Machine learning methods are perfectly suited for typical tasks in product development: accelerating development steps, providing well-founded statements about feasibility and possible variants in a matter of seconds, e.g. in apps for non-simulation experts.

The training of the models must be based on the existing knowledge in the company and this knowledge is in the existing simulation projects. Simulations also play a crucial role in generating training data when no real test data is available.

Traceability is a key element

The possibilities that arise from this will increase exponentially. This makes it all the more important to focus on traceability and organization. If ML is to be integrated into the development process as a tool, the training data must be structured and versioned. Over time, new findings will emerge that require retraining. This must be just as traceable.  
 
The use of an AI model itself must also be traceable if decisions are made based on it.  It is essential to keep track of when which AI model was trained with which data and which decision was made with it.

Structured and traceable with simulation data management

The underlying training data, the AI models and the apps derived from them are typical CAE data. It therefore makes sense to organize them by means of a simulation process and data management system:

• Ansys Minerva specializes in mapping CAE processes, such as those required for the automated generation of training data.

• As a simulation data management system, it provides a data model that is designed to manage the large number of training sets and the associated volume of data.

• The core of the ML application in engineering is the traceability of dependencies, which data was used when and for what purpose.  

Contact us or take a look at our CADFEM webinars on the new release for further advice and answers to your questions.

All CADFEM update webinars at a glance