How to Implement a Simulation Strategy

Authored & Presented By
Andy Richardson (Phronesim)

Description
This short course introduces the core components necessary to build an effective and efficient simulation capability. The course then provides a framework for building and implementing a simulation strategy targeted at achieving the organisations product and business goals.



• Goals for Modelling and Simulation [M&S]

• Leadership Questions for M&S

• Trends and Challenges Affecting your M&S

• The Essential Elements for Effective and Efficient M&S

• Engaging Stakeholders

• Strategy Framework

• Building your Strategy

• The Essential Elements Mini Deep Dive (Process – Methods – Tools)

• What’s your Status? - Maturity Assessment

• Improvement Actions

• Strategy Implementation

• Strategy Governance

Andy is a Fellow of the Institution of Mechanical Engineers. He has over 30 yrs experience in Automotive Engineering at Jaguar Land Rover, with 20 years at Senior Management level, including 10 years as Head of Engineering Simulation.

In 2019 formed PHRONESIM Ltd to provide consulting in all aspects of Engineering Simulation Strategy helping multinational organisations with; simulation capability maturity assessment, simulation strategy development, improvement roadmaps, and simulation strategy training and coaching.

Andy is a member of the NAFEMS UK Steering Committee, Assess Business Theme Committee, and the Simulation Governance and Management Working Group .

Unlocking your Organisations Simulation Capability through Maturity Assessment

Authored & Presented By
Andy Richardson (Phronesim)

Description
Engineering Simulation is a crucial capability for developing, refining and optimizing products. It is therefore, vital that organisations engineering simulation capability is fast and efficient, but can also be trusted to provide reliable results to enable important engineering decisions to be made with confidence. This is especially true with the increased use of AI techniques, as simulation is a critical source of trustworthy data used to train AI models. Is your engineering simulation capability ready?

Engineering Simulation is like a complex machine. It depends on many components working together in harmony to be effective, efficient, and reliable. To assess the status of their Engineering Simulation, organisations can benefit from conducting a Maturity Assessment. This enables them to identify their strengths, but importantly to identify areas to target improvement across.

This mini workshop will present an approach and framework that organisations can use to assess the status of their organisations engineering simulation capability, addressing the following topics:

• Organisation Goals

• Core Components of Engineering Simulation

• Simulation Maturity Assessment Types

• Approach to conducting a Maturity Assessment

• An Example Maturity Framework

• Some Practicalities

• Taking Action

• The Benefits of Conducting a Maturity Assessment

Share your experiences and learn from others via interactive mini surveys and workshop discussions:

• Your goals and challenges

• Your experiences and lessons learnt

• Benefits of maturity assessment

Andy is a Fellow of the Institution of Mechanical Engineers. He has over 30 yrs experience in Automotive Engineering at Jaguar Land Rover, with 20 years at Senior Management level, including 10 years as Head of Engineering Simulation.

In 2019 formed PHRONESIM Ltd to provide consulting in all aspects of Engineering Simulation Strategy helping multinational organisations with; simulation capability maturity assessment, simulation strategy development, improvement roadmaps, and simulation strategy training and coaching.

Andy is a member of the NAFEMS UK Steering Committee, Assess Business Theme, and the Simulation Governance and Management Working Group .

Findings From 6 Years of Applied Research: A Foundation for AI Use Cases - Illustrated with Two Practical Examples

Authored & Presented By
Christopher Woll (GNS Systems GmbH)

Abstract
In engineering, the implementation of artificial intelligence methods can be a useful tool to break the boundaries of data silos and efficiently process data in the field of simulation-based design. In addition to implementing comprehensible and efficient data management systems, identifying and capturing the relationships between the respective engineering data elements provides usable context that sustainably enriches the data used. Contextualization makes the data highly structured, making it ideal for feeding into AI models to gain deeper insights, identify patterns and support informed decision-making.

This is also the idea behind the concept of the Data Context Hub (DCH): which was developed as a research project over 6 years at Virtual Vehicle Research (Graz) together with automotive and rail OEMs. The platform brings together information from R&D and manufacturing data sources as well as from telemetry data streams or storage locations like data lakes. The DCH then creates an explorable context map in the shape of knowledge graphs from area-specific data models. These are essential for streamlining processes, reducing risks and identifying new opportunities in the data-driven development of innovative products. The use of state-of-the-art AI models also supports developers in gaining deeper insights from the data, predicting trends and automating tasks.

In the joint presentation by Context64.ai and GNS Systems, the experts will present the findings from applied research on contextual graph databases. Using two practical examples from the automotive and manufacturing industries, the experts will demonstrate how Data Context Hub helps companies transform complex data into clear, actionable insights. Use case 1 concerns buildability studies in the automotive industry, where simulation times have been significantly reduced. Due to the high complexity of the product configurations, an exponential number of possible combinations and the associated slow development cycles, conventional tools and methods proved insufficient for the company. The Data Context Hub's AI-based platform served as a virtual buildability checker by providing a solution specifically for capturing, simplifying and querying complex configuration dependencies. The implemented solution presents the configuration rules to the user as nodes in a directed graph structure that can map the relationships and dependencies between different components. By consistently capturing dependencies between assembly constraints, configuration rules and special manufacturing processes, far-reaching insights can be gained that give engineers in the production process of innovative vehicles, for example, conclusions about the impairment of the rear axle by a selected battery pack. This approach allows a faster and more precise determination of the configurations that can be built, thereby shortening development times.

In a second use case, it becomes clear that the platform optimally transforms data-intensive AI applications by creating a structured, context-aware basis for information retrieval and decision-making. The use of Large Language Models (LLM) as a chatbot provides precise, relevant answers to specific user queries about certain data contexts in technical use cases. Like the currently best-known ChatGPT, the chatbot uses language queries to independently generate content in the form of correlations between the data. For example, the chatbot would respond to a request for a summary of all simulations carried out with a specific material in the last month with an interpretation of the underlying knowledge graphs. To do this, the trained model uses the internal “Memory for Your AI” (M4AI) module to search through the linked data relationships and find relevant information based on semantic similarity. This works as follows: In the first phase, a defined data model searches a large dataset for relevant information. In the following phase, retrieval augmented generation models (RAG) are used to optimize the initial search. The model then identifies relevant information based on semantic similarity and not by searching for exact matches. The information obtained in this way flows into the subsequent generation process. This enables users to better understand the answers provided and make decisions more quickly.

The concept behind the Data Context Hub combines the ability to retrieve information from a large database with the generative capabilities of state-of-the-art AI technologies such as LLMs in a pioneering way in the field of simulation-driven design.

Biography
- Christopher Woll completed his studies with Diplom-Ingenieur (BA) for Information Technology and Master of Science in Computational Sciences in Engineering - Working for GNS Systems since 2004 - Subject area: HPC specialist and CAE-IT consultant, expert for innovative solutions for virtual product development - Extensive experience with well-known international customers from the industrial manufacturing, automotive and life science - Current position: Christopher Woll has been managing director of GNS Systems GmbH since 2015.

CFD Workflow Automation with Generative AI and Specialized Approaches for Storing and Querying Data

Authored & Presented By
Maria Bonner (Siemens)

Abstract
In the dynamic field of Computational Fluid Dynamics (CFD), the integration of Generative AI and Large Language Model (LLM) agents presents significant opportunities for workflow automation and enhancing efficiency. By employing specialized approaches for storing and querying source data, such as vector databases and graph-based databases, overall precision can be further improved. This presentation explores the innovative application of Generative AI combining it with advanced data retrieval mechanisms from vector databases and graph-based databases to automate the CFD workflow. It specifically focuses on the use case of employing LLM agents for Java macro generation within the Simcenter STAR-CCM+ tool.

By leveraging traditional Retrieval-Augmented Generation (RAG) techniques in conjunction with the GraphRAG technique, we can significantly enhance the accuracy of CFD workflow processes in a CFD tool. Extraction of sources into structured data plays an important role in this integration, providing a structured and formalized source of engineering knowledge. These structures capture and interlink diverse data points, enabling access to relevant information and facilitating the generation of Java macros, which stand behind a workflow process in a CFD tool. The combination of knowledge graphs with AI-driven techniques accelerates and brings more precision to the retrieval process.

Our presentation will begin by outlining the fundamental concepts of the RAG technique, knowledge graph technology, and its application in the CFD workflow automation. We will demonstrate how AI, when fed with structured data, can rewrite deprecated code, generate, modify and extend Java macros for a CFD specific tool. This not only improves the efficiency of the workflow but also ensures that the generated macros are up-to-date with the latest releases. Additionally, we will showcase real-world examples, where these techniques have shown the promises.

We will discuss the implications of this approach, focusing on the quality of data and the preprocessing stage, and its impact on the accuracy and reliability of the final result. Emphasizing the importance of data integrity, we will explore how ultracareful preprocessing can prevent errors and enhance the overall performance of the AI models. Finally, we will share feedback from users of our very first Copilot prototype, highlighting their experiences and the tangible benefits observed in their CFD workflows. This feedback will provide valuable insights into the practical applications and future potential of integrating GenAI with workflow automation processes.

Biography
Maria Bonner is Solution Architect for Semantic Web, AI-based and MBSE solutions at Siemens Digital Industries in Nuremberg. With a background in mathematics and computer science, Maria worked on various topics related to requirement-to-model, verification and simulation methods. Throughout her professional journey at Siemens, Maria contributed to the enhancement of Siemens products by incorporating AI-based methods into their features. Maria's educational achievements include a Diploma in Mathematics from the Novosibirsk State University in 2006, followed by a PhD in Computer Science from the University of Paderborn in 2011.

Back to Basics for CAE: Demystifying Input Files for and with Generative AI

Presented By
Subham Sett (Hexagon)

Authored By
Subham Sett (Hexagon)
Curtis (Brody) Kendall (Northwestern University)

Abstract
Finite element analysis (FEA) has been a pillar of computer-aided engineering (CAE) since the 1960s. The longevity of this simulation technology presen¬ts both opportunities as well as unique challenges for leveraging generative AI for both new and experienced users. While there are mushrooming large language model (LLM) based assistants available for such users, many of the assistants operate at a superficial level based on training documents and manuals; they are not sufficient for supporting a user intent on understanding, debugging and quickly resolving issues that lie at the input file level, the gateway to the FEA solver.

In this context, there are three unique challenges to leveraging generative AI. First, the inputs to FEA solvers are text-based with syntaxes, definitions and descriptions exposed through keywords that follow a non-intuitive, unique taxonomy and can be documented in manuals that can have thousands of pages. Next, these input file formats never adapted beyond the original implementation intended for punch cards which allow input to be unsorted but create the burden of ID management. This approach is in direct conflict with modern input decks that are geared towards HPC and can easily be gigabytes in file size. The sorting of such a file could be specific to a pre-processor, company best practices or simply the historical build-up of a model. The unsorted nature of this directly contracts with the natural flow of human language.

In this work, we present a unique approach to solving the problem by establishing a graph representation of the input file, establishing edges based on cross-referencing ID schemes and traversing this graph. This approach enables input-file specific document retrieval and provides users of every skill-level the ability to obtain prompts and responses at a higher level (pure documentation) and deeper level (input file). Future ongoing work explores optimizing the developed algorithms that reduce token consumptions and help users leverage their compute resources in a more cost-effective manner.

In this proof-of-concept work, we combine a parser, graph, a graph traversal method, and documentation retrieval (based on the graph) to deliver an in-context Generative AI experience relative to the exact features of an input file a user is interrogating. In ongoing trials, we expect reduced onboarding times, reduced debugging times, and reduced touches of traditional documentation.

AI-driven Design Optimization of Mechanical Structures in CAx-Processes Chains

Presented By
Libin Mao (Ostfalia Hochschule f. angew. Wissenschaften)

Authored By
Libin Mao (Ostfalia Hochschule f. angew. Wissenschaften)
Martin Strube (Ostfalia, University of Applied Sciences) Mathew Mathew (Ostfalia, University of Applied Sciences) Felix Schneider (Ostfalia, University of Applied Sciences) Martin Mueller (Ostfalia, University of Applied Sciences)

Abstract
Optimisations in the product development process are very computationally and time-intensive and do not always generate the maximum possible potential. In order to improve and methodically analyse this process, automatable CAx process chains are to be linked with AI agents in order to improve the optimisation result and to accelerate the development process by reducing simulation time.

For the training process of the AI agents, a CAx process chain is used as an environment that includes a parametric-associative coupling of CAD and CAE in order to realise automated and update-stable iterations. The current object of investigation for the AI agents is the use of Deep Reinforcement Learning (DRL). In particular, the following DRL-approaches were used in the research project and examined for their usability in terms of time expenditure and optimisation quality:

• DQN + Rainbow (Deep Q-Network)

• PPO (Proximal Policy Gradient)

• DDPG (Deep Deterministic Policy Gradient)

• TD3 (Twin Delayed DDPG)

• SAC (Soft Actor Critic)

However, there are a number of challenges in training AI agents for the optimisation process, which will be discussed in the presentation. AI-based optimisations are therefore in direct competition with traditional optimisations in order to justify the greater effort required for the necessary training processes. A decisive hurdle is the data generation for the training process, which is carried out by the numerical simulation as part of the CAx process chain, which is very time-consuming. For this reason, various options such as design of experiments, simplification of the simulation models, use of surrogate models and optimisation of the workflow were investigated in order to reduce simulation times, which makes the training processes more efficient. As well-known the definition of reward functions have a major influence on the convergence of a training process.

In this case, the particular challenge lies in the fact that the trained agents are to be applied to different components. With this in mind, the reward functions must be designed using generic parameters. This offers the advantage that trained AI agents can be used as a tool for similar problems (transfer learning). The findings on the challenges mentioned will be discussed in the presentation. The AI-based optimisations will be presented using several structural-mechanical examples from automotive engineering.

Pioneering Virtual Calibration: The Role of AI in Performance Optimization

Presented By
Morgan Jenkins (Secondmind)

Authored By
Morgan Jenkins (Secondmind)
Victor Picheny (Secondmind Ltd) Henri French (Secondmind Ltd)

Abstract
In the rapidly evolving field of automotive engineering, calibration is crucial for optimizing a vehicle’s control systems to achieve peak performance, efficiency, and regulatory compliance. Traditionally, calibration has been a time-consuming, manual process. However, as modern powertrain systems become more complex and interconnected, there is a growing need for innovative approaches to keep pace with technological advancements. Powertrain Calibration Engineers are increasingly adopting virtual environments to conduct these processes, marking a pivotal shift from traditional methods.

Virtualizing the calibration process offers numerous advantages for OEMs and suppliers. By reducing reliance on physical prototypes, companies can significantly cut down on costs and enhance time efficiency. Virtual calibration provides unprecedented flexibility and scalability, allowing engineers to navigate vast parameter spaces with improved accuracy and data utilization. Additionally, this approach opens up new opportunities for enhancing safety and achieving comprehensive integration of diverse systems and subsystems.

In this talk, Morgan Jenkins, Chief Product Officer of Secondmind, will explore the transformational impact of artificial intelligence and more specifically advanced machine learning techniques on the calibration process. Machine learning models, which can predict outcomes and suggest optimal calibration settings, act as powerful tools to streamline and refine calibration tasks. The talk will delve into how optimization algorithms efficiently explore calibration spaces without the need for physical testing, bolstered by integrating advanced simulation tools to create hybrid models that accelerate the calibration timeline.

Furthermore, the presentation will discuss the use of transient and global machine learning models that extend their utility into the vehicle verification and validation phases of development. By drawing on real-world examples from EV powertrain calibration, this talk will showcase how such cutting-edge approaches dramatically accelerate development cycles, thus supporting the industry's drive for innovation and alignment with environmental goals.

Ultimately, the adoption of these virtualized strategies sets a new standard for performance optimization, positioning the industry for continued advancement and success in a digital-first landscape.

AI-driven 3D Design of Cables and Hoses

Presented By
Christine Schwarz (Noesis Solutions)

Authored By
Christine Schwarz (Noesis Solutions)
Jiajun Gu (fleXstructures Gmbh)

Abstract
The increasing complexity of modern automotive systems is driving the need for advanced solutions to optimize the design and routing of sensor cables, a critical component in ensuring vehicle safety, performance, and efficiency. Traditional manual methods, such as cable design in CAD and validation through physical prototypes are increasingly insufficient for addressing the challenges posed by intricate geometries, constrained design spaces, and stringent performance requirements. To address these challenges, this study introduces an innovative approach that integrates automation and AI-supported optimization algorithms, reducing time and resources, while enabling an efficient and intelligent design cycle.



In this case study, the workflow developed to automate cable routing while adhering to constraints such as minimum bending radii, safe distances from components, and load minimization on cables and clips, highlights the integration of automation and optimization tools with complex cable simulation software. By simulating the nonlinear behaviour of cables in the design environment, this approach eliminates the need for physical prototypes, significantly reducing design cycle times and resource consumption.



The core of this advancement lies in the application of evolutionary optimization algorithms, enabling exploration of a high-dimensional design space defined by variables such as clip positioning and cable segment lengths. The optimization workflow consistently identified efficient and feasible solutions, satisfying all constraints, and achieving significant improvements over initial manual designs.

The results illustrate substantial improvements over traditional methods. Constraints that were previously challenging to meet are addressed, demonstrating the system’s ability to manage complex requirements while ensuring compliance. This integration of automation and AI not only accelerates the design process but also improves reliability by systematically respecting all constraints.

This study highlights the potential for automation and optimization to transform the cable design process in the automotive industry. By replacing manual trial-and-error methods with data-driven workflows, engineers can achieve faster development cycles and more efficient, robust designs. This approach represents a shift in how critical components are engineered, contributing to the development of smarter and more efficient vehicle systems.

Thickness optimization of a fuel tank using ML based physics model

Presented By
Jayant Pawar (Dassault Systèmes)

Authored By
Jayant Pawar (Dassault Systèmes)
Girish Patil (Dassault Systems) Rajkumar Natikar (Dassault Systems)

Abstract
Sustainability is a critical challenge for industries today, primarily due to the extensive use of plastic in component design. Companies are actively seeking ways to minimize plastic consumption by optimizing component thickness through Finite Element Analysis (FEA). Fuel tank is an important component of a vehicle that undergoes through different loading conditions. The key challenge lies in determining the ideal thickness distribution that can withstand desired loading conditions while achieving this optimization in shortest time possible. Many design iterations might be performed during early concept design stage that increases the go to market time of any product. Extrusion blow molding is very a non-linear and un-predictive simulation that requires Finite Element Analysis expertise. All the subsequent simulations are dependent on thickness distribution achieved at the end of Extrusion Blow Molding simulation.

Recent advancements in data-driven technologies, particularly machine learning (ML), offer a promising solution to expedite the design exploration process while conserving computational resources. To address the challenge of exploring wider design space in short span of time, Dassault Systemes has developed a novel methodology that employs a parametric machine learning (ML) physics model to efficiently predict the thickness distribution after blow molding and stress contours of subsequent structural load cases. By leveraging a neural network-based model, the richness of 3D simulation results is maintained while significantly reducing execution time, facilitating quasi-interactive design exploration and optimization. The ML physics model undergoes an optimization loop to identify the optimum thickness distribution of a fuel tank to maintain sufficient thinning in EBM while maintaining safety standards of structural loading. Machine learning can help predict Finite Element Analysis validation results faster in early concept design stage to reduce overall product development cycle. This paper highlights the potential of combining advanced ML techniques with traditional simulation methods to achieve design optimization quickly and efficiently.

Silver Sponsors: 4a engineering / Altair Engineering

Description
4a engineering

The accurate representation of material behavior in finite element simulations (e.g. crash simulations) is crucial for generating reasonable predictions during virtual product development and design optimization. However, in many cases engineers do not have access to high fidelity material data or validated material cards which can be applied right away. 4a’s products and services are exactly designed to close this gap of knowledge and availability and help engineers to generate better predictions.

During this talk, specialized testing systems which are precisely designed for the purpose of generating high fidelity material data for simulations will be introduced, together with a unique holistic software solution and modelling approach which highly automates the process of generating even complex material cards based on test data. By using these methods, the time and effort to generate validated material cards can be significantly reduced. Finally, several examples of successful applications and use cases will be shown.



Altair Engineering

AI in Design and Engineering – ready to deploy

Highly engineered products improve every aspect of life, from safety and sustainability to comfort. Whether it’s cars, phones, or everyday appliances, these innovations make life easier, safer and better. Though we rarely think about the design and manufacturing behind them, each product involves detailed designing, planning, developing, and making countless competing decisions. Hundreds, if not thousands, of engineers contribute to their development and some of them will focus on speeding up the design process through virtual product development.

Engineers create large amounts of data, so it is only natural that AI is a good fit for virtual product design.  There has been significant progress made in this area, but even the most renowned AI companies have focused on anything but virtual product development. As such, impressive AI advancements, such as the latest LLMs, do not have direct applicability to virtual product development, leaving engineers to discover AI themselves.

This is why Altair has heavily invested in AI-powered engineering. We believe the future of product development lies in combining physics-based insights with data-driven learning. Using physical test and virtual design data to train AI models accelerates innovation while cutting infrastructure costs shortening the distance from ideas to product.

In this session, you will hear about our vision on how to implement AI in Design and Engineering, ready to deploy today.

Graph Neural Networks for Semantic Feature Identificaton

Presented By
Tim Newman (National Composites Centre)

Authored By
Tim Newman (National Composites Centre)
Jakub Kucera (National Composites Centre) Jon Taylor (National Composites Centre)

Abstract
Whilst computers can store, access, render and manipulate Computer Aided Design (CAD) models they have no understanding of what is being represented. Many processing steps in CAD are still completed manually. This is because a machine lacks the engineer’s tacit understanding of part features and what they are for. These manual tasks can be a significant proportion of the cost in design iteration. Enabling machine feature identification can enable the automation of downstream processes like design for manufacture.

Current research focuses on lower-level features, specifically the process used to make the feature, e.g. punch or subtraction. It does not examine how those features combine into a useful unit in the design. The purpose of this research is to demonstrate identification of high-level features using graph neural networks (GNN).

CAD designs are generated automatically, forming a range of complex parts, with a focus on the structures seen in aerospace composites. Holes, fillets, and radii are chosen as selected features and investigated. These parts were labelled and converted into step format. After further processing, they were converted into GraphML format, which is compatible with Pytorch Geometric. The features added for nodes include face orientation, count of edge type (per the step definition), the basis of the faces, the total degree and for edges; the specific curve type of that edge. The graphs are tagged at node level with binary classes, i.e hole or not hole. These classes are imbalanced with a bias away from the class of interest. This is more prevalent in smaller features as these have fewer nodes that make them up.

Models were trained to maximise against Area Under Curve Receiver Operating Characteristic (AUC RoC) and down selected to a Graph attention network architecture (GAT) over GNNs and Graph Sage. Initial results showed that the models were learning relative to the origin. New samples were generated at random positions compared to the point (0, 0, 0). The models are binary classifiers as part of a related project. The models demonstrate a high level of capability (AUC ROC ~ 99%) against the generated synthetic data. We also hand labelled a set of real CAD models for the hole feature. There was significant degradation in performance against the real CAD models (AUC ~91%).

Potential factors focus on improving the variability of the generator to generate more partially over-lapping features. Work identifying additional features would also be beneficial overall.

This paper demonstrates:

• A new mechanism for generating part models with composite features present in aerostructures

• GNN models show a great deal of promise on synthetic data with high accuracy and AUC RoC

• Demonstration of performance degradation against real data

Advanced AI driven Exploration Possibilities to Link Model Changes and Effects

Presented By
Daniela Steffes-lai (Fraunhofer SCAI)

Authored By
Daniela Steffes-lai (Fraunhofer SCAI)
Jochen Garcke (Fraunhofer SCAI) Rodrigo Iza-Teran (Fraunhofer SCAI) Tom Niklas Klein (Fraunhofer SCAI) Mandar Pathare (Fraunhofer SCAI) G. N. Devdikar (Stellantis-Opel)

Abstract
The CAE field is facing enormous innovative opportunities offered by the rapid development of artificial intelligence (AI), particularly in text and image processing, as long as the unique challenges of complex CAE data are addressed. CAE engineers foresee AI as a potential tool not only for data management but also for enhancing complex engineering design processes across development stages. This presentation examines how AI could transform CAE with novel data representation and exploration techniques. We focus on AI-driven methods for visualizing and exploring model evolution and linking it to resulting effects, enabling inference and intelligent data retrieval. Implementing AI effectively in CAE requires transparent learning algorithms capable of handling limited but complex data, like mesh functions and geometries. Off-the-shelf AI models, such as large language models (LLMs), are not directly applicable. Therefore, we discuss emerging methods tailored to CAE challenges, highlighting AI's potential in engineering product development.

First, we present the latest extensions of our Laplace-Beltrami shape feature approach (LBSFA) implemented in our software tool SimExplore to investigate similarities or exceptions in deformations and mesh functions, called events, in many simulations. The combined representation of model changes in geometry and input parameters, and results allows to set the model changes into context with the insights from simulation results analysis. This provides a fundamental basis for further post processing of results, like sensitivity and correlation analysis, up to envisioned learning of relationships. However, the engineer still needs to select mesh functions and components of interest out of a ranked list of components that are influenced by the model changes. To address this, we have investigated some more sophisticated filter and grouping methods to ease the decision on which component to look first. On the one hand, a method that filters out all parts without deformations but just influenced due to rigid body motion has proven to be very effective. On the other hand, concentrating on jumps in a development tree identifies the measures with the highest impact. In this sense, a jump means that the corresponding result jumps from one behavioral mode to another.

Second, we explore how LLMs can be leveraged for CAE data. We present a proof of concept of using a Retrieval-Augmented Generation (RAG) method together with our structured data representation to enable easy exploration of relationships within the data of a development project. The RAG approach uses a knowledge base outside the training database and thus offers an extension or specialization in domain-specific knowledge. With this approach we could reach an automatic knowledge driven preselection of important components. Further on, fast metamodels can be generated based on this selection which are an efficient alternative for data-intensive convolutional neural networks (CNNs).

Finally, we outline the different parts needed for an AI support system which learns measure-effect relationships. This contribution illustrates how far our AI driven solutions specialized for CAE applications can already reach to support the complex simulation data analysis tasks giving the engineers more resources for interpretation of results and decision making.

Biography
Dr. Daniela Steffes-lai is a senior researcher at Fraunhofer SCAI, department of numerical data-driven prediction, Germany. She graduated from University of Cologne, Germany with a PhD in Mathematics. Her current research is focused on the development of data analysis methods for CAE data. She has many years of experience in leading projects in the field of data analysis for engineering applications. Furthermore, she is the main developer of the SCAI Software product SimCompare which provides a comparison of two FE simulation results.

Empowering Organizations with Engineering Intelligence to Revolutionize Product Development

Presented By
Paul Mc Grath (Neural Concept Ltd.)

Authored By
Paul Mc Grath (Neural Concept Ltd.)
Michael Straub (AVL Deutschland GmbH)

Abstract
In today’s highly competitive landscape, engineering organizations face increasing pressure to deliver complex, high-performing products at an accelerated pace, all while managing tighter margins. To maintain an edge, teams must shorten design cycles and enhance product performance and quality simultaneously. However, traditional simulation and optimization tools often fall short of meeting the demands of today’s fast-moving production environments. Design teams struggle to fully utilize the insights from these tools and seamlessly integrate them into their workflows. Engineering Intelligence (EI) represents a significant advancement, leveraging modern AI techniques to overcome these challenges. EI enables a paradigm shift by offering multi-physics and multi-component generative design, more flexible automation, and the ability to harness historical data to expedite simulations. Adopting EI offers a transformative approach, integrating simulation, CAD, and engineering data into an intelligent, cohesive platform that accelerates the product development lifecycle from concept to market. By drawing on enterprise engineering knowledge and existing processes, EI facilitates near-autonomous design, significantly reducing development time.

At the heart of this transformation is a new breed of engineers—the CAE Data Scientists—who blend simulation expertise with data science skills.

To address the engineering challenge of accelerating multi-hour modal analysis, Neural Concept in collaboration with AVL developed a 3D deep learning surrogate model to predict results within seconds. The model takes 3D bracket geometries (used in engine compartments) as input and predicts the high-fidelity 3D stress field (in MPa) and the first four frequency modes. Using 62 training samples and 11 testing samples, a proprietary 3D deep learning architecture based on geodesic convolutions was trained to replace simulations for new geometries.

Key results demonstrate exceptional accuracy, with frequency mode errors below 0.2 Hz (well within the 1 Hz workflow threshold) and stress errors averaging below 2 MPa, with extreme value errors under 3 MPa. An ablation study revealed that as few as 10 training samples would suffice for this level of precision. By leveraging data science tools in engineering and upskilling engineers as CAE data scientists, significant gains in efficiency and design capabilities can be achieved.

A key limitation of this approach is the model’s reliance on the boundaries of its training design space, which constrains its applicability to broader scenarios. Overcoming this challenge requires not only a deeper understanding of the data and problem space but also the deployment of advanced engineering intelligence tools such as generative design and automated re-simulation. AVL is planning to implement these tools at scale, to expand the model’s robustness and adaptability, enabling it to address a wider range of use cases effectively.

Moreover, this case study highlights the transformative role of workforce upskilling in engineering.

The integration of AI into CAE workflows requires a new breed of professionals: CAE data scientists capable of bridging engineering expertise with advanced data science techniques. By fostering such skills, we can unlock the full potential of data science for a wider variety of engineering challenges. The project highlighted efficient onboarding, with a young engineer at AVL rapidly building their own engineering intelligence workflows.

Methods and Applications of Image and Sound Processing in Engineering

Authored & Presented By
Kambiz Kayvantash (Hexagon)

Abstract
Traditional engineering computations require data representing the loading, material properties and geometry. In general these are provided in form of numbers (scalars, matrices, CAD ). Recent advances in machine learning and associated technologies have allowed us to explore the feasibility and limits of various other type of data, which were not fully exploited up to now. In particular we are now capable to acquire, thanks to huge advances in the sensing technology, various other types of data such as geometrical features and forms, categorical (colors, Booleans, groups, etc.), images, CT scans, radar or lidar signals and sound. In this paper we intend to present some important applications and underlying methods allowing to conduct everyday engineering modelling tasks in a much more powerful manner resulting from various machine learning techniques allowing for increased performance and reduced cost of model preparation.

For characterization of images many solutions are at hand ranging from basic PCA (SVD) type solutions to CNN (Convolutional Neural Networks) and more recently Unet approaches which can be assimilated to deep learning. Transfer learning has also allowed us to explore existing data bases as complementary support for learning. Numerous variants of the above exist too. The only problem with these solutions lies in the fact that they are highly customized and further adjustments per application. In this paper we will present a global framework allowing for the generalization of many apparently different applications under a unified approach. In particular we shall demonstrate how various forms of data, which we refer to as information, can be handled in a systematic way, applying nearly identical feature extraction and prediction methods.

For the sake of demonstration, we shall present five different cases: 1) A structured data matrix with prediction of time dependent outcome, 2) A CAD model used for cost estimation within a CNC application, 3) An image based fault detection solution and 4) An image based stress field prediction and finally 5) A fault detection based on sound recordings.

Building Surrogate models for Physics Simulation using a no-code approach

Presented By
Asparuh Stoyanov (Key Ward)

Authored By
Asparuh Stoyanov (Key Ward)
Suman Sudhakaran (TriMech) Farid Benvidi (TriMech) Chris Duchaine (TriMech)

Abstract
This project demonstrates a no-code methodology for building surrogate models for engineering simulation. Using such methods, physics simulation analysts can tap seamlessly into the potential of surrogate models, transforming traditional simulation workflows to be more efficient and flexible. In this abstract, we present a workflow of how to use simulation result data to build a 3D surrogate model that any analyst can utilize without requiring programming skills—enhancing the usability of AI-driven simulation tools for broader adoption.

Finite Element Method (FEM) simulations are often computationally intensive and challenging to scale, especially for complex structural applications. Our methodology minimizes these resource-heavy processes with a graph-based surrogate model optimized for computational efficiency. To achieve this, we utilized automated extract, transform, and load (ETL) workflows to process the raw simulation data into a shape and format suitable for AI ingestion. We show how, through no-code data processing automation, analysts can focus on deriving insights rather than getting lost in technical details.

The dataset used comprised linear static analysis results of a Press Bench model, performed using SOLIDWORKS Simulation. Parametric variables included back height, feet width, and plate length, and the results predicted were displacement and stress. Using data processing and management tools, we first extracted and converted the surface field and volumetric field data, from the original raw format into an open-source “AI-ready” format (. csv,.vtk). This allowed us to gather all simulation data in one place to better understand the data distributions, patterns, and correlations between variables. In the next step, we cleaned the collected data while maintaining different data versions and keeping track of changes. As a final step, using the cleaned and processed dataset, we trained a Graph Neural Network. The model was trained to predict accurate stress and displacement fields within seconds (>90% accuracy), using the 3D volume mesh data as inputs. The whole process from raw data to a trained model took approximately one workday to develop. The same approach will be tested on large deformation nonlinear structural analysis.

This project demonstrates how structural simulation data can be used to build surrogate models that accelerate the design process. Advances in AI modeling tools now make these models widely accessible, enabling engineers to leverage physics simulation data without coding or deep machine learning expertise—expanding the possibilities in product design optimization.

Reshaping Simulation Data for an AI Future

Presented By
Sam Zakrzewski (Rescale)

Authored By
Sam Zakrzewski (Rescale)
Romain Klein (Rescale)

Abstract
The convergence of High-Performance Computing (HPC) and Artificial Intelligence (AI) is revolutionizing engineering simulation, marking the dawn of a data-centric era in innovation and optimization. This paper proposes a comprehensive framework for simulation data management tailored to harness the power of AI for engineering applications. The ability to efficiently manage, analyze, and extract insights from high-quality simulation data is pivotal to realizing the transformative potential of this convergence.

Central to the proposed approach is the establishment of a centralized data repository, designed to unify diverse datasets, including experimental data, numerical simulations, and third-party resources. Such a repository serves as the foundation for streamlined data organization and access, enabling engineers to manage the growing complexity of simulation data effectively. Leveraging AI-powered analytics, advanced machine learning and deep learning algorithms can be applied to these datasets to identify patterns, uncover insights, and drive data-driven decision-making. This capability significantly accelerates the design and optimization processes while reducing reliance on trial-and-error methodologies.

Custom machine learning models play a critical role in this framework, offering tailored solutions for predicting performance metrics, optimizing designs, and automating routine engineering tasks. For instance, in the automotive sector, AI-driven simulations can predict vehicle performance, fuel efficiency, and safety parameters with unprecedented accuracy. This not only shortens development cycles but also enhances the overall quality and competitiveness of the final product.

The concept of digital twins is another cornerstone of this approach. By leveraging simulation data to create high-fidelity digital replicas of physical systems, engineers can perform predictive maintenance, optimize system performance, and conduct virtual testing. These digital twins facilitate rapid design iterations, minimize prototyping costs, and accelerate time-to-market. A collaborative environment further enhances this framework, fostering seamless knowledge sharing among engineers and scientists. Such platforms encourage interdisciplinary innovation, amplifying the impact of AI-driven insights across diverse domains.

The integration of cloud-based HPC offers a flexible and cost-effective computing solution, enabling engineers to scale resources dynamically based on project demands. This ensures optimal utilization of computational power while maintaining cost-efficiency. Moreover, cloud-based storage provides scalable solutions for managing large simulation datasets, ensuring secure, centralized access for global teams.

By adopting this data-centric strategy, industries such as automotive, high tech, life sciences, and manufacturing can unlock the full potential of AI to accelerate innovation, improve product performance, and reduce development costs. This framework exemplifies the synergistic possibilities at the intersection of AI, HPC, and engineering simulation, paving the way for a transformative future in digital engineering.

Accelerating R&D Processes with AI-Driven CAE

Presented By
Simon Mayer (dAIve GmbH)

Authored By
Simon Mayer (dAIve GmbH)
Alexander Koeppe (PDTec AG)

Abstract
The European automotive industry faces immense pressure to stay competitive amid growing demands for innovation, stricter safety standards, and ambitious sustainability goals. Development cycles remain longer than those of international competitors, delaying market entry. Limited budgets and resources further amplify the need to replace physical prototypes with digital simulations and virtual testing. This requires faster, more accurate simulation workflows without compromising quality.

To address these challenges, AI is integrated into CAE workflows to accelerate simulations, improve accuracy, and reduce redundant efforts. The solution relies on three interconnected pillars that streamline the process and demonstrate how AI can be successfully implemented in practice. The first pillar is structured data management, where simulation and test data from tools such as ANSYS, NASTRAN, and Abaqus are centralized into a structured platform. We show how valuable legacy data—often scattered and underutilized—can be accessed, reused, and prepared for AI applications in a structured and automated way. The second pillar involves the use of small, task-specific AI models for pre-screening. Unlike large, generic machine learning models, these smaller, highly focused models predict outcomes based on historical simulation and test data. In our presentation, we will demonstrate how these AI models serve as an intelligent pre-screening mechanism, identifying the most promising design solutions early in the process. This allows engineers to focus computational resources effectively, reducing unnecessary simulations and redundant iterations. Participants will see how straightforward it can be to train and deploy these models with existing tools and workflows. The third pillar focuses on continuous improvement and collaboration. AI-generated predictions, along with the trained models, are integrated back into the data management system, creating a continuous improvement loop. We will showcase how engineers can reuse pre-trained models for new tasks without the need for retraining or additional data preparation. This approach significantly reduces the number of required simulations while fostering collaboration: teams across projects can access validated models, apply them to new challenges, and build upon previous insights. Our practical examples will prove that AI implementation is not a “black box,” but an accessible and tangible solution for engineering teams.

In this presentation, we will go beyond describing the solution. We will demonstrate its practical implementation step by step, proving that integrating AI into CAE workflows is not complex or exclusive to experts. By showcasing real-world examples and a concrete workflow, we will highlight how AI tools can be effectively deployed to centralize data, train task-specific models, and create a self-improving simulation ecosystem. Participants will see firsthand how AI empowers engineers to accelerate design processes, optimize resources, and deliver innovative results. The key message: applying AI is achievable and accessible for every team willing to leverage their data effectively. The Impact This AI-driven approach accelerates R&D processes by reducing unnecessary simulations, streamlining workflows, and enabling engineers to focus on high-value tasks like design exploration and system optimization. The result is shorter development cycles, greater resource efficiency, and improved decision-making without compromising quality. While AI does not replace traditional simulations, it serves as a powerful complement, enhancing speed, precision, and collaboration. By the end of the presentation, participants will not only understand the value of AI in CAE but also leave with actionable insights to implement AI-driven workflows within their own teams—proving that AI is not magic but a practical tool for transforming CAE into a strategic enabler of innovation.

Autonomous 3D-CAE Agents – Rethinking complex 3D-simulation workflows

Presented By
Dirk Hartmann (Siemens)

Authored By
Dirk Hartmann (Siemens)
Robin Bornoff (Siemens) Stefan Gavranovic (Siemens) Dirk Hartmann (Siemens)

Abstract
Today's engineering systems, whether in aerospace, automotive, or in other domains, are becoming increasingly complex. Manually setting up and running (multiphysics) 3D-simulations for these complex systems is time-consuming and error-prone. At the same time, the trend towards faster innovation cycles and the need for improved product performance require more excessive design space exploration. As the number of design iterations and thus the number of simulations increase, the effort and expertise required to setup complex simulation becomes a major limiting factor. And this trend will become even more aggravated due to the relative shortage of CAE experts, which prevents leveraging ever increasing available compute power.

While in the recent years many investments have been targeted at accelerating solver processing, pre-and post-processing workflows have remained mostly unchanged. Increasing the level of autonomation in CAE workflows offers therefore a major opportunity. The latest advancements in Generative AI have unlocked unprecedented possibilities of automation through autonomous AI Agents. AI agents are autonomous software systems that can perceive their environment, make decisions, and take actions to achieve specific goals.

Within this presentation, we demonstrate along two specific examples how such autonomous agents can be tailored to independently perform various tasks associated with the modelling, pre-and post-processing of simulation, and analysis of complex engineering systems. The first example addresses how autonomous AI agents can realize the vision of fully autonomous design validation within Computer Aided Design workflows. We demonstrate such an autonomous design validation workflow on a 3D structural simulation of a jet engine bracket. Starting from the specification given to the designer, the copilot autonomously infers loads, boundary conditions, and material properties from the geometry of the model and any accompanying textual specifications, and executes the simulation. Once the simulation is complete, the results are summarized in a report, providing non-expert users with sufficient information to determine whether the design is viable.

The second example demonstrates how Generative AI, specifically their vision capabilities, can be used for CAD part recognition and classification. This in turn is a prerequisite for automated modelling defeaturing and abstractions. Preprocessing times are then substantially reduced, especially when considering massive CAD assemblies. Furthermore, we outline how Knowledge Graph based Simulation Data Management allows to reduce hallucinations of the CAE Agents.

The concepts will be demonstrated within the Simcenter Tools along the two concrete examples but allow for a generalization also to other contexts.

This approach tremendously reduces the complexity and effort of CAE workflows ultimately allowing non-expert users to perform virtual design validation.

Interactive Search of Crash Deformation Patterns in a Database Comprising Several Hundred Full Model Crash Simulations

Presented By
Stefan Müller (Sidact)

Authored By
Stefan Müller (Sidact)
Dominik Borsotto (SIDACT GmbH) Vinay Krishnappa (SIDACT GmbH) Nouran Abdelhady (SIDACT GmbH) Kirill Schreiner (SIDACT GmbH) Tobias Weinert (SIDACT GmbH)

Abstract
In order to achieve and fulfill design and crash criteria as a central component of a vehicle development project, the adaptation and redesign of the simulation model is the central task of the engineer. These changes are validated by new simulation runs, which ideally show improved behavior. The challenge here is, on the one hand, to analyze the changes and, on the other hand, to exclude the possibility that the changes have undesired side effects. These effects can either be known but undesirable, or unknown and therefore not yet evaluated in terms of their impact on crash behavior. In both cases, it is the engineer's task to find and document all the effects of the applied changes and, if necessary, adapt the model accordingly.

Comparing two or more simulation results is time-consuming. This process can be automated using machine learning. For this purpose, a database is created, for example for a construction kit or a development tree. This database is expanded with each analysis if new crash behavior in terms of deformation or the behavior of a post variable is detected. All behaviors contained in the database thus represent the event horizon against which each new simulation is compared. If a behavior is known but unintentional, this behavior can be tracked. This means that the user is warned for every simulation that exhibits this behavior.

The manageability of the database depends heavily on its size. Thanks to data compression techniques and the reduction to the essential components of a crash result, the database only requires a fraction of the original simulation results. For a use case in which the sum of all simulation results amounted to more than 7 TB, the database is only 14 GB, which corresponds to a reduction by a factor of over 500.

But what if, for example, you have seen a certain behavior of a component in a test, but this does not match the final simulation result? Or you are interested to know whether a certain crash behavior occurred in the set of all simulations. In this case, a geometric search can be carried out in the database.

By applying Model Order Reduction (MOR) the new deformation pattern is projected to the low dimensional latent space of the simulation results available in the database. In the low dimensional space the distances between the deformation pattern of interest and all deformation patters of all simulations results for all time steps are analyzed and a similarity score is determined. This provides an overview of which models and at which time steps the behavior was similar. It is therefore possible to search for early buckling behavior as well as for the final state of a component deformation.

Even for data bases containing several hundreds of full vehicle crashes, the time of search is just a few seconds making it an interactive task.

Biography
Since 2008, working in the field of “FEMZIP: Data compression for simulation results” from 2008 - 2012 at Fraunhofer SCAI and after the Spin-off at SIDACT GmbH. 2021 PhD in Mathematics: "Compression of an array of similar crash test simulation results" at HU Berlin. Since 05/2024: CTO at SIDACT GmbH.