These slides were presented at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

In recent years, the hype surrounding artificial intelligence (AI) has opened many doors, but many companies are still stuck in the orientation phase. Although they would like to jump on the AI bandwagon, they often struggle to identify valuable use cases and achieve tangible results. To overcome this phase, it is crucial to approach AI implementation in the right order: starting with a solid business case and the selection of the appropriate technology. The combination of AI with engineering expertise can lead to significant advancements. It enables a safer and faster quotation phase, it allows to accelerate time-consuming development steps and it is key to drive a "Shift Left"; approach in order to explore variants earlier in the development process. Although this sounds logical, several technical challenges still need to be addressed: -Handling a limited number of training data and various data sources from experiments and simulations -Robust validation methods for AI predictions to support decision-making -Efficient and traceable integration of AI-based tools and methods into thedevelopment process We will discuss the following methods to address these challenges using examples from fluid dynamics, process automation, and HF technology: 1. ML Tool Stochos: Using a machine learning (ML) model specifically suited for engineering applications, which includes confidence values and can process any type of numerical data. 2. Management of CAE Data for and from AI Use with Ansys Minerva: For training, existing data, tests or specially created simulation data are used. The trained AI is thereby linked to the knowledge available at the time of training and needs to be versioned. It will be used to provide predictions, optimizations, or applications for users outside the group of simulation experts. As a matter of fact the knowledge base used for training changes over time, resulting in new versions of the trained models and the resulting applications. To answer which decision was made based on which AI model and training status, CAE data organization is necessary. This starts with the structured storage of knowledge from previous CAE projects, includes traceability of the training, and extends to the provision of managed apps whose use is traceably documented. Conclusion: The successful integration of AI into engineering processes requires a strategic approach that begins with a clear business case and includes the selection of the right technology for the task. These practical examples illustrate how these challenges can be addressed, from the actual ML tool to a traceable integration into the development process.