Leveraging Machine Learning in Designing Structures

Machine Learning is applied in a variety of applications including engineering and design. One of the challenges after building a machine learning (ML) model is to make it available to a wider group of users to derive some business value out of the models. Think about the following questions.

• How is the ML model made available to other users?
• How do you track who's using your model and if they have proper authorization?
• How do you track the performance of the model if the predictions are relevant or not?
• How do you upgrade a model?
• How do you protect your intellectual property?  
• How do you set up the infrastructure and scale up resources as needed to meet increasing payloads?
• How do you control cost for model deployments?
• How can you integrate the model predictions to other applications seamlessly?

These questions and many others are encountered most frequently by the practitioners of Data Science that is addressed by MLOps - a set of tools and processes that automate and streamline machine learning model deployments. Altair SmartWorks Analytics operationalizes machine learning models at scale and simplifies model deployment.

Consider a use case of designing an overhang bridge using simulation, where the cost of manufacturing and assembling the structure is obtained from an ML model. The objective is to minimize the cost while meeting all design requirements.

Problem Statement
Design a structure meeting performance targets, minimizing cost of manufacture and assembly

Design challenges

• Part Thickness
  • Increase thickness -> Stiffer design -> Increases weight, increases cost
  • Decrease thickness -> Decreases cost -> Performance not met
• Material or Design Changes
  • Stronger or lighter material -> Higher raw material & manufacturing cost
  • Mix materials -> Assembly or fabrication cost

Solution

• Altair HyperStudy for problem set up and optimization
• ML based costing model and prediction service (deployment) in SmartWorks Anlaytics
• Structural analysis and design performance metrics evaluation in Simulation Solvers

Benefits

• No distribution of files or modules or software
• Protect proprietary information, control access
• Track usage, version control
• Monitor model performance, retrain if needed

Learn more about SmartWorks Analytics and model deployments from the video playlist.