Product design has a major, often under-estimated impact on sustainability. This article describes the process of a new sustainable product design aiming for integrating circular engineering and social aspects as an integral part of the design process. A methodology is presented that enables a holistic, systematic approach to product development with the product design goal of a ‘Circular System’ with three subordinates but interdependent key elements: the ‘Product’, its ‘Life Cycle System’ and the related ‘Context System’. Social and ecological responsibility is integrated into product design from the beginning, ending up in a cyclic ‘Convergent Creation Model’. The associated methodology can be understood as a valuable opportunity to structure and control the complexity of sustainable product design for circularity. Two representative use cases - an established high-volume circular product (smart water metering unit) and a strategic design and R D perspective on “green” robots, contributing as sustainably engineered production systems to zero emission production – are discussed. Both demonstrators for the general applicability of the proposed ‘Circular System Design’ approach underline the potential of the methodology for advanced sustainable systems engineering in different phases of product design and circular production.
The work concerns the use of D-Wave’s quantum cloud service to solve the NP-hard flow shop scheduling problem with due dates and with the criterion of maximizing the weighted number of tasks performed on time. Constrained Quadratic Model, Binary Constrained Quadratic Model, and Binary Unconstrained Quadratic Model were proposed. Load experiments were carried out in a hybrid D-Wave LeapHybridCQMSampler environment using a combination of metaheuristics and quantum annealing, and DWaveSampler natively implementing quantum annealing. Calculations in the DWaveSampler environment are performed very quickly, but their practical application is currently limited due to the relatively small number of available qubits.
In this paper we tackle the dynamic stacking problem by introducing a framework for incremental online optimization. The dynamic stacking problem features continuous uncertain arrival and delivery of blocks via a crane controlled by the solver. The problem is implemented as a discrete event simulation and the solver runs asynchronously. We develop a framework that can use our existing offline solver for the dynamic stacking problem and turn it into an online solver capable of incrementally updating optimized plans. We test our framework by comparing to our previously published iterative approach as well as a rule based baseline solver on a diverse set of problem instances. Using the new framework, the solver improves our key performance indicators across the benchmark instances. We also investigate the reasons for the performance differences both in the aggregate as well as the level of individual simulation runs. The framework not only works well on this specific stacking problem, but is general enough to be used in many online dynamic optimization problems.
This paper aims to analyze and compare several technologies to measure their energy efficiency, with the goal of defining sustainable software design and architectural patterns. The study presents a methodology designed for assessing and optimizing energy consumption in IT systems and demonstrates how it contributes to sustainable development within the industry.
Trustworthiness in the design of AI software is crucial for acceptance and performance, particularly in complex socio-technical systems. The AI-DATA model, introduced by the funded 506 GEMINI project, addresses this need through a human-centered approach to Onsite Customer Journey Optimization in a case study of an online shop for nutritional supplements. This model combines systems thinking, process mining, and Large Language Models (LLMs) to enhance the Customer Journey in E-Commerce. By evaluating various prompt techniques, AI-generated marketing claims for onsite activation are tailored to a specific audience within specific phases of the AI-DATA model and utilized as triggers for interventions. This integrated methodology ensures optimization of the Onsite Customer Journey by providing relevant and trustworthy content at the right Moments of Truth in order to increase conversion rates, such as purchases, thereby fostering sustainably revenue growth.
Cerebral Palsy (CP) requires individualized interventions due to its complex nature affecting movement and coordination. Medical Human Digital Twin (MHDT) technology offers significant advancements in CP management by creating virtual representations of patients’ physical and neurological states. This paper reviews MHDT applications in CP diagnosis, therapy, and rehabilitation. Advanced imaging techniques and sensor integration enhance early and accurate diagnosis through detailed patient modeling. Therapeutic applications focus on personalized treatment plans using AI-driven models, VR/AR, and robotic-assisted devices, ensuring continuous optimization through real-time monitoring. Rehabilitation strategies benefit from immersive technologies and adaptive feedback, improving patient engagement and outcomes. The integration of human empathy with data-driven insights enables more precise and effective care. Ethical considerations, data management challenges, and future research directions, including interdisciplinary collaboration, are discussed. MHDTs present a transformative approach to CP care, promising improved patient outcomes and personalized healthcare solutions.
Caenorhabditis elegans as an in vivo model organism provides the potential for higher throughput substance testing, leading to reduced animal testing, substance use, and experiment costs. In this work, white light and fluorescence images of closeup captures of C. elegans worms were used as a modality of measurement for the protein expression. To measure worm morphology and the effect of substances on the nematode’s behavior, fitness, and survivability relevant features will be extracted automatically. With automated segmentation and localization of worms in both modalities, important features can be extracted allowing conclusions on substance effects. For the segmentation, we used a Mask R-CNN to extract single worm instances and to allow the separation of close instances. Different effects on the training process and the combination of both image modalities were investigated. This results in a low MAPE and a high R^2 on unseen C. elegans images for important morphological and protein expression features such as mean intensity ( R^2 = 0.995), length ( R^2 = 0.952) and area ( R^2 = 0.983).
We address the role of food demand and the agents’ mental models for the future of water scarcity. After briefly introducing fundamental facts concerning the water cycle and human water withdrawals, a system dynamics model containing a stylized water cycle and the interaction with a food producer is presented. The producer’s task is to maximize profits while conserving sufficient water reserves in the mid- and long-term. Simulation of various stylized decision policies shows a range of different outcomes, making the simulator a valid instrument for decision-making experiments. We close the contribution by discussing the role of decreasing precipitation in the water system and describing future steps.
Suicide is a major health and social problem worldwide; and, therefore, family members and friends of people who have suicidal ideation, or just people seeking for information, require a simple access to reliable and useful sources of information about suicide. This information can be provided by means of chatbot tools; however, the reliability and topicality of the chatbots’ answers should be ensured. In order to reach that goal, it is necessary to have datasets that are able to provide a ground truth for a proper evaluation, as well as contain reliable and unbiased information for the purpose of fine-tuning these chatbots. In this work, we present the creation of three question-answering datasets about suicide in Spanish and the methodology followed in their creation. In particular, we have created three datasets with three quality levels: bronze, automatically generated; silver, derived from bronze but verified by professionals; and gold, manually generated by professionals. This is a first step towards building reliable chatbots that answer questions about suicide; and, additionally, it can serve as a basis for other contexts where it is necessary to access to verified and reliable information.
This paper investigates practical strategies for improving manufacturing processes through computer simulations in the domain of industrial automation. The focus is on employing Model-Based Systems Engineering (MBSE), utilizing AutomationML—an XML-based framework—and Visual Components, a tool for simulating industrial processes at the layout and kinematics levels. The study specifically addresses the optimization of the simulation modeling process by examining the integration of AutomationML data into Visual Components. The research involves adapting an existing Visual Components XML-Importer add-on to interpret AutomationML data. This modification allows for the updating of an XML file with information related to the entry station in Visual Components, serving as a illustrative example within a broader mechatronic production system. The findings contribute to the broader MBSE framework by demonstrating a practical application of AutomationML and showcasing its role in enhancing simulation capabilities within industrial processes. Furthermore, this research underlines its potentially valuable contribution to the educational context by facilitating teaching with simulations.
In the last decades the measures of uncertainty are of growing interest. Since there are many versions of entropy and of its dual version, unified formulations for entropy and extropy have been introduced to study their properties and to compare them.
In this article we report about our efforts to employ detailed physical simulation models (so called digital twins) and advanced optimization algorithms to identify optimal investment decisions into renewable energy systems to reduce energy costs and carbon emissions. From our own experience and discussions with several energy consultants, we learned that even when presented with very good investment options, decision makers are very often hesitant to act quickly. In this work we argue that to tackle this problem, the psychological aspects of the optimization and decision making process need to be addressed explicitly. We review some of the main insights from psychology of decision making and propose trust as the key factor with transparency and explainability as the main necessary contributors for building trust.
In the field of time series analysis, the scarcity of comprehensive datasets poses a significant challenge for the development of reliable predictive models. This study addresses the difficulties in forecasting solar module outputs and enhancing data accessibility for modeling, especially in residential sectors. We propose a general method to establish a distribution of photovoltaic module parameters across a country and, from this, generate a synthetic dataset for simulation and modeling pv module output. This approach integrates multiple freely available data sources. The study is focused on Germany, utilizing the Marktstammdatenregister as its main source for the module parameter distribution. The data is then enriched using publically available data. Based upon this, a crawler is developed to gather fair-weather module outputs from the Photovoltaic Geographical Information System for training, testing, and benchmarking purposes. One benchmark has fixed locations and the second one has fixed module parameters. Additionally, we provide a data loader with artificial degradation for all datasets. In the last step we test multiple state of the art models on the dataset and show that the proposed forecasting task is not trivial. All the code and data is publically available.
Patients with transradial hand amputations have long been able to be fitted with myoelectric arm prostheses. The current state of technology offers users a range of highly developed mechanical and bionic hand prostheses with individually movable fingers. The control of such prostheses, however, is not intuitive, requires much training, and not all possible gestures can be performed. The reason for this is often cited as the number and quality of the myoelectric signals, the amount of training data, and resulting the achievable classification quality. Since the incremental improvement of our existing approach allowed only minor improvements, we experimented with new methods for classification and control. This contribution presents an integrated solution consisting of a smart sensor, individualized feature extraction, distributed classification, and inverse kinematics to control individual fingers.
Reinforcement Learning (RL) has emerged as a pivotal technology in enhancing production systems, offering solutions for optimizing complex, dynamic processes. This study presents a quantitative trend analysis of RL algorithms in production systems, addressing a significant gap in the literature. We propose a methodology for conducting quantitative literature reviews and apply it to assess the current state-of-the-art and temporal development of RL applications in this domain over the past decade. Our findings reveal a marked increase in research activity since 2017, with significant contributions in robotics, scheduling, and energy management. Model-free RL algorithms, particularly Q-learning, DDPG, and DQN, are the most frequently utilized, reflecting their broad applicability and effectiveness. To ensure the robustness of our methodology, future work will compare our quantitative results with existing qualitative studies. Additionally, we plan to replicate this analysis periodically to monitor the evolution of RL in production systems and explore the applicability of our methodology in other fields.
The digitalization of the economy and society continues unabated. Generative artificial intelligence, especially ChatGPT, has triggered a new wave of digitalization since its availability in November 2022. In addition to new technologies such as generative AI, companies and their IT function are still busy with the transformation of applications and IT infrastructure into the cloud, the expansion of capabilities in data analytics, the agilization of application development, new regulatory requirements, the increase in IT security and the shortage of skilled workers. Furthermore, many companies are developing new digital business models that place new demands on the IT function. Social developments, such as sustainability and diversity, are also leading to new challenges for the IT function. The pace of change in companies is also accelerating. The complexity that IT management must deal with has grown in recent years and continues to grow at a rapid pace. Traditional IT management models and IT operating models, such as ITIL or COBIT, are not able to cope with these challenges. Against this backdrop, this contribution discusses whether and how systems thinking can be an integral part of future IT management models and IT operating models and can help deal with the increased complexity brought about by digitalization.
The ongoing development of autonomous vehicles requires introducing advanced technologies and protocols to ensure road safety and efficiency. Conventional data recording devices in standard vehicles have limitations in data storage and accessibility, which hinders efficient information sharing and analysis. In addition, connected vehicles are vulnerable to targeted attacks that can lead to potential data breaches. EU regulation 2019/2144 has mandated the introduction of standardised data recording systems from 2024 to address these challenges. Based on this, we propose a standardised event detection and response system for autonomous vehicles introducing a Client2X architecture to improve data collection, storage and analysis. This architecture enables efficient machine learning-based event analysis, faster data retrieval and data integrity. An external monitoring system complements the in-vehicle data storage and ensures comprehensive data analysis. The proposed system aims to accelerate incident resolution and improve vehicle safety and data management.
In the face of climate change, it is a necessity to reduce the environmental impact of AI models. The pursuit of more energy-efficient AI is referred to as Green AI. A key ingredient in Green AI strategies is to reduce the amount of data used for training and inference. This is especially important for energy intensive sectors like the manufacturing industry, which is already under pressure to reduce their production-related CO2 emissions. This study evaluates empirically the impact of sample rate reduction on time series data as a possibility to implement Green AI for manufacturers that use CNC machining. A real-world dataset is used for this research to evaluate the performance of the decimation pre-processing step for our sample application, semi-supervised deep anomaly detection (DAD). The results show that decimation has the potential to assist manufacturers to reduce resource consumption and to make advances in operational sustainability. In addition, the DAD performance is not only stable, but even improves to a certain extent in the course of the sample rate conversion. This can arguably be attributed to the beneficial effect of noise reduction. This research contributes to the ongoing discourse to develop energy efficient AI methods and is especially applicable to the field of manufacturing processes.
Large Language Models are of fundamental importance for future AI usage in industry and society. The environmental impact of training and inference of such models however is devastating, which necessitates research into more efficient model mechanisms. In the simple context of the well-established BERT architecture, this study performs a comprehensive comparison between classical fine-tuning, which involves most or all of the model weights, and parameter-efficient Adapter Tuning, in which shallow trainable layers are introduced throughout the model while the vast majority of original weights remain unchanged. A series of experiments were carried out in which the BERTbase-architecture was trained on the LexGLUE benchmark with fine-tuning and Adapter Tuning for each data set. Through extensive comparisons in the experiments, it was found that Adapter Tuning is advantageous from a green AI perspective at epoch level, supporting the claims made in the literature. However, in our experiments reaching model convergence with Adapter Tuning requires significantly more time, making this training method in total less environmentally friendly than the conventional fine-tuning method, at least for the chosen model family.
This paper proposes a combination of multi-objective beam search and the pilot method. The idea is to use the pilot method as a problem-independent metaheuristic to guide the search process of multi-objective beam search. Through this combination, the newly created search algorithm is less dependent on problem specific heuristics, which would otherwise be required to guide the search process. Especially when dealing with problems where no well-known heuristic functions exist this approach can be feasible. It must be noted, that in this situation, the design of the required heuristic functions is a challenging task because a deep understanding of the underlying problem as well as the chosen search algorithm is required. The proposed methodology does not use multiple heuristic functions to evaluate nodes. Instead, a sub search process in the form of the pilot method is started to obtain an estimation for the node’s quality. This sub search process is a single target search process and can only provide an estimation for one of the problems target functions. Therefore, the target functions are now also used as heuristic functions. Applying this new search algorithm to well-known benchmark problems yielded promising results and generates high quality solutions compared with current state-of-the-art algorithms.
