Multi-disciplinary design optimization of heat exchangers using Gaussian processes

Abstract

Heat exchangers are important equipment in many engineering systems. They are equally complex and resource-intensive to design and manufacture, requiring several hours of both simulation and experimentation budget before reaching a design freeze. Various configurations within the sub-components or their designs make the design process even more challenging, as each of those variants will require respective analysis before reaching an optimal design. The literature also reveals that such analysis are often done in isolation and not in an integrated fashion. The first aim of this work is to perform a multi-disciplinary optimization of a charge air cooler that allows us to tweak several of its design properties within the framework, as they are often necessary for design analysis. More specifically, the framework studies the impact of using corrugated tubes against smooth tubes within the multi-disciplinary framework, highlighting the importance of using such methods over traditional ones and allowing for improvement in resources required to reach a design freeze. The disciplines applied within this framework are aero-thermal and vibrations, making this approach the first of its kind in multi-disciplinary optimization for heat exchangers. Furthermore, another important component of the design process addressed in this work are the challenges posed by experimentation and the constraints it puts on the degree of freedom to explore the design space along with the time it takes to do so. This is done using surrogate-based approaches. Namely, Gaussian processes, multi-fidelity modeling, and multi-output Gaussian processes that are used to build prediction models that help swiftly explore new tube topologies (in this case, corrugated tubes), with much control and flexibility to explore the design space effectively. This work not only recommends how novel tube designs can be tested but also reduces the burden on the design process, which traditionally is resource intensive. The models constructed for tubes are then applied to a whole heat exchanger design process, providing another layer of performance for these models that cater to the entire heat exchanger design process. Lastly, an understanding about the reuse of data is formulated using multioutput Gaussian processes, where the chosen tube topologies are themselves utilized as source and target data sources. This allows us to assess when such topological decisions are helpful and when they are not.

More information

Identifiers

ORCID for Atul Udaivir Singh: orcid.org/0000-0002-6878-6391

ORCID for David Toal: orcid.org/0000-0002-2203-0302

ORCID for Edward Richardson: orcid.org/0000-0002-7631-0377

Catalogue record

Export record