Themenpublikation Industrielle KI englisch

Industrial AI for sustainability

Topic Industrial AI for a Sustainable Economy Author: Dr. Christiane Plociennik The use of digital technologies is seen as an opportunity assemblies (so-called remanufacturing). Such sustain- to save resources and optimize production processes. able production also means that the products must be At the latest, since large language models and applica- competitive and economically sustainable – it must sim- tions based on them, such as ChatGPT, have become ply be worthwhile to produce sustainably. Otherwise, mainstream, artiﬁcial intelligence is also on everyone‘s the company will hardly be able to hold its own on the lips as a potential enabler for improving processes and market in the long term. Social sustainability means that products. However, these improvements alone will not be enough in the fu- ture. Our economy must not only become more efﬁcient but also more sustainable. The United Nations stipulates this with its Sustainable De- velopment Goals (SDGs), and the German govern- « Can AI not only improve efficiency but also sustainability in production and throughout the entire life cycle of the product? » people should work under good conditions and be paid fairly. In the future, this should not only ap- ply to the company itself but also to its suppliers. This is the aim of the new Supply Chain Act, which came into force in January 2024. ment is currently working on the National Circular Econ- omy Strategy. It is due to be adopted this year. Reason Towards a circular economy with AI One way to promote sustainability is to transition from a enough, therefore, to look into whether AI can improve linear economy to a circular economy. The circular econ- efﬁciency and sustainability in production and through- omy attempts to keep raw materials “in circulation” for out the entire life cycle of products. as long as possible, i.e., to generate as little waste as Basically, sustainability means not consuming more re- possible by reusing products, repairing them as often sources than can be replaced, whereby there are three as possible, and recycling them as much as possible at dimensions: ecological, economic, and social sustaina- the end of their life cycle. Reuse, repair, and recycling bility. In this sense, production is sustainable if it works are three of the ten so-called R-strategies of the circular resource-conserving and resource-efﬁciently, for exam- economy (see R-Graphic). ple, by avoiding or reusing waste as far as possible or by For the circular economy to work, it needs data above incorporating used individual parts or even entire used all – data on the product itself, the production process, 4

Topic and a possible repair (e.g., repair instructions) or suita- players in the circular economy is crucial. But how can ble recycling processes at the end of the product’s life. this be organized? This data helps the various parties involved in the circu- The digital product passport is the ideal solution. Start- lar economy: a product designer, for example, can (re) ing with batteries with a capacity of over 2 kWh, this will design a product with a view to improved sustainability become mandatory for more and more product groups if they know which recycling processes are possible in in the EU from 2027. The exact product-related infor- principle. Information on why a product cannot yet be mation it should contain, as well as the format and recycled helps improve the design. The recycler, in turn, data exchange procedure, has not yet been ﬁnalized. can parameterize his systems much better if he knows One of the purposes of the product passport is to cre- the composition of the products he receives. Therefore, ate the transparency required by the Supply Chain Act. the exchange of information between the individual However, it is also suitable for implementing the Circular economy R-ladder R0 Refuse R1 Rethink R2 Reuse R-Graphic: The ten R-strategies for circular value creation. R4 Repair R5 Refurbish Use R3 Reuse R8 Recycle R6 Remanufacture R9 Recover energy R7 Repurpose In a circular economy this is avoided

Topic R-strategies of the circular economy with the help of AI. DFKI is also researching this. DFKI tests the digital product passport DFKI is implementing various product passport con- cepts as prototypes in its research projects. In the CircThread project, DFKI and 30 other Europe- an partners are exploring the possibilities of the dig- ital thread concept, a kind of digital thread across the entire product life cycle. In an information system, it is The Digital Lifecycle Passport (DLCP) enables role-based access to product information. It can be read by both humans and machines. not the information about a product itself that is stored SMEs over a period of six months. The applications cover but where it can be found. This information can then be a wide range of topics, from optimizing logistics and the used at the end of the life cycle to decide what to do with use of goods in the skilled trades to digitally integrat- the product – is it worth remanufacturing, or should the ed solutions for plant maintenance or closing material product be recycled? flows in industry. The Digital Lifecycle Passport (DLCP), a product passport To make the topic of “AI for resource efﬁciency” tan- based on the Asset Administration Shell, was designed gible for interested companies, a mobile demonstrator and implemented in the ReCircE (“Digital Lifecycle Re- was developed as part of the Green-AI Hub Mittelstand. cord for the Circular Economy”) project. The DLCP ena- The demonstrator clearly shows how companies can in- bles role-based access to product information. It can be tegrate the R-strategy “Reuse” into their production. read by both humans and machines. This means that the The quality control and storage of used parts are shown information can be displayed user-friendly via a web ap- using the example of a model truck. To do this, an AI- plication but can also be used to control machines. This based optical quality control system ﬁrst assesses the was also implemented in the ReCircE project: combined condition of a used product and sorts it into one of four with AI-based object recognition, the information from quality categories on this basis. A price is then deter- the DLCP is used to control a sorting system. The result mined for the used part, the potential CO2 savings are is a hybrid AI system that sorts e-waste according to in- predicted, and a storage space is reserved in the con- novative sorting criteria – for example, whether a device nected parts warehouse. All this information is stored in is potentially dangerous because it contains a battery. the truck’s digital product passport and published via a Such a device should be sorted out before recycling to digital marketplace. Potential customers and manufac- prevent it from catching ﬁre. This approach can, there- turers can access the information via this marketplace fore, improve the sorting and recycling process for old and decide whether used parts should be installed in a electrical appliances. new truck to be produced. These can then be ordered The digital product passport is also an essential topic as required and taken from the parts warehouse for for the Green-AI Hub Mittelstand. The Green-AI Hub the new production process. DFKI is thus expanding its Mittelstand is an AI initiative of the Federal Environ- demonstrator landscape to include the topic of the cir- ment and Consumer Protection Ministry, coordinated by cular economy and bringing it to Hannover Messe 2024. Zukunft Umwelt Gesellschaft gGmbH and implemented by DFKI. The Green-AI Hub Mittelstand helps compa- nies use modern AI solutions for greater resource efﬁ- A look into the future of the circular economy With CircThread, ReCircE, and the Green-AI Hub, DFKI ciency. At the heart of the Green-AI Hub are 20 AI pilot demonstrates how the interaction between the circular applications, each of which will be implemented at pilot economy and AI works. However, there is still a lot of 6

Topic research to be done, as the data situation in the circular still needs to be added and in what format it needs to economy is currently still patchy. There is often a lack of be available. Procedures are then collected and tested information at key points to enable decisions to be ma- in order to create comprehensive sustainability assess- de – by AI or by humans. The relevant community also ments from this data, e.g., AI-based analyses for specif- does not yet have sufﬁcient expertise to harness and ic products. Ultimately, scientists, companies, and the analyze the appropriate data. This is where the Data general public should beneﬁt from these ﬁndings. This Competence Center for Circular Economy Data – DACE brings the AI-based circular economy a good step closer for short – comes in. It brings together computer sci- to the mainstream. ence and AI researchers from different universities and institutions with experts in the circular economy and www.green-ai-hub.de life cycle assessment so that everyone can learn from www.recirce.de each other. The ﬁrst step is to understand where data https://circthread.com The mobile demonstrator of the Green-AI Hub Mittelstand shows how companies can integrate the R-strategy „Reuse“ into production.

Interview Resource-saving AI Through Algorithms How high do you estimate as relevant as research into electric the potential for ener- and hybrid drives and the potential gy savings through re- intelligence? uses of green hydrogen. It is a known fact that The power consumption of the data centers are ener- Internet is still relatively low gy-hungry. Many compa- compared to other industries and nies are relocating their sectors, but it is rising steadily. centers to colder regions In Germany alone, the internet to save on electricity Every question you ask ChatGPT has consumes around 13 TWh per year, costs for cooling. Does its price. Every time you develop and corresponding to the consumption this make operations more use AI, you leave an ecological foot- of all private households in Munich. sustainable? print. In this interview, Wolfgang - The CO2 emissions are about as The operator reduces the electric- high as air trafﬁc or steel produc- ity consumption for cooling but telligence can be sustainable and tion. From a global perspective, the heat is still generated. We are which indicators need to be met for the electricity consumption of data currently unable to use this energy this to be the case. He is a professor centers is expected to double from sensibly, so we are heating up the of business administration, particu- 460 TWh in 2022 until 2026. Con- environment instead. The multi- larly business informatics in the ser- sidering that the essential services disciplinary team in the ESCADE vice sector, and head of DFKI‘s Smart of the internet are now based on research project wants to get to Service Engineering research de- AI, there is enormous potential for the root of the problem and is partment. He is currently coordinat- savings here. We can tap into this if researching more energy-efﬁcient ing and leading the ESCADE project, we make AI-based services on the algorithms and hardware. in which a consortium of numerous - internet signiﬁcantly more efﬁcient. Developing new concepts for the The ESCADE project is a cient data centers. climate-neutral Internet is at least medium-sized project 8

Interview funded by the Federal Min- of our considerations is the ques- At the same time, energy proﬁles istry for Economic Affairs tion of which AI algorithms we can are created to measure the energy and Climate Action, with a use to come closer to the economic consumption of the AI application volume of around 5 million functioning of the human brain in and compare it with the positive euros. What is unique terms of design. The concept of sustainability effects of the appli- about it? neuromorphic computing, which cation and other scenarios. Using Compared to the resources that big makes it possible to switch areas optimized hardware with highly technology providers can spend, of the chip on and off depending on efﬁcient AI algorithms, we are that’s a little. However, more money the computing task via algorithmic trying to create a mobile envi- does not necessarily mean more control, is very promising here. ronment that delivers maximum success. When it comes to provid- performance with minimum ing a proof of concept, a small team You said ESCADE addresses energy input. can be just as successful. ESCADE the entire technology stack is not just a project on resource from the chip to the use What usable results can efﬁciency but also a resource- case. What does this look the economy expect? efﬁcient project that follows the like? We hope ESCADE will become a guidelines of frugal innovation. It One application example is the use pioneering concept for resource- is also a full-stack project that of visual computing to efﬁciently efﬁcient data centers and AI use considers all components from sort steel scrap for recycling using cases with proven business sustain- algorithmics to implementation, a distributed architecture that tests ability. To make this measurable, including chip design, chip and evaluates various AI models ESCADE is developing a framework production, and the use case in an and scenarios in terms of their model for measuring the sustain- integrated manner. Furthermore, energy efﬁciency. Using a drone ability of AI systems. This will the entire technology stack is in and an NVIDIA Jetson board, we provide the German economy with German hands. classify steel scrap on the site of a a model for balancing AI energy DFKI’s contribution to the enabling more informed decisions project is the algorithms. on how to recycle the scrap steel. https://escade-project.de steel mill on-edge and in real-time, consumption. What role does software play in improving the energy balance of data centers? Software design can start with the selection of large language models. It doesn’t always have to be GPT-4. For specialized tasks, you can also work with smaller models derived from larger ones. Not all computing operations have to be carried out on mainframes, but they can also be trained and executed on edge close to the application. The focus

More on the topic Edge Devices in Production Agriculture, healthcare, transport and logistics, the In- of the Franco-German cooperation in the ﬁeld of AI. ternet of Things (IoT), production, and manufacturing – FAIRe (Frugal Artiﬁcial Intelligence in Resource-limit- wherever fast data processing and real-time decisions ed environments) aims to enable the deployment of AI are required, specialized, compact hardware is used applications on mobile devices through an innovative on-site. In industrial production, edge devices record approach to reducing the model size and computational and analyze sensor data or detect faulty products. They effort through network quantization, optimization of the promise lower latency times and complete control over network architecture, optimization of the computation- locally processed data. Edge computing is booming: al flows, and, ﬁnally, network execution on low-power studies predict that the Internet of Things will grow high-performance hardware. by more than 16% by 2028. Data processing on-site already offers a certain savings potential per se: No energy has to be used to transfer the data to the data center and retrieve the results from there. But there is more to it. Frugal AI is the name of the concept for eco- nomical artiﬁcial intel- ligence, which aims to „Our goal is to develop a comprehensive ap- proach that addresses the different layers of implemented, running AI systems. To do this, we consider all levels of abstraction of an AI application: The actual AI algorithms, the hard- ware, and the compiler layer in between, which is responsible for the ef- ﬁcient translation of the reduce power consumption for peripheral computing algorithms to speciﬁc hardware,“ explains Prof. Chris- by using less memory, less computing power, and more toph Lüth, who is leading the project on the DFKI side. data efﬁciency. A case study will demonstrate the practicality of the ap- The French computer science institute Inria (Institut proach. The interaction between a human and a robot national de recherche en informatique et en automa- will be used to illustrate how the various layers of ab- tique) and DFKI are working on this in the FAIRe pro- straction work together and how adaptation to unknown ject. The project itself is a particular case: it is funded contexts and continuous robotic learning can also suc- by the research ministries of both countries as part ceed in a resource-conscious manner. 10

About us Human-Centric AI Intelligent Solutions for the Knowledge Society DFKI has been researching human-centric AI for over 35 years, focusing on social relevance and scientiﬁc excellence in the key future-oriented research and application areas of artiﬁcial intelligence. The German Research Center for Artiﬁcial Intelligence (DFKI) was founded in 1988 as a non-proﬁt public-private partnership. It has locations in Kaiserslautern, Saarbrücken, Bremen, Lower Saxony, laboratories in Berlin, Darmstadt and Lübeck as well as a branch ofﬁce in Trier. In 26 research departments, ten competence centers, and eight living labs, product functions, prototypes, and patentable solutions in the ﬁeld of information and communication technology are developed on the basis of application-oriented basic research. Financing is provided through public funding and development contracts from industry. Project results and milestones are periodically reviewed institutio- nally and by an international panel of experts (Scientiﬁc Advisory Board). In addition to the federal states of Rhineland-Palatinate, Saarland and Bremen, numerous renowned German and interna- tional high-tech companies from a wide range of industries are represented on the DFKI Supervisory Board. www.dfki.de 11

Lübeck Oldenburg Bremen Berlin Osnabrück Trier Darmstadt Kaiserslautern Saarbrücken Imprint: dfki ai next Issue April | 2024 Published by: German Research Center for Artiﬁcial Intelligence GmbH (DFKI) Trippstadter Str. 122 D-67663 Kaiserslautern Phone: +49 631 20575 0 E-Mail: news@dfki.de Responsible: Heike Leonhard, Armindo Ribeiro Layout: Lando Lehmann, Annemarie Popp Production: One Vision Design Credits Cover: Adobe Firefly, AI-generated image, Prompt: sustainability transformation with the help of artiﬁcial intelligence, resource conservation; Page 3: DFKI, Jürgen Mai; Page 5: Fig. adapted from Minguez, Rikardo & Lizundia, Erlantz & Iturrondobeitia, Maider & Akizu-Gardoki, Ortzi & Saez de Camara, Estibaliz. (2021). Fostering Education for Circular Economy through Life Cycle Thinking. 10.5772/intechopen.98606. CC-BY-SA 3; Page7: DFKI, Maximilian Spengler; Page 8: DFKI, Jürgen Mai; Page 9: SHS/ Saarstahl AG; Page 10: Kanchana, adobestock.