AI for Engineering: A Beginner’s Guide to Getting Started

Artificial Intelligence (AI) is revolutionizing engineering workflows by making them smarter, faster and more efficient. This beginner’s guide answers common questions about implementing AI for Engineering, from infrastructure needs to compatibility with existing tools, helping you take your first steps with confidence.

 

What is the infrastructure needed to set up a functional AI?

The easiest way to implement AI functionality is through cloud-based solutions. Most providers of AI solutions offer Software-as-a-Service (SaaS) platforms that give you access to a secure domain within their cloud environment. These platforms come with advanced security features and eliminate the need for setting up costly on-premises infrastructure.

A major reason for the preference for cloud solutions is the high cost of GPU hardware required for AI model training. By leveraging cloud services, companies can bypass the need for upfront investments in such hardware and start using AI without extensive infrastructure setup.

In summary, with cloud-based AI platforms, you can get started immediately without the need for dedicated infrastructure on your end.

 

How many simulations and data points do you need to start with AI?

The amount of data needed to train a reliable AI model depends on the specific use case. For applications with high variability or complex input parameters, larger datasets are typically required. Conversely, simpler use cases with limited design space can succeed with smaller datasets.

For example, in the case of Convolutional Neural Networks (CNNs) used in solutions like Neural Concept and Ansys SimAI, you might need a minimum of 50 simulation cases to start developing an effective AI model. However, this is just a rule of thumb, and requirements can vary based on the desired accuracy and complexity of the model.

 

How long time does it take to train the AI model?

Training an AI model, which involves computational work on GPUs, typically takes anywhere from five-six hours to several days. The duration depends on the complexity of the model and the size of the dataset.

However, the most time-intensive part of AI model development is data cleaning and preprocessing. Challenges such as inconsistent naming conventions, non-converged models, and outliers in the design space must be addressed before training can begin. Depending on the data complexity, this phase can take days, weeks, and sometimes even months.

Fortunately, once this work is completed for a model, it doesn’t need to be repeated when adding new data or refining the setup. AI platforms such as Neural Concept provide pipelines – a series of scripts and functions that capture all preprocessing operations – enabling efficient workflows for managing multiple models.

 

Will AI-tools replace traditional parametric tools like Ansys optiSLang?

AI tools do not replace traditional tools like Ansys optiSLang – they complement them. Traditional parametric tools, such as Ansys optiSLang, are very effective for working with parameter spaces, parametric models and creating surrogate models based on parameters. For many use cases, these tools are sufficient and applying an AI framework would most likely be unnecessary.

However, AI tools excel where parametric tools reach their limits. Scenarios where AI proves advantageous include:

  • Difficulty parameterizing CAD models
  • The number of significant parameters makes the DOE impractical
  • Changing input and output requirements over time

AI methods offer a next-level approach, building on the foundation of parametric workflows. You might want to start off with the parametric methods and when you hit the limits you move onto the AI methodology, building on the data and the work generated parametrically. Further, running several different parameterized workflows could be a great way to generate the data needed to train the AI-models.

 

What CAE tools do AI-tools work well with?

As most AI-solutions are agnostic to where the data is coming from, the tools work with most, if not all, CAD and CAE tools you use today. One reason for this is that the AI-models prefer to have the data on certain formats to do the computations and training efficiently, and no CAE tools are using these formats today. Hence, all data needs to be converted, regardless of the software that has been used to generate it.

 

 

Unlock the Power of AI with EDRMedeso

Whether you’re looking to optimize simulation workflows, enhance design processes, or explore entirely new possibilities, AI for Engineering offers transformative potential. With the AI solutions that we offer, Neural Concept and Ansys SimAI, you can overcome traditional limitations and drive innovation in your projects. Reach out to us if you want to discuss the possibilities with implementing AI for Engineering to your specific organization.

 

Watch our on-demand webinar

ajax-loader-image