Rapidly evaluate vehicle performance and make smarter decisions early in the vehicle design process
Machine Learning applications are having a profound impact on traditional timelines associated with creating predictive models during the CAE led design process. In this webinar, we will present case studies and machine learning applications that are changing the way we run vehicle dynamics simulations.
Find out how, Pratt Miller are using AI to derive more insight from their Adams simulations
At Pratt Miller modeling and simulation has always been in the forefront of our vehicle development process. Design of Experiments (DOE) and optimization are used to gain insight in design and tuning of all aspects of our vehicle design. Lately Pratt Miller have been moving more towards direct optimization, though time consuming it can often provide a satisfactory design in a relatively short period of time. There is one aspect of the DOE results that is usually lost in the optimization process; the parameter sensitivity of the result, both locally around the optimum and globally over the parameter range.
Utilizing machine learning and artificial intelligence is a novel way to look at multi-run simulations. Pratt Miller employs Lunar from CADLM to perform Reduced Order Modeling (ROM) to decrease the number of simulations while increasing the information output from the simulations. Based on an initial DOE setup and run, Lunar will learn the influence of the parameter on the results data and are able to not only replicate the whole time series for a new parameter set, it will do it in a matter of seconds even if the underlying simulations took hours or days to perform the original DOE. It will also return the parameter sensitivity for the results at any point or interval allowing for advanced insight into the system response to parameter variations. A built-in optimizer allows for very fast optimization as each iteration takes only take a few seconds.
This presentation shows how to use Adams Car to run a parameter sweep for a couple of different maneuvers, how to pull the requested data into Lunar, and study the parameter effects at critical points of the simulations. We will also discuss how we can use the Nova, the optimizer in Lunar, to perform lightning fast optimizations and how to use Quasar, the Lunar engine, to connect to external optimizers or embed a Lunar database in an external process such as a control system observer model.