E-Thesis 313 views 117 downloads
Reduced Order Modelling of Hydraulic Valves / PETER HALL
Swansea University Author: PETER HALL
DOI (Published version): 10.23889/SUthesis.63612
Abstract
There is a need for accurate models of small and highly dynamic one-way valves such as those found in engine oil systems. Such reduced order models can then be included in larger oil system models that are too complex to model at high resolution. Two one-way valves are introduced and analysed using...
| Published: |
Swansea, Wales, UK
2023
|
|---|---|
| Institution: | Swansea University |
| Degree level: | Doctoral |
| Degree name: | EngD |
| Supervisor: | Dettmer, Wulf. and Peric, Djordje. |
| URI: | https://cronfa.swan.ac.uk/Record/cronfa63612 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Abstract: |
There is a need for accurate models of small and highly dynamic one-way valves such as those found in engine oil systems. Such reduced order models can then be included in larger oil system models that are too complex to model at high resolution. Two one-way valves are introduced and analysed using a fluid structure interaction simulation. The behaviour of the valves is examined, and the resultsgenerated are then used to develop two reduced order methodologies. The first is physics based. It combines physical phenomena and steady state simulation results to build up an accurate valve model. This model provides good results but is limited by the applicability of steady state simulations to dynamic behaviour, which is itself dependent on valve geometry and fluid properties. The second approach is based on a type of continuous-time recurrent neural network. This allows training on fluid structure interaction simulations with no need for physical insight. The network is trained directly on the dynamic behaviour and accurately reproduces it. Training methods are discussed. A multiple particle swarm methodology and in situ training technique is presented. The neural network-based model is applied to a more complex valve. Network architecture and size are investigated. A comparison is done over a range of time step sizes. Finally, an example workflow is given presenting the steps to train a neural network and use the resulting reduced order model within a larger oil system model. |
|---|---|
| Keywords: |
Reduced order modelling, neural network, hydraulic value, fluid structure interaction |
| College: |
Faculty of Science and Engineering |