AI Assisted Optimization

Use of AI in pre-processing, Engineering Design, Materials and Manufacturing

Wednesday 9th & Thursday 10th December 2026 | Online Seminar

(3 hours/ each day)

08:30 (Los Angeles) | 11:30 (New York)
16:30 (London) | 17:30 (Berlin)

We know that traditional optimization approaches often struggle with expensive-to-evaluate functions, high-dimensional design spaces, and complex multi-physics problems. Meanwhile, physics-informed neural networks (PINNs) are embedding physical laws governing datasets into the learning process, enhancing information quality while enabling robust predictions even when the training data is limited: AI-assisted optimization is here.

In this 2-day online event (3 hours each day), we will cover the application of AI methodologies to different areas of engineering, including Pre-processing, Engineering Design, and Materials and Manufacturing processes.

The event will bring together leading researchers and practitioners to explore AI-assisted optimization in engineering design across manufacturing, materials science, and computational mechanics.

Topics Covered

• The use of AI in Manufacturing Optimization: Exploring the growing role of AI in manufacturing optimization and its potential to drive to greater efficiency, productivity, and operational performance
• Reliability-Based Design: Incorporating uncertainty and reliability into optimization workflow
• T​he Legal Issues: Related to the AI models - data and model ownership

 

Who Should Attend?

Simulation engineers, analysts and academics working in design optimization and exploring AI applications in CAE, and engineering managers evaluating AI-driven design

Agenda

Day 1 - Wednesday, 9th December 2026

W​elcome & Introduction
N​adir Ince, GE Vernova

16:30-17:15 GMT
AI Agents in the Simulation Loop for Automated Structural Studies
Clint Nicely & Zach Zutavern, RTX

17:15-18:00 GMT
Topology Optimization and AI: Multi-Fidelity Pathways for Physics-Based Design
H Alicia Kim, University of California San Diego

Topology optimization (TO) delivers unintuitive, high-performing designs, but multiphysics coupling, uncertainty and the cost of large-scale simulation limit its use in practice. AI and machine learning may help ease these limits, while physics-based TO offers what purely data-driven design lacks: physical consistency, sensitivities and reliable data where experiments are scarce. This talk explores how the two might work together. I will first present a modular level-set TO framework that couples with existing and legacy analysis codes, with multiphysics examples including fluid-structure interaction, thermo-fluid heat exchangers with millions of elements solved on a laptop, and battery packs for urban air mobility with phase-change materials. I will then discuss design under uncertainty, where robust TO and multi-fidelity Monte Carlo methods combine models of different cost and accuracy. This multi-fidelity structure suggests one possible pathway for integrating AI, as learned models could potentially serve as additional low-fidelity models within the framework. I will close with open questions on where learned models may help, where they may risk degrading optimization, and what is needed to integrate these tools into system-level design.

18:00-18:45 GMT
P​resentation details tbc

Day 2 - Thursday, 10th December 2026

W​elcome & Introduction to Day 2
Melike Nikbay, Istanbul Technical University

16:30-17:15 GMT
Data-Driven and Physics-Informed AI for Optimizing Materials from Microstructure to Manufacturing
Pınar Acar, Virginia Tech

The performance of engineering components depends on their microstructure, which is in turn determined by how the material is manufactured. Optimizing materials and processes therefore requires models that connect processing, microstructure, and properties across length scales. Physics-based tools such as crystal plasticity and process simulations capture these links accurately but are often too expensive to run inside optimization loops, while purely data-driven models require large datasets that are rarely available for engineering alloys. This presentation shows how data-driven and physics-informed AI can bridge this gap, with examples from aerospace alloys and additive manufacturing. First, a physics-informed neural network (PINN) framework for crystal plasticity is presented for predicting the mechanical properties of Ti-Al alloys. Next, machine learning models coupled with crystal plasticity are used to predict microstructure evolution during metal additive manufacturing, including directed energy deposition and additive friction stir deposition, running about 600 times faster than full physics-based simulations. The presentation also shows how property closures, which map all attainable property combinations of a material, turn microstructure design into an efficient optimization problem, including design under microstructural uncertainty. Finally, ongoing work on an AI-driven digital twin for polymer 3D printing is briefly introduced. The presentation closes with practical lessons for integrating these methods into industrial materials and CAE workflows.

17:15-18:00 GMT
Product Optimization Workflow Acceleration using Physics-informed AI
Keith Harris,McKinsey & Company

Engineering design has been limited by how slowly each candidate can be evaluated, so teams test a few options one after another and settle on the first design that works. Integration of physics-informed AI changes this by learning to predict physical performance in place of slow simulations, which makes closed-loop optimization practical: the system generates geometries, evaluates them across several physics domains and refines them on its own. Because manufacturability, cost and robustness become constraints inside that loop, engineers can optimize trade-offs across the whole design space at once rather than fixing problems downstream. The session demonstrates that this optimization capability is the foundation, and that once it is proven on one component it can extend to full subsystems, manufacturing and in-service operations.

18:00-18:45 GMT
AI for Manufacturing Productivity: Delivering Practical Value in Aerostructure Assembly
Windo Hutabarat, University of Sheffield & Daniel Horn, Airbus

This presentation explores how AI can improve productivity in aerostructure assembly, tracing the development from rule-based analysis to data-driven methods, including deep learning. Using work from 2006 and 2026 as reference points, it brings together research and industrial perspectives on applying these approaches to deliver practical value on the shop floor.

This online seminar is being hosted in conjunction with the NAFEMS Optimisation Technical Working Group.

Sponsorship

We would like to extend an invitation to your company to be part of this event. There are several outstanding opportunities available for your company to sponsor the seminar, giving you maximum exposure to a highly targeted audience of delegates, who are all directly involved in simulation, analysis, and design.

P​lease contact the event organiser
Jo Potts, for further information
jo.potts@nafems.org; +44 (0)1355 225688