Keywords: Electrical, Electronic and Power Systems, Automation of Modelling and Software tools, incl. Computer Modelling
Abstract: Performance spaces represent the full range of attainable performances of an integrated analog circuit, including gain and bandwidth. Based on the knowledge of the optimal and least favorable values, simplified simulation models can be employed to optimize the system design process. We present a modified variant of the well-established Normal-Boundary Intersection method, which is used to approximate five-dimensional performance spaces in reasonable time. Based on these approximations, we develop a concept for implementing a system-level simulation that is used to design an operational transconductance amplifier for an analog-to-digital converter to improve its sampling rate. To demonstrate the usability of our concept, we perform several numerical experiments and visualize the data in parallel coordinates plots for better user comprehension.

Keywords: Electrical, Electronic and Power Systems, Automation of Modelling and Software tools, incl. Computer Modelling
Abstract: Performance spaces obtain all possible combinations of competing performance parameters for analog integrated circuits, like gain and bandwidth. The best combinations of those performances form the so-called Pareto front. While these spaces contain a lot of information, visualizing them - especially in high dimensions - can be overwhelming and difficult. Therefore, we present a combination of Parallel Coordinates plots together with a clustering algorithm in order to allow the exploration of performance spaces in a simple and intuitive manner. To compute approximations of those high-dimensional spaces, we use a parallelized version of a state-of-the-art box-coverage algorithm. Several numerical simulations for operational transconductance amplifiers demonstrate the functionality and efficiency of our concept for up to eight-dimensional performance spaces.

Keywords: Fitting Models to Real Processes, incl. Identification and Calibration, Electrical, Electronic and Power Systems, Data-Driven Models, Neural Networks
Abstract: Accurate models of semiconductor devices are a key component in modern circuit development. Although parameters of such models are often derived from fundamental physical laws, device geometry or material properties, empirical models from extensive measurements have become increasingly popular. Sparse identification of non-linear dynamics (SINDy) is one approach that emphasizes interpretability. Although several extensions have been developed for this procedure, it still lacks the ability to accommodate function compositions that are very common amongst established semiconductor models. This work introduces an approach fusing the capabilities of neural networks with the core principles of SINDy to address this shortcoming. The results are compared to well-known bipolar transistor models such as Gummel-Poon and indicate promising approximation capabilities.

Keywords: Physical and Multiport Modelling, Bondgraphs, Mechanics, Mechatronics, incl. Robotics, Automotive, Aerospace, Transportation Systems
Abstract: In this paper, a bond graph model of a 4 × 2 truck, integrating both multibody and functional approaches is presented. The relevance of the model is assessed by numerical simulation results comparison with regard to a reference multibody model. Then, using the inverse-based methodology in the bond graph framework, the sizing of the electric power steering assistance is addressed. Simulation of the inverse model identifies the performance requirements that the electric power assistance system must meet to respect the specifications.

Keywords: Data-Driven Models, Neural Networks, Machine Learning, Deep Learning, Data Analytics, Big Data, AI for Modelling, Electrical, Electronic and Power Systems
Abstract: The global energy transition increasingly relies on renewable energy sources and the use of batteries for electrical energy storage. Efficient battery utilization necessitates accurate state estimation algorithms and appropriate control mechanisms. This paper presents and evaluates a data-driven approach for estimating a battery's dynamic model using tensor trains, that efficiently reconstruct complex multidimensional systems with respect to time and memory, enabling the development of adaptive models capable of capturing real-time variations in system parameters. In this study, the proposed method is applied to reconstruct a dynamic battery model from operational data and is tested upon a solid-state lithium-ion battery cell. The method's explanatory capabilities are demonstrated through the extraction of key parameters such as open circuit voltage and impedance in the form of relaxation times distribution. The accuracy is further validated against the results of conventional battery characterization tests. Owing to its intrinsic scalability and low computational cost, this method holds potential for integration into artificial intelligence-driven battery management systems, enhancing battery longevity and safety while optimizing time-intensive battery characterization processes.

Keywords: Medicine, Physiology, Health Care and Biology, ODE, DAE, SODE, SDAE Systems, Model Reduction, Model Simplification and Optimization
Abstract: At least since the COVID-19, crisis wastewater-based epidemiology (WBE) has been recognised internationally as a reliable surveillance and early warning system to track circulating pathogens. By probing and sequencing wastewater, WBE allows to detect, quantify and characterise pathogens circulating in the connected catchment. The main advantage of WBE over traditional case-based surveillance is its scalability, allowing one sample to cover thousands without requiring active participation, reducing costs and bias. While cost-effective, nationwide surveillance at reasonable granularity incurs significant taxpayer costs. Deciding which wastewater plants to sample and how often is crucial for public health but must be economically justified to convince policymakers and the public of WBE's value. In this work, we present a differential-equation-based SIRS-network model extended by a drainage and a probing model that simulates the regional spread of a new pathogen, its concentration at wastewater plants, and the limit of detection of pathogen specific assays as a function of wastewater catchment characteristics. Finally we state an optimisation problem with the goal to minimise detection time by varying the selection of plants and sampling intervals. We show preliminary results with manually varied strategies and propose ideas for using a simheuristic to solve the optimisation problem.

Keywords: Machine Learning, Deep Learning, Data Analytics, Big Data, AI for Modelling, Operation Research, Logistics and Planning, Automotive, Aerospace, Transportation Systems
Abstract: This study presents a novel approach for disaggregating aggregated train delays into primary and secondary components using Gated Graph Convolutional Networks (GatedGCNs). We develop a graph-based representation of railway traffic that captures complex spatiotemporal relationships and long-range dependencies. Our method is applied to synthetic delay data generated from an agent-based simulation model of the Austrian railway network. We evaluate the model on classification and regression tasks, demonstrating high accuracy in distinguishing between primary and secondary delays. The classification task achieves 96% accuracy and 0.99 AUC, while the regression task attains an R-squared value of 0.86. These results significantly outperform a naive baseline model. The findings suggest that GatedGCN is a promising method for delay disaggregation and has potential for broader applications in capturing delay propagation processes. However, while the results on synthetic data demonstrate strong performance, further validation on real-world data is essential to confirm its practical applicability.

Keywords: Operation Research, Logistics and Planning, Automotive, Aerospace, Transportation Systems, Discrete and Discrete-Event Systems, incl. Petrinets
Abstract: The reduction of greenhouse gas emissions has gained more and more relevance in railway traffic over the past years. We want to propose a simulation based possibility to assess the quality of circulation plans from an energy-consumption perspective, the Green Markup, which should indicate the performance of circulation plans both in terms of the corresponding energy consumption as well as their robustness against delays in a realistic setting. This is achieved by using a simulation model that calculates the delay propagation within a time table based on injected primary delays and comparing the delays and energy consumption of the simulated circulation plan to an idealized base scenario. We define two seperate markups, the delay markup and the energy markup and define the green markup as a weighted sum of the other two. The paper uses real world use cases to calculate reasonable weights for the green markup.

Keywords: Automotive, Aerospace, Transportation Systems, Operation Research, Logistics and Planning, Comparison of Methods for Modelling
Abstract: The planning of traction units and drivers is a crucial strategic and tactical planning problem for railway companies. Due to the complexity of the individual problems, large railway undertakings typically solve these problems in a sequential manner, i.e., they first generate circulation plans and then use them to generate shift plans. As a consequence, the achievable quality of the shift plans depends on the operational efficiency of the circulation plan, and information dependencies between the two planning steps cannot be exploited. This paper presents different types of integrated circulation and shift planning. An investigation of the inherent challenges and benefits of an integrated planning approach is presented, initial approaches for data structures and algorithms are introduced, and first results are shown.

Keywords: Manufacturing and Process Engineering, Machine Learning, Deep Learning, Data Analytics, Big Data, AI for Modelling, Model Reduction, Model Simplification and Optimization
Abstract: Planning and scheduling problems can be challenging to manage, and the related optimization problem is often too complex to solve in real-time. The typical approach to address this issue is to apply heuristic policies, which perform reasonably well. Machine learning algorithms can be used to replace heuristics, but this raises the issue of obtaining unexplainable solutions. The application of Reinforcement learning with policy extraction technique can help produce explainable results to improve the dynamic system of the planning and scheduling problem. As a case study, we use the simulation model of a dynamic flexible job shop to learn a solution strategy for the associated scheduling problem with deep reinforcement learning. Based on this, we were able to extract a decision tree that is superior to classical dispatching heuristics for a wide variety of scenarios. It also provides valuable information about the criteria for decision-making. Here we offer a new approach to analysing and solving these problems - leading to an improvement in dynamic systems.

Keywords: Model Reduction, Model Simplification and Optimization, Bio-Technical, Bio-Chemical and Chemical Engineering Processes
Abstract: For model reduction of non-linear state space models the empirical Gramian framework is frequently used in various disciplines (e.g., chemical and mechanical engineering). Numerical computation of empirical Gramians is known to be computationally demanding and not always very accurate. To remedy these inaccuracies perturbations from steady states are often used, thereby omitting essential dynamics. By analysing the non-linear model reduction problem from a system identification perspective where the initial conditions x0 are viewed as parameters, we show that these issues are better handled. In addition, our approach allows for an accurate nullspace computation of the well-known observability matrix for non-linear systems. These (lack of) observability results can be verified by solving a well posed problem (in terms of complexity) with computer algebra software. We demonstrate both reduction and observability of a non-linear dynamical system in a few examples.

Keywords: Education in/for/with Modelling
Abstract: The Capacity Building Project entitled “Enhancing Higher Education Capacity for Sustainable Data Driven Food Systems in INDonesia” (FIND4S, “FIND force”) aims to boost the institutional and administrative resources of seven Indonesian higher education institutions (HEIs) in Central Java by upgrading and developing innovative BSc and MSc degree curricula. These curricula will provide advanced knowledge and skills to address the challenges and opportunities in the field and will focus on the integration of big data, quantitative modeling, and engineering tools, enabling students to solve complex problems in food systems and respond to industry demands. Through collaboration with European Higher Education Institutions, Indonesian universities will gain access to global expertise in food science, data science, computational methods, and advanced engineering techniques. This partnership will support the development of high-quality programs that align with international standards while addressing local challenges. The project also focuses on training academic staff to utilize big data analytics and quantitative modeling, allowing them to incorporate these tools into their teaching and research activities. A key component of the initiative is the creation of a dedicated research center equipped with modern laboratories that will foster innovation and practical learning. By leveraging big data and engineering tools, the project will engage various stakeholders, including industry partners, to ensure that the programs meet market needs and contribute to a greener, more sustainable economy. Ultimately, this initiative will drive the transformation of food science education in Indonesia, equipping graduates with the skills and knowledge necessary to lead the shift toward socially, environmentally, and economically sustainable food systems. By integrating advanced technological tools into the curriculum, the project aims to support Indonesia’s transition to a resilient and innovative food economy.

Keywords: Model Reduction, Model Simplification and Optimization, Bio-Technical, Bio-Chemical and Chemical Engineering Processes
Abstract: A sensitivity-based algorithm to establish state controllability is extended to establish output controllability being the ability to control the outputs of a nonlinear dynamical system instead of the full state. Due to the exceptional efficiency of the sensitivity-based algorithm, large-scale nonlinear dynamical systems can be handled, as demonstrated by several examples in this paper. As a final contribution, this paper starts with a simple analysis of state and output controllability properties of nonlinear dynamical systems in terms of connectivity’s and sensitivities. The latter relate directly to the sensitivity-based algorithm.

Keywords: Bio-Technical, Bio-Chemical and Chemical Engineering Processes
Abstract: Global sensitivity analysis (GSA) is used to explain the dependence of the model output on every individual parameter in the model. While several GSA approaches are available in literature, variance-decomposition based Sobol sensitivity indices are widely used. This approach is based on decomposing the variance of the model output into variance terms associated with each parameter. In this paper, we demonstrate the utility of Sobol GSA in understanding how bioparts (i.e., model parameters) affect the engineered gene expression. As a case study, a synthetic gene circuit is considered in which the expression of green fluorescence protein (GFP) is triggered by a transcription factor.

Keywords: Machine Learning, Deep Learning, Data Analytics, Big Data, AI for Modelling
Abstract: In recent years, Natural Language Processing (NLP) has experienced a paradigm shift from fine-tuning pre-trained large language models (LLMs) on task-specific data to in-context learning, where task descriptions are embedded directly into the LLM input, allowing the same model to perform multiple tasks. While both approaches have demonstrated impressive performance across various NLP tasks, their opaque nature poses challenges in understanding their inner workings and decision-making processes. In this talk, I will discuss the research my team has conducted to address interpretability concerns surrounding neural models for language understanding. This includes interpreting uncertainty in text classifiers built on LLMs, developing explainable methods for scoring student answers in science exams, and exploring monosemanticity in LLMs through feature decorrelation. I will conclude my talk by sharing perspectives on potential future directions for interpretable language understanding.