How AI can transform product development with Rescale Data Intelligence

Too much data. Too little data. Both are a problem — and both can be solved. Many CAE teams think AI isn’t for them: “Our simulations are too different to compare.” “We don’t have enough data.” With DEP MeshWorks, we can generate, enrich and structure simulation data directly — enabling accurate ML predictors even with incomplete or inconsistent datasets. That means AI/ML for Simulation is now accessible to every engineering team, not just those with massive data lakes. Presented by @Radha Krishnan, President of Detroit Engineered Products, and @Stephan Hunkeler, PhD, Technical Manager of Dynas+ Engineering Products - Europe. 📅 Thursday, Nov 6 – 02:00 PM CET 🔗 Register here → https://lnkd.in/eW_fbv9E 💬 Where do you stand today — too much data or not enough? Dynas+ #AIML #Simulation #MeshWorks #Engineering #DEP #Webinar #Innovation

Adaptive Grasp Planning with Hierarchical Reinforcement Learning for Varied Industrial Gripper Configurations This research proposes a novel approach to adaptive grasp planning in industrial robotic arms by leveraging hierarchical reinforcement learning. The system dynamically adapts grasp strategies based on the specific gripper configuration and object geometry, enabling robust and efficient manipulation across diverse manufacturing environments. We achieve a 30% improvement in grasp success rate compared to traditional methods, with substantial potential to reduce downtime and increase throughput in automated assembly lines. Our rigorous methodology incorporates a modular training pipeline, real-time simulation, and validation through physical robot experiments, culminating in a scalable and readily deployable solution. Introduction Related Work Proposed Methodology Our system employs a two-level HRL architecture. The high-level policy selects a grasp strategy from a predefined set, while the low-level policy refines the grasp execution to adapt to object variations and uncertainties https://lnkd.in/gmsptjbF

AI-Driven Automated Fiber Tension Optimization in Composite Winding Processes This paper presents a novel AI framework for real-time optimization of fiber tension during composite winding, addressing inconsistencies and defects prevalent in automated processes. Current winding systems rely on static tension profiles, failing to adapt to variations in material properties and process parameters. Our framework employs a Reinforcement Learning (RL) agent integrated with a physics-based simulation to dynamically adjust tension, guaranteeing uniform fiber distribution and improved composite structural integrity. The system is projected to improve composite part quality by 15% and reduce material waste by 8%, with a pathway to industrial adoption within 3-5 years. Composite winding is widely used to manufacture lightweight, high-strength structures such as pressure vessels, pipes, and aerospace components. Achieving consistent fiber tension during winding is critical for ensuring optimal laminate properties and preventing defects such as wrinkling, bridging, and distor https://lnkd.in/gpUxsRaj

📚 Data ingestion...Recognition...Prediction... 🤖 That’s how this machine learning (ML) model learns to think like a process. First, it ingests data, thousands of synthetic RTM injection cycles (every dot in the plot), each one describing how resin, pressure, temperature, and vacuum interact inside the mold. Then it begins recognizing patterns, understanding relations like, among others, when resin temperature rises, viscosity drops, and flow improves... Finally, it reaches prediction, learning to anticipate which parameter combinations will minimize porosity and which will trap air. 🎥 In this short simulation, the NN (neural network) learns that balance step by step: The map shifts as it learns the boundaries between OK and High-porosity zones. The white contour traces that frontier. The loss curve drops as the model improves. And in the end, a single point emerges, the optimal process window predicted by the model. Even with synthetic data, it’s fascinating to watch how ML captures composite physics: Viscosity and permeability govern flow, temperature drives cure, and pressure–vacuum balance defines porosity risk. Next natural step would be...→ feeding it real RTM manufacturing cycles to validate and refine the predictions. 📊 Model: MLP (Multi-Layer Perceptron (Scikit-Learn)) 📈 Dataset: 10.000 synthetic RTM6-based cycles 📍 Built in Jupyter Lab at My Lab #CompositeManufacturing #MachineLearning #ML #RTM #MLComposites #AerospaceEngineering #DigitalCompositeLab #MyLab

AI-Driven Predictive Maintenance Optimization for Collaborative Robotic Workcells via Dynamic Bayesian Network Integration Detailed Research Paper Abstract: This paper presents a novel framework for Predictive Maintenance Optimization (PMO) within collaborative robotic workcells, leveraging Dynamic Bayesian Network (DBN) integration and Reinforcement Learning (RL) for adaptive decision-making. The system analyzes real-time sensor data from multiple robotic agents and integrates this information with historical failure data and environmental factors to provide optimized maintenance schedules, minimizing downtime and maximizing operational efficiency. This framework offers a 30-40% reduction in unplanned maintenance events compared to traditional time-based scheduling, yielding significant cost savings and improved workflow predictability. 1. Introduction: 2. Related Work: 3. Proposed Methodology: 3.1 Data Acquisition & Preprocessing: 3.2 Dynamic Bayesian Network Modeling: Mathematically, the DBN can be represented as: P(Xt | Xt-1, …, X0) = ∏i P(Xi,t | Parents(Xi,t)) Where: P denotes probability. Xt repre https://lnkd.in/gx8Yv_xM

Proceedings of the 16th International Modelica & FMI Conference Published - Special Highlight on Free Modelica Libraries! The Proceedings of the 16th International Modelica & FMI Conference are now published via the electronic press of Linköping Universitet. With over 1000 pages, they give a detailed account on the technical progress that was discussed at our biennial conference! In addition to papers and presentations, there is another important form to contribute to the common body of knowledge that we want to highlight now: these are the free open-source Modelica libraries. * Pharmacolibrary: This library introduces Modelica to the field of Pharmacology and thereby enables the mechanistic modeling of drug behavior and response that is essential for rational drug development and personalized therapy. * URDFModelica: The Unified Robot Description Format is commonly used within its community. This free software enables the automatic Modelica model generation from this format and thereby makes Modelica much better accessible and useful for the robotics community. * SMArtInt Library: This library offers users a straightforward and efficient approach of integrating artificial intelligence via neural networks directly into Modelica models. Furthermore, five more libraries have been handed in for award. All quality contributions that are listed here in alphabetical order: * Absolut Modelica Library: A library specialized on absorption heat exchangers. Multi Energy Systems: A library for the dynamic modeling and simulation of district heating and gas betworks. * ShipSim: a library for the assessment of ship maneuverability. Components for the propulsion system as well as for hydrostatics and hydrodynamics make it especially useful for the conceptual design phase of ship building. Toroid: This library specializes on efficient models for (magnetic) toroid flux tubes verified against FEM analysis. * VDCWorkbench: Modelica-based library that can be used as a workbench for the design, testing and validation of vehicle dynamics controllers and energy management algorithms of electric vehicles. Last but not least, there were larger and smaller updates to already well-established open-source libraries * Chemical 2.0: features an improved interface for substance properties and introduces chemokinetics that helps to avoid difficult non-linear equation systems and increases robustness. * Transient Library: Designed for integrated energy systems. Many updates including improvements for large district heating networks. DLR ThermoFluidStream: a significant update on the pump models that increases usability and accuracy. The Modelica Association want to express its highest gratitude to all authors and developers for all these highly valuable contributions to our beloved field of mathematical modeling and simulation of dynamic systems. https://lnkd.in/esFJ7EVU

Engineers today have access to more powerful tools than ever before - for design, simulation, visualization and more. But with more power comes more complexity - engineering is just as difficult as ever. This is where AI can come in - by being an assistant to the engineer, helping them iterate more quickly than before - helping them navigate challenging problems and information-dense work. Watch as Nexus, our AI Agent, helps navigate an engineer through a complicated safety test using Finite Element Analysis - making the entire process nearly 50 times faster, without any loss of accuracy.

How a Digital Twin Is Really Built Digital twin has become a buzzword — but behind it lies something far more meaningful. In R&D, a true digital twin connects what happens at the molecular scale with what happens at the plant scale. It starts with computational chemistry — molecular simulations that reveal how ingredients interact, react, or self-assemble. Those insights define parameters for physics-based models: CFD for fluid flow, DEM for particles, kinetic and thermodynamic models for transformations. Then comes data — experimental and process information that calibrate and validate these models in real time. Finally, machine learning adds adaptability — learning from deviations, predicting performance, and updating the virtual model continuously. When chemistry, physics, and data are unified, the digital twin stops being a visualization — it becomes a living model that mirrors and guides R&D decisions. That’s the real promise of digital twins: not replacing experiments, but extending how we understand and design reality. #DigitalTwin #ComputationalChemistry #HybridModelling #ProcessInnovation #DigitalR&D #AIinScience

What if your FEA or CFD simulation that took hours could be done in minutes — thanks to AI? 🚀 Engineers, here’s the game-changer you’ve been waiting for. In today’s fast-moving engineering world, speed and accuracy matter more than ever. In our latest blog, we dive into how Artificial Intelligence (AI) is transforming both Finite Element Analysis (FEA) and Computational Fluid Dynamics (CFD) simulations. Here’s what you’ll discover: - How AI can help reduce simulation turnaround times dramatically — enabling more design iterations in less time. - Ways AI supports predictive modelling, enabling engineers to forecast performance before full-scale simulation runs. - Optimisation possibilities: AI becomes the catalyst for smarter, faster simulation-driven design decisions. - Real-world implications: For industries from oil & gas to automotive, using AI in simulation workflows is becoming a competitive edge. Why this matters for you: Whether you’re managing simulation workflows, reducing cost, or aiming to shorten time-to-market, adopting AI-powered FEA/CFD isn’t just “nice to have” — it’s quickly becoming “must-have”. If you want to stay ahead in engineering, understanding this shift is critical. 🔗 Read the full blog here: https://lnkd.in/dcjUeWR8 If you’re curious about implementing AI in your simulation workflow — drop a comment or DM me, and I’d be happy to chat!