ICE intake ports design through ROM and VOLTA

Jun 5, 2024 | 12:15 - 12:40 PM

This work was carried out by Optimad srl, a consulting engineering company, and Dumarey as a supporting R&D activity for GM, with contributions from I. F. Cozza, C. Carriero, D. Menghini and R. McAlpine.
In this work, we provide a framework for the design of intake ports for an internal combustion engine (ICE), with the aim of enabling the extensive exploration of the design variable space from early design stages, to meet increasing expectations for efficiency and emissions control, and to pursue the global goal of moving towards net-zero emissions in the transport sector.
To this end, we develop a cost-effective but accurate CFD model and use it in an automated process to generate a dataset of over a thousand fluid-dynamic solutions corresponding to different geometries and valve lifts. This data is used in romBOX to train a non-intrusive Reduced-Order Model (ROM) based on Proper-Orthogonal Decomposition (POD). By encoding the solution space through POD, we can capture the most relevant flow features of the problem and describe it using a much smaller set of degrees of freedom compared to the full CFD model. A regression model is then employed to determine the coefficients of the POD expansion and predict the flow field for unseen configurations. This allows us to evaluate the flow field over specific sections and, consequently, to derive integral quantities of interest.

In order to provide designers with a simple numerical tool capable of effectively screening a large number of promising design solutions, we publish this workflow on VOLTA, efficiently enabling the quasi real-time evaluation and analysis of different configurations even for non-expert users.

The effectiveness and limitations of this design approach are highlighted through the study of a parametric intake port problem representative of a real-life industrial application, with particular focus on the performance comparison between the proposed ROM and standard response surface methods and machine learning surrogate models trained on integral quantities.

Subscribe to our newsletter

Stay up to date with news, events, upcoming webinars and innovative applications.

This field is required.
This field must be a valid email address.

By clicking I accept the privacy policy

Thank you for your subscription!
There was a problem with your subscription

Seems there was a problem with your subscription, please check the form fields or try again later.