Lars Gräning, Thomas Most, and Sebastian Wolff from ANSYS Germany GmbH presented their work at the NAFEMS Seminar on Artificial Intelligence and Machine Learning in CAE-based Simulation. Their presentation focused on the integration of machine learning (ML) within the robust design optimization workflow using optiSLang. This approach is aimed at enhancing design understanding by investigating parameter sensitivities, reducing complexity, and generating optimal metamodels. It also addresses model calibration and design improvement for optimizing design performance and ensuring design quality and reliability. The team introduced the concept of automatic AI/ML in optiSLang, where meta-models are utilized for optimization and robust design. They discussed various iterative approaches, including manual, adaptive multi-objective optimization (MOP), and the One Click Optimizer. Their system leverages Design of Experiments (DoE) across various parameters and uses automatic ML to optimize settings, filter parameters, and extend functionality with plugins. The focus of optiSLang is on data-based parametric metamodels, enabling the full automatic creation of metamodels using AutoML in Product Data Integration and Optimization (PIDO) workflows. The team also highlighted advanced statistical post-processing for inspecting identified metamodels. Gräning and his team presented the Metamodel of Optimal Prognosis (MOP) and its application in different dimensions, from 0D (scalars) to multidimensional fields such as stress fields and deformations. The MOP approach includes various ML techniques like polynomials, Kriging, and RBF, among others, and integrates customer and partner models for a comprehensive analysis. Their presentation also covered the automatic competition and creation of metamodels, detailing the signal and field MOPs, and the use of Deep Infinite Mixture Gaussian Process (DIM-GP) for scalar, signal, and field inputs and outputs. This approach allows for a statistical model of the complete signal based on Karhunen-Loeve expansion and the approximation of scalar factors using the scalar MOP. Error Measures for Model Assessment were discussed, focusing on the Coefficient of Determination (CoD) and Coefficient of Prognosis (CoP) for scalar outputs. These measures assess the approximation quality and are essential for understanding the model's predictive capabilities.The presentation concluded with various applications of their methodology, including transient thermal analysis of circuit boards and signal approximation with DIM-GP, demonstrating how their approach can be applied to practical engineering problems.
Reference | aiml23_10 |
---|---|
Authors | Graning. L Most. T Wolff.S |
Language | English |
Type | Presentation |
Date | 25th October 2023 |
Organisation | ANSYS |
Region | DACH |
Stay up to date with our technology updates, events, special offers, news, publications and training
If you want to find out more about NAFEMS and how membership can benefit your organisation, please click below.
Joining NAFEMS© NAFEMS Ltd 2025
Developed By Duo Web Design