Computational Design Optimisation of a Centrifugal Pump
Lead Research Organisation:
University of Cambridge
Department Name: Engineering
Abstract
The aim of this project is to introduce automation in the design process of a centrifugal pump, and therefore this project falls under the engineering design research area. This is achieved by introducing automatic integrated design optimization into the design process. In order to achieve this aim, there are a number of questions that must be answered.
To begin with, flow through the pump must be accurately simulated using CFD simulations. Suitable solvers must be determined that give accurate flow fields, but do not have prohibitively large computational costs. On the other hand, design optimization is most easily implemented over a reduced number of parameters, as optimization over a large set of parameters is usually not feasible. Therefore, a design vector consisting of a sufficiently small set of parameters will need to be determined.
If evaluating potential pump design using CFD simulations is significantly time-consuming, then a surrogate model will need to be determined. By evaluating a number of designs over the design space, the response of the objectives can be determined for this small number of designs. A surrogate model may then be created over this design space to 'fill in the blanks'.
Once the response of the objective functions over the design space has been created, optimization over this space can then proceed. There are a number of optimization methods that may be used; these include local and global optimizers. Both of these methods of optimization have their advantages and disadvantages, and efficacy of both methods will need to be tested for this application.
These processes must be automatically performed, which adds another level of complexity to the problem. This integrated design process must be robust and able to handle a number of potential issues that may arise during implementation. Furthermore, this may be a time-consuming process and therefore considerations must be made for the computational cost of any of the above processes.
To begin with, flow through the pump must be accurately simulated using CFD simulations. Suitable solvers must be determined that give accurate flow fields, but do not have prohibitively large computational costs. On the other hand, design optimization is most easily implemented over a reduced number of parameters, as optimization over a large set of parameters is usually not feasible. Therefore, a design vector consisting of a sufficiently small set of parameters will need to be determined.
If evaluating potential pump design using CFD simulations is significantly time-consuming, then a surrogate model will need to be determined. By evaluating a number of designs over the design space, the response of the objectives can be determined for this small number of designs. A surrogate model may then be created over this design space to 'fill in the blanks'.
Once the response of the objective functions over the design space has been created, optimization over this space can then proceed. There are a number of optimization methods that may be used; these include local and global optimizers. Both of these methods of optimization have their advantages and disadvantages, and efficacy of both methods will need to be tested for this application.
These processes must be automatically performed, which adds another level of complexity to the problem. This integrated design process must be robust and able to handle a number of potential issues that may arise during implementation. Furthermore, this may be a time-consuming process and therefore considerations must be made for the computational cost of any of the above processes.
People |
ORCID iD |
| Geoffrey Parks (Primary Supervisor) | |
| James Gross (Student) |
Publications
Gross J
(2021)
Optimization by moving ridge functions: derivative-free optimization for computationally intensive functions
in Engineering Optimization
Studentship Projects
| Project Reference | Relationship | Related To | Start | End | Student Name |
|---|---|---|---|---|---|
| EP/N509620/1 | 01/10/2016 | 30/09/2022 | |||
| 2091853 | Studentship | EP/N509620/1 | 01/10/2017 | 30/09/2021 | James Gross |
| Description | Due to the high-dimensionality of the pump design, it was found that many traditional design optimisation methods were not suitable for this problem. In particular, it was found that stochastic optimisation methods, although they give good results initially, did not get a optimal solution to a desired accuracy within our small computational budget. Moreover, we found that although polynomial models often require too many function samples to create, we could create polynomial models over a small number of linear combinations of the design parameters, allowing us to dramatically reduce the effective dimension and samples needed to construct accurate models for further optimisation. We initially tried to leverage recent results in global polynomial optimisation to optimise these models, but this lead to a negative result, as the optimisation was too costly, even for dimension reduced polynomial models. Inspired by trust region methods, we developed a novel method of high-dimensional model-based optimisation named Optimisation by Moving Ridge Functions (OMoRF) which allowed us to more effectively optimise high-dimensional problems than other model-based approaches. This work is still ongoing, but further research questions have already been developed, such as "How can this approach be extended to constrained optimisation where the constraints also require models?". |
| Exploitation Route | The benefit of this result is that the optimisation algorithm may be applied to any case where the objective is high-dimensional and gradients are unavailable. Example applications include, aerospace, machine learning, chemical engineering, and more. |
| Sectors | Aerospace, Defence and Marine |
| URL | https://www.researchgate.net/publication/338399758_Optimisation_with_Intrinsic_Dimension_Reduction_A_Ridge_Informed_Trust-Region_Method |
| Description | Computational Design Optimisation of a Centrifugal Pump |
| Amount | £50,936 (GBP) |
| Funding ID | 2091853 |
| Organisation | Engineering and Physical Sciences Research Council (EPSRC) |
| Sector | Public |
| Country | United Kingdom |
| Start | 09/2017 |
| End | 03/2021 |
| Description | Computational Design Optimisation of a Centrifugal Pump |
| Amount | £28,000 (GBP) |
| Funding ID | 1946827 |
| Organisation | Engineering and Physical Sciences Research Council (EPSRC) |
| Sector | Public |
| Country | United Kingdom |
| Start | 09/2017 |
| End | 03/2021 |