AI-Driven Engineering Models

From Raw Engineering Data to Real-Time Optimization - How is an AI model built? 🔍 This visual from Narnia Labs breaks it down beautifully, and it’s not just a generic ML pipeline. It’s specifically designed for engineering data like 3D CAD models, simulation outputs, and performance metrics. What stood out to me: → Engineering-first data pipeline: It all starts with 1D/2D/3D design or simulation data - no text or tabular shortcuts here. → AI without parameter chaos: Instead of manually defining design variables, generative models learn directly from shapes and performance outcomes. →From model to action: Once trained and tested, the AI can be deployed as an API or GUI for real-time design evaluation and optimization. 🔍 This approach is grounded in a recent peer-reviewed paper from KAIST/Narnia Labs exploring eight application scenarios for generative AI in engineering - from 3D shape generation to simulation prediction and optimization. Instead of optimizing in a high-dimensional parameter space, they compress the design into a low-dimensional latent space - which makes real-time generative optimization possible. That’s a game-changer for anyone working on simulation-heavy or geometry-rich products. 📄 Full paper: Generative AI-driven Design Optimization (Kang, JMSTA 2025) || 🔗 DOI: 10.1007/s42791-025-00097-1 #engineering #ai #generativedesign

⚡️Big step toward software defined product development: this new AI model turns raw 3D point clouds into editable code for fully programmable 3D parts 😳 A new AI model MeshCoder converts raw 3D point clouds - the kind captured by LiDAR or photogrammetry - into editable, parametric code. Not mesh files. Not STLs. Executable CAD instructions that AI systems can read, modify, and regenerate. 🗄️ 𝗜𝘁 𝗽𝘂𝘀𝗵𝗲𝘀 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗳𝘂𝗿𝘁𝗵𝗲𝗿 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝘀𝗮𝗺𝗲 𝘄𝗼𝗿𝗹𝗱 𝗮𝘀 𝘀𝗼𝗳𝘁𝘄𝗮𝗿𝗲 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴, where products 𝗮𝗿𝗲 𝗲𝘅𝗽𝗿𝗲𝘀𝘀𝗲𝗱 𝗶𝗻 𝗰𝗼𝗱𝗲, 𝘃𝗲𝗿𝘀𝗶𝗼𝗻𝗲𝗱, 𝗽𝗮𝗿𝗮𝗺𝗲𝘁𝗲𝗿𝗶𝘇𝗲𝗱, 𝗮𝗻𝗱 𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗼𝗼𝗱 𝗯𝘆 𝗔𝗜 systems at a structural level. When a model can convert a physical shape into code, it becomes possible for AI to reason about the part, modify it, and generate new versions without manual CAD work. This sits perfectly in the broader shift we're already seeing: 🔹 Geometry becomes readable and editable code, not opaque mesh files 🔹 Part variations can be generated automatically for simulation and testing 🔹 AI systems can analyze and transform components programmatically 🔹 Product development becomes more traceable, reproducible, and automatable 🔹 Designers, engineers, and AI can operate on the same representation: code There is also an amazing arXiv paper in the comments, which shows how this conversion works at the technical level. AI is getting closer to understanding physical products through their code representation, and this is an early glimpse of how future design loops might run end to end in software 👏 Dr. Dirk Alexander Molitor Christian Heining Daniel Spiess

What if you could model airflow - using ONLY AI? 🤯 If you’ve ever waited HOURS for a CFD simulation to converge,  or spent DAYS testing your design in a wind tunnel - this is for you. A team of researchers recently did something remarkable: they modeled the entire 3D velocity and pressure fields of air rising from an espresso cup, without the use of CFD simulation or using a complex sensor setup. How? 🤔 They combined two powerful tools: (1) Tomo-BOS - a technique that shows how light bends as it moves through heated air, allowing you to build a 3D temperature map. (2) Physics-Informed Neural Networks (PINNs) – AI models that don’t just fit data, but also obey physics laws like the Navier–Stokes equations. When you put them together, the AI learns how the air must have moved -predicting flow patterns, plume shape, velocity, and pressure distribution. And when they compared those predictions to real experiments, the results were astonishingly close (!) Why should you care as a mechanical engineer? (1) You could extract full flow fields from something as simple as temperature images. (2) You could rely on lighter, inexpensive CFD simulations. (3) You could Achieve faster results for convection, cooling, and airflow models. This isn’t just an academic demonstration. It’s a glimpse into how we’ll soon analyze and design physical systems - not only with solvers and sensors, but with AI that understands the physics behind them 🦾 What a time to be alive 😉 Where in your workflow would you want this kind of AI first? 👇🏼

Why AI-Native Systems Engineering Is the Next Frontier - and Why It Matters Now As systems grow ever more complex - spanning automotive, aerospace, medical devices, and advanced software - traditional tooling and manual processes simply can’t keep up. The result? Fragmented requirements, siloed data, costly rework, compliance risk, and slow innovation cycles. But we’re at a turning point. AI is no longer an “add-on” feature - it’s becoming the foundation of next-gen systems engineering workflows. Instead of stitching automation onto legacy platforms, we now have tools built from the ground up with AI at their core - enabling engineers to shift from labor-intensive coordination to strategic problem solving. One standout example is Trace.Space (https://www.trace.space/) – AI‑Native Requirements & Systems Engineering Platform - a platform that demonstrates what this new paradigm looks like in practice: AI-Driven Traceability & Risk Detection: AI continuously maps relationships between requirements, tests, designs, and changes - identifying broken links, gaps, and compliance risks before they become costly issues. Structured Collaboration at Scale: By ingesting data from PDFs, JIRA, Git, Confluence, and more, the platform creates a living trace graph that keeps teams aligned and version history transparent - hardware, software, and systems engineers working in sync. Augmentation, Not Replacement: Rather than replacing engineers, AI suggests and supports - proposing links, surfacing blockers, flagging missing coverage, and enabling engineers to focus on high-value decisions. The result? Faster cycles, stronger compliance, fewer surprises, and better outcomes - from electric vehicles to satellites and regulated software systems. This is more than automation - it’s AI-augmented engineering intelligence. If your team is still wrestling with static requirements docs, siloed data, or manual trace matrices, it’s worth asking: Is your tooling enabling your engineers to lead, or is it slowing them down? #AI #SystemsEngineering #RequirementsEngineering #DigitalEngineering #EngineeringTools #Innovation Janis Vavere, Trace.Space

AI use cases in engineering span the entire product development lifecycle and have the potential to accelerate engineering processes in a truly sustainable way. One use case that, in my view, still receives far too little attention is the AI-supported transformation from EBOM to MBOM at the interface between Engineering and Manufacturing. The handover from the Engineering BOM (EBOM), which reflects the functional and design intent, to the Manufacturing BOM (MBOM), which represents the production-ready view including assemblies, routing logic, and manufacturing constraints, is time-consuming, resource-intensive and heavily dependent on expert knowledge. Yet most manufacturing companies already possess a wealth of historical transformation data and mapping rules that have been developed over years. AI agents can leverage these assets (past EBOM→MBOM mappings, domain ontologies and mapping rules) to propose MBOM structures automatically or semi-automatically. By doing so, they can massively speed up the engineering-to-manufacturing transition and help ensure that Engineering and Manufacturing teams “speak the same language.” Typical tasks AI agents can support include: - inferring manufacturing assemblies from engineering components, - identifying missing manufacturing attributes and - validating consistency across versions and product variants. However, this requires that historical transformation data becomes machine-readable, and that ontologies, mapping rules and human-in-the-loop checkpoints are embedded into agent-based workflows. The raw data exists in many organizations! Now it’s about preparing it, structuring it and developing robust workflows to unlock these high-value use cases. The potential is enormous: faster handovers, fewer inconsistencies, less manual rework and a scalable way to capture engineering–manufacturing knowledge for future generations. Vlad Larichev | Laurin Prenzel | Rick Bouter | Jülich Sebastian | Jiangyue Zhao #AI #EBOM #MBOM #Manufacturing #Engineering #DigitalThread #ProductDevelopment #SmartManufacturing

Artificial intelligence-driven approach to accelerate discovery of directed energy deposition of GRCop-42 Our recent publication in the Virtual and Physical Prototyping (https://lnkd.in/gg42KXGJ), Impact Factor 8.8, is focused on the benefit of AI-driven approaches in process optimization for difficult-to-manufacture materials using additive manufacturing. GRCop-42 and GRCop-84 alloys are designed for high heat flux combustion chamber applications. GRCop-42 offers high thermal conductivity, high creep resistance, extended low-cycle fatigue lifetime, enhanced oxidation resistance, and high tensile strength. GRCop alloys with high thermal conductivity hinder efficient laser energy absorption, demanding higher laser powers or reduced scan speeds to maintain melt pool stability and avoid incomplete fusion. In this work, we employ an artificial intelligence (AI) driven framework to rapidly identify feasible process-parameter configurations for successful laser-directed energy deposition of GRCop-42 across a wide range of laser powers. To achieve this, we have developed an AI-guided discovery approach referred to as Bayesian Experimental design for Additive Manufacturing (BEAM). Using this BEAM process, we were able to print GRCop-42 between 500 and 700 Watt laser powers. Samples were characterized for their microstructure, phase analysis, and mechanical properties. Optical imaging of the samples shows that the distribution of the Cr2Nb varied depending on input energy density and related processing parameters. Compressive Yield strengths varied between 257 ± 31 and 332 ± 17 MPa, while the Vickers microhardness varied between 71 ± 5 HV0.2 and 142 ± 7 HV0.2. The full-text article can be accessed at https://lnkd.in/g3HZ4QuX   Full citation – Zuckschwerdt, N. W., Fadhel, A., Deshwal, A., Doppa, J. R., Bose, S., & Bandyopadhyay, A. (2026). Artificial intelligence-driven approach to accelerate discovery of directed energy deposition of GRCop-42. Virtual and Physical Prototyping, 21(1). https://lnkd.in/gbyJR7pm   #additivemanufacturing #3dprinting #wsu #metallurgy #nasa

Artificial Intelligence for Drilling Optimization In modern drilling operations, artificial intelligence is no longer an experimental concept; it is a practical engineering tool for optimizing drilling performance and reducing operational risk. The volume, frequency, and complexity of surface and downhole data generated during drilling exceed what can be effectively processed through manual interpretation in real time. AI based drilling optimization systems integrate high frequency surface data (WOB, RPM, torque, standpipe pressure, flow rate, ROP) with downhole measurements from MWD/LWD (vibration, shock, stick-slip, toolface, inclination, azimuth, gamma, resistivity). Machine learning models are trained using large historical datasets to establish correlations between drilling parameters, formation response, BHA behavior, and resulting performance. Unlike rule-based optimization methods, AI models continuously adapt to changing conditions. As formation properties, bit wear, or BHA configuration evolve, the models update their predictions and recommend optimal operating windows for WOB, RPM, flow rate, and differential pressure. This allows drilling to be maintained within mechanically stable and directionally efficient limits, rather than reacting after dysfunctions are observed. From a dynamics perspective, AI algorithms are particularly effective in early detection of drilling dysfunctions such as stick-slip, lateral and backward whirl, and bit bounce. By identifying precursor patterns in torque, RPM fluctuations, and vibration signatures, the system can flag instability before it escalates into tool damage or loss of ROP. This capability is critical for protecting motors, MWD/LWD tools, and bits, especially in hard or interbedded formations. AI-driven optimization also improves consistency and repeatability across wells. By standardizing parameter selection based on data-driven models rather than individual judgment, performance variability between crews, rigs, or shifts is reduced. This is especially valuable in pad drilling and factory style operations, where small inefficiencies are multiplied across multiple wells. From an economic standpoint, the primary value of AI lies not only in maximizing instantaneous ROP but in minimizing total well cost. Reduced NPT, fewer corrective trips, improved tool life, and better wellbore quality directly translate into lower days on well and improved drilling economics. In my experience, AI is most effective when used as a decision support system rather than an autonomous controller. It provides objective, real time insight into drilling performance and risk, allowing engineers and directional drillers to make informed, timely decisions. When properly integrated, artificial intelligence shifts drilling operations from reactive troubleshooting to predictive performance control.