheduling means deciding which production steps are carried out when on which machine,” continues Comploi-Taupe.

In the SAELING research project, we want to generate machine learning methods and combine them with symbolic optimization methods in order to solve scheduling problems such as in our use case in the metal industry.

Project Manager Richard Comploi-Taupe, Research Department Siemens Austria

© Siemens

The combination of symbolic or logic-based AI on the one hand and data-driven AI (e.g. machine learning) on the other is a current topic in cutting-edge AI research and at the same time the area of expertise that Siemens will contribute to this project. The “Configuration Technologies” research group at Siemens Austria, which operates within the global “Data Analytics and Artificial Intelligence” technology field, has more than 30 years of experience in the application of constraint technologies to industrial use cases. In recent years, this expertise has been extended to other areas of symbolic AI (such as answer set programming and recommendation technologies), to data-driven AI (data analysis, machine learning) and to semantics and knowledge graphs (graph databases, ontologies, stream reasoning).

Connecting AI worlds to create hybrid AI

“Through the interplay of symbolic and data-driven or sub-symbolic AI, the application potential of this combination is expected to far exceed either area on its own. We expect that this hybrid AI with both symbolic and sub-symbolic approaches will be useful in solving problems such as those being investigated in the SAELING project. By uniting the two AI worlds, new possibilities such as deriving logically correct conclusions in uncertain environments or improving the performance of logic-based methods by learning from data become possible,” says Herwig Schreiner, head of the Configuration Technologies research group.

Uniting the two AI worlds opens up new possibilities such as deriving logically correct conclusions in uncertain environments or improving the performance of logic-based methods by learning from data.

Herwig Schreiner, Head of the Configuration Technologies Research Group, Siemens Austria

© Siemens

“The SAELING research project will therefore develop methods and algorithms for optimized decision-making. The challenge of multi-criteria optimization lies in the uncertainty of the predictions of machine-learned models. SAELING will therefore explore robust and explainable methods by combining different AI techniques,” says Comploi-Taupe.

The result of the research project will be a proof-of-concept system that demonstrates the solution to the problem for the voestalpine use case. SAELING’s goal is to increase energy efficiency by 20 to 30 percent and thus achieve annual energy savings of 4 to 6 GWh. This also reduces CO₂ emissions and the overall total cost of ownership.

The “Configuration Technologies” research group at Siemens Austria is also contributing its hybrid and neurosymbolic AI research to the MATISSE project (Model-based engineering of Digital Twins for early verification and validation of Industrial Systems). This project is funded by the FFG and the EU, but as an EU project with 29 partners from seven countries, it has a larger scope.

The MATISSE project is developing a system based on AI and digital twins for the continuous validation, reconfiguration and redesign of physical systems. Applications include the manufacture of trains and buses as well as the maintenance of bridges and satellites in orbit.

Microgrid use case

The Siemens use case relates to the intelligent micro-grid at the Siemens City campus in Vienna (hi!tech has already reported on this several times), consisting of a PV system, an electricity storage unit, charging points for electric cars, a load management system and an interface to the building management system. “In the Siemens use case, the vision is to have a system that continuously ensures that the rules applied in a microgrid are optimal for the current context,” says project manager Danilo Valerio from the research department at Siemens Austria. The e-mobility charging stations of the microgrid on the Siemens campus in Vienna are being used as a test environment.

© Siemens

The charging stations at the Siemens City Campus in Vienna are the test environment for the development of a system that optimizes the operation of microgrids.

A microgrid is a system of interconnected electrical installations with defined boundaries that acts as a single controllable unit. The electrical components interact with each other and can generate, consume or store energy at any time. The way in which the energy flows within the system depends on a series of static rules that are defined during the installation phase. However, these rules do not allow the full potential of the systems to be exploited. In addition, changing the rules and tracking the changes can be a lengthy process that in some cases is not even possible.

“The aim is to develop a solution that works via plug and play, without having to define all the operational details of the microgrid in advance. To achieve such an ambitious goal, the solution must be able to continuously complete the knowledge base. We plan to do this with neurosymbolic AI approaches and inductive logic programming,” Valerio continues.

In the Siemens use case in the MATISSE project, the vision is to have a system that continuously ensures that the rules applied in a microgrid are optimal for the current context.

Project Manager Danilo Valerio, Research Department Siemens Austria

© Siemens

Another focus of the research group’s contribution is the reconfiguration of the system using a digital twin based on multi-objective optimization. It is essential to combine optimization and predictive analytics, taking into account that a solution must be optimal for the energy demand, the weather and the energy generation mix predicted for the near future. Hybrid AI approaches are used for this purpose, which integrate the optimization function into the machine learning training.