Agenda

Pre-Seminar Short Course: AI for Simulation Engineers

Authored & Presented By
Max Kassera (yasAI UG)

Welcome & Introduction

Presented By

Authored By

Pär-Ola Jansell (member of the NAFEMS NORDIC Steering Committee / Altair Engineering)
Albert Roger Oswald (NAFEMS

Platinum sponsor presentation: Pushing the boundaries of Innovation with AI

Authored & Presented By
Joakim Lindholm (Altair Engineering AB)

Data-based Design of a Tracking Controller for Planar Closed-loop Mechanisms

Presented By
Sanam Hajipour Talkouei (BTU Cottbus-Senftenberg)

Authored By
Sanam Hajipour Talkouei (BTU Cottbus-Senftenberg)
Dieter Bestle (Brandenburg University of Technology Cottbus–Senftenberg)

Path tracking problems have always been of engineers’ interest whether in purely mechanical or mechatronic mechanisms to follow a certain trajectory with desired velocities. Setting up a trajectory controller requires quite some kinematics and dynamics expertise, especially in case of closed-loop mechanisms. The method introduced in this paper tries to minimize expertise and deep understanding of system kinematics and dynamics by utilizing artificial intelligence (AI) methods as a substitution of such advanced knowledge. In a first example, a user-defined tracking path is processed by an artificial neural network to define the corresponding geometry of a four-bar linkage. Simultaneously, a desired velocity profile is defined by splines. An inverse kinematics model and an inverse dynamics model are established by means of neural feedforward networks. Both are trained just from simulation data without any knowledge on the internal structure of the mechanism. Together they provide feedforward control for torque control of the crank. An additional simple feedback control loop eliminates any remaining errors due to kinematic nonlinearities and inertia effects. To address the applicability of this design assistant and the control framework, a lambda robot is used as second example, which is a closed-loop mechanism with the shape of the Greek letter lambda. The concept works also for this two-degree-of-freedom system. Both examples demonstrate that the proposed concept allows almost perfect tracking without much information about the system equations or internal mechanism structure.

Gold sponsor presentation: Sidact

Event detection and cause correlation for crash simulation results

Authored & Presented By
Dominik Borsotto (SIDACT GmbH)

Reaching and fulfilling several design and crash criteria during the development process is what makes the engineer adapt and redesign the simulation model over and over again. Ideally resulting in new simulation runs with in best case improved performance, matching the intention of the applied changes. For the more demanding case of unforeseen results which do not necessarily fit to the expectations of the actual changes, methods and a workflow are being introduced here, which allow to identify the root cause of this behavior. In a first instance every new simulation run is being added into an analysis database, which is continuously being used to compare new simulations against. Previous studies have already shown that this process can assist the engineer in automatically highlighting new behavior and pin pointing the engineer to the regions of interest. Rather than only highlighting the new behavior now a second phase is being triggered additionally. In this second phase the previously detected event is being isolated and analyzed against the gathered data of the development history. The analysis methods used are based up on the Principal Component Analysis, a reduced order modelling technique. This allows not only identifying structures in the data but also correlating deformation patterns against each other. Especially the latter one is of interest for an automated process, as it allows automatically detecting and suggesting possible root causes to the engineer. As an outcome of this process the engineer receives a list of correlating parts, so that he can focus on deriving a better engineering solution to achieve a deterministic behavior, rather than searching for the root cause of the event. To provide additional information about the type of cause, as for example failure or buckling, the identified parts are also forwarded to a classification prototype. This type of classification shall assist the engineer in deriving a possible design adaptation.

Machine Learning Tools and Applications for Optimization and Design Exploration

Authored & Presented By
Dimitrios Drougkas (BETA CAE Systems SA)

Machine learning tools have become increasingly prevalent in various software applications, offering numerous advantages and transforming the way we interact with technology. The challenge in the CAE-domain is to adapt and learn from the huge amount of Simulation data plus demanding load cases and complex physics. The ML-workflow involves data collection and preparation, ML-model training and evaluation, and finally Prediction. The Prediction of scalar key values, plots and field Simulation Results give us the opportunity to use Machine learning for variety of applications. Predictors enable CAE engineers to rapidly explore various design configurations without relying on time-consuming simulations. Engineers can now optimize their processes, streamline workflows and expedite the realization of innovative designs. In this presentation we will present the latest features along with some of the most important applications.

Gold sponsor presentation: Towards AI-Empowered Design Exploration

Authored & Presented By
Jan Granlund (Dassault Systemes AB)

AI-Empowered 3D Surrogate Modelling: Case Study of Thermo-Mechanical Simulation of a Glass Table

Presented By
Jan Granlund (Dassault Systemes AB)

Authored By
Jan Granlund (Dassault Systemes AB)
Barakat Bokharaie (Velux)
Jan Granlund (Dassault Systemes)

Accelerated product development cycles required in today’s fast-paced world present both opportunities and challenges in the present industry – demanding innovative designs, sustainable manufacturing and an enhanced product experience. This sense of urgency leads to several challenges such as: design flaws, compromised performance, and higher production costs. Moreover, this coupled with the traditional development methods, which typically involve a siloed data exchange between CAD designers, CAE simulation analysts and manufacturing experts, restricts efforts towards design optimization, therefore resulting in sub-optimal products. To tackle these challenges, Dassault Systèmes’ 3DEXPERIENCE® platform seamlessly integrates the modelling and simulation processes through unifying the well-established CATIA® and SIMULIA® solutions widely adopted across industries. This paradigm, referred to as MODSIM, helps maintain a digital continuity across CAD-CAE experts throughout the development process. This promotes avenues for easier parametric model development, simulation-driven design optimization, and Design of Experiments (DoE) based data generation. Furthermore, leveraging this generated DoE data, our latest machine learning and AI-driven simulation techniques can act as surrogates to physics-based simulations. This enables accelerated simulation predictions, and therefore a rapid data-driven discovery of optimal design alternatives. In this work, through a collaborative use case with VELUX, we demonstrate how data-driven AI-empowered MODSIM on the 3DEXPERIENCE® platform helps accelerate thermo-mechanical simulation of a glass table. We demonstrate fast executing neural networks which are trained as 3D surrogates using parametric DoE data (ranging across table geometries, loads and materials) generated from physics simulations. Trained models are then deployed in a collaborative design environment for rapid exploration. This allows designers to evaluate numerous parameters within a vast design space, thereby enabling robust optimized products. By delivering high-fidelity physics predictions, which cover both transient responses and 3D fields (temperatures/displacements), this proposed technique can significantly enhance the efficiency of science-based product development. By embracing this systematic and collaborative approach on the 3DEXPERIENCE® platform, coupled with the efficiency of AI-empowered surrogate modelling technology, industries can navigate the complexities of swift timelines while ensuring the delivery of optimized, sustainable and innovative products to the market.

Memory for Your AI: Lessons learned from 6 years of applied research on contextual graph-data-bases as enabler for AI use cases

Presented By
Nikhil Mittapalli (GNS Systems GmbH)

Authored By
Nikhil Mittapalli (GNS Systems GmbH)
Christopher Woll (GNS Systems GmbH)
Marco Lah (Context64.ai GmbH)

In engineering, the implementation of artificial intelligence methods can be a useful tool for breaking down siloed data limits and efficiently processing data in the field of simulation-driven design. This is also the idea behind the concept of the Data Context Hub (DCH):, which was developed as research project over 6 years at the Virtual Vehicle Research (Graz) together with automotive and rail OEMs. due to the success, the scalable data integration and management platform is now available to the market. 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 from Context64.ai and GNS Systems, the experts will present the lessons learned from 6 years of applied research at the Virtual Vehicle Research (Graz) and two OEMs on contextual graph databases. Using two practical examples from the automotive and manufacturing industries, the experts will outline how the Data Context Hub supports companies in transforming complex data into clear, actionable insights. Use case 1 addresses buildability studies in automotive, where simulation times could be reduced in orders of magnitude. The contextual graph databases of the DCH also offer a significant approach for the use of artificial intelligence methods in engineering use cases. For example, the use of Large Language Models (LLM) can provide precise, relevant answers to concrete user queries on specific data contexts as chatbot. To accomplish this, the trained model searches the interlinked data relationships to find relevant information based on semantic similarity. This enables users to understand the answers provided better and reach faster decisions. This is presented in Use case 2. The advanced 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 LLM's in a future-oriented way in the field of simulation-driven design.

Automated Damage Localization in Carbon Fiber Composites via Machine Learning and Deep Learning

Presented By
Andre Tavares (Siemens Industry Software NV)

Authored By
Andre Tavares (Siemens Industry Software NV)
E. Di Lorenzo
B. Cornelis
B. Peeters
(Siemens Industry Software / KU Leuven)
W. Desmet
K. Gryllias (KU Leuven / Flanders Make @ KU Leuven)

The increasing use of composite materials in key industries like automotive and aerospace has intensified the focus on Non-Destructive Testing (NDT) and damage detection techniques for these components. In this context, optical measurement techniques, such as Laser Doppler Vibrometry (LDV), offer an effective experimental setup for obtaining measurements from dense point grids without adding mass to the structure. In order to analyze the large measurement campaigns with high-frequency vibration data for defect classification, Machine Learning (ML) and Deep Learning (DL) techniques hold significant promise for developing reliable and automated frameworks. In this study, various ML and DL methods were evaluated for detecting defects in a Carbon Fiber plate using frequency response data. These approaches were enhanced by incorporating simulated datasets generated through Finite Element Analysis (FEA) within a Transfer Learning framework. Since simulation data is easier to produce than experimental data, any benefits gained from its use are advantageous. The results of the damage detection process are detailed, along with a comparative analysis of the different techniques employed.

Artificial Neural Network as a Surrogate to Calibrate a Proposed Hardening Model

Presented By
Eskil Andreasson (Tetra Pak Packaging Solutions AB)

Authored By
Eskil Andreasson (Tetra Pak Packaging Solutions AB)
Md Shafiqul Islam
Nils Kopp
Shadrick Kapambwe (Blekinge Institute of Technology)
Viktor Petersson (Tetra Pak Packaging Solutions AB)

In the packaging industry, High-density Polyethylene (HDPE) is commonly modeled using elastic-plastic constitutive laws with isotropic hardening due to their computational efficiency and simplified representations. However, traditional hardening models fail to capture the complex behavior of HDPE, particularly under stable necking deformation. To address this limitation, we propose an improved hardening model for HDPE and utilize an artificial neural network (ANN) as a surrogate for model calibration. The ANN is trained on an extensive dataset consisting of virtual hardening parameter sets and their corresponding force-displacement (F-D) responses generated via finite element simulations of uniaxial tensile tests. By applying curve-fitting techniques, we reduce the highly nonlinear F-D responses to a lower-dimensional representation, enhancing computational efficiency. The calibrated hardening model demonstrates excellent predictive accuracy, enabling instant calibration for various HDPE variants with differing properties based on simple uniaxial tensile F-D data. This approach provides a robust and efficient solution for modeling the mechanical behavior of HDPE, improving model calibration efficiency and simulation accuracy.

Modelling of visco-plasticity using neural networks

Authored & Presented By
Martin Kroon (Linnaeus University)

Neural networks are used to enhance and generalise an Eulerian formulation of plasticity. Neural networks are used to model the rate-dependence and hardening of the material. Functional forms for these material properties are then replaced by neural networks. The neural network-based model is applied to both theoretical reference data as well as actual experimental data in the form of stress–strain data. Simulated annealing is used to optimise/train the neural networks. The model was able to reproduce both the theoretical reference solutions as well as the experimental data very well. An implicit FE formulation was also provided in the form of a subroutine (UMAT) in Abaqus. The implementation was applied to two 3D examples, and the implementation seems to be robust and shows nice convergence properties. Overall, the present neural network-enhanced framework seems to be promising and there is potential for further development, such as inclusion of directional hardening and a more general neural network-based treatment of rate-dependence and material hardening.

AI-Driven Design Parameter Optimization for Sheet Metal Forming Processes

Presented By
Kai Diethelm (GNS Gesellschaft für numerische Simulation mbH)

Authored By
Kai Diethelm (GNS Gesellschaft fü
r numerische Simulation mbH)
Semih Burak (RWTH Aachen University)
Koutaiba Kassem-Manthey (GNS mbH Gesellschaft für numerische Simulation mbH)
Seyed Ali Mohammadi (TU Darmstadt)
Lukas Moj (GNS Systems GmbH)
Ahmad Tarraf (TU Darmstadt)
Christian Terboven (RWTH Aachen University)
Felix Wolf (TU Darmstadt)
Christopher Woll (GNS Systems GmbH)

We present an AI-based approach for optimizing sheet metal forming processes like deep drawing. Given the desired shape of the deformed metal sheet, our methodology varies process parameters such as the applied forces and the initial thickness of the undeformed sheet to optimize objective functions like the maximum thinning and thickening of the sheet or the amount of wrinkling. Our workflow comprises five main components: a deep neural network, a finite-element simulation software system, a decision component, active-learning functionality, and a Bayesian optimization routine. The deep neural network is pre-trained, and we use it to generate a set of possible initial process parameter vectors for the given geometry. We then perform a finite-element simulation to check whether any initial value vectors have yielded a sufficiently good result. If this is not the case, the decision component considers the user's requirements, the precise objective function, the given data, and the available resources. Based on this information, it selects whether to proceed with the parameter optimization process via Bayesian optimization or active learning. The work described in this talk is supported by the German Federal Ministry of Education and Research (BMBF) under Grant No. 16ME0631.

Applications of AI in Engineering

Authored & Presented By
Gustaf Jönsson (Altair Engineering AB)

Combining Simulation with Machine Learning During Various Design Phases of Turbomachinery

Authored & Presented By
Erik Munktell (Siemens Industry Software AB)

To effectively do turbomachinery development, we want to evaluate as many designs as early in the process as possible. Taking the next steps into the future means combining this with machine learning. Since the design space can become large quickly and with many disciplines involved it can take some time. In this presentation we show how we can use machine learning algorithms to train itself in real time on the design space that is currently being evaluated with computational fluid dynamics and finite element method. There is also a large shift in using Machine learning as well traditional mathematical reduced order models to represent a digital twin. We give a short overview over current state of the art when it comes to how Siemens have applied these technologies for both design and operations opportunities as well as limitations.

Data-driven optimization of exhaust valve geometry for wear reduction

Presented By
Mohammad Haghnejat (FS Dynamics Finland Oy Ab(Espoo))

Authored By
Mohammad Haghnejat (FS Dynamics Finland Oy Ab(Espoo))
Jouko Hintikka (Wärtsilä)
Niclas Liljenfeldt (Wärtsilä)
Luc St-Pierre (Aalto University)

A well-engineered exhaust valve significantly impacts both the performance and durability of an engine. This study presents a data-driven approach to optimizing exhaust valve geometry, focusing on reducing wear loss and enhancing sealing effectiveness between the valve and seat ring in a diesel engine under thermal and structural analysis. Utilizing a design of experiments (DOE) approach, the study considers six geometric parameters of the valve and a friction coefficient to evaluate their effects on wear loss and sealing performance. The friction coefficient was identified as the most critical factor for wear loss, while the contact angle and length were key in influencing sealing between the components. Two machine learning models using random forest were developed to predict sealing effectiveness and wear volume loss. These models were then used in an optimization procedure employing a Multi-objective Genetic Algorithm (MOGA). The optimal design was selected through multi-criteria decision-making methods (MCDM), including the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) and VIKOR. The optimized model selected by TOPSIS outperformed the initial model by improving the sealing of the contact by 24% and reducing wear loss by 58% compared to the initial model. This study demonstrates the potential for significant advancements in exhaust valve design through data-driven optimization, leading to improved performance and efficiency in diesel engines.

Empowering Organizations with Engineering Intelligence to Revolutionize Product Development

Presented By
Mariana Osorio (Neural Concept Ltd.)

Authored By
Mariana Osorio (Neural Concept Ltd.)
Aleksandr Timofeev
Théophile Le Clerc (Neural Concept)

To tackle new challenges, engineers need radically new capabilities, including more effective ways to harness our computational resources. Because of their historical origin, simulation tools are not well adapted to design optimization in fast-paced production and design environments. It is extremely hard for design teams to leverage insights provided by advanced simulations and by the specialized teams who develop them. Deep Learning technologies can be used to integrate knowledge from simulation and design optimization tools in the workflow of the design engineers, instead of delegating this task to separate expert teams. In this talk, we explain how recent algorithms based on Geometric Deep Learning, allow shortcutting any simulation chain through a predictive model that outputs post processed simulation results and optimization suggestions, right from the CAD design. These models are being used in engineering companies to simplify processes and to emulate the expertise of simulation engineers in the hands of product or design engineers early in the development process. Thus, the number of iterations between teams are reduced while accelerating the design activities. We will also present recent developments and application examples in Geometry Generative Design tools, which can be combined with the predictive models, as well as real use cases of these technologies, enabling optimization within a fragment of the time necessary with traditional simulation techniques. non-exhaustive list of possible applications: heat exchanger design (cfd simulations) durability (FEA simulations, transient and static) external aerodynamics (cfd simulations) internal flows (cfd simulations, turbomachinery, pipes, etc.) PCB drop tests (transient fea simulations) process simulations (warpage prediction for injection molding, stamping, etc.)

Discussion: AI and ML in Simulation Driven Design - How do these Technologies Add Value?

Authored & Presented By
Pär-Ola Jansell (Altair Engineering AB)