By Travis Sheperd
Calculating fluid flows are difficult in even the simplest of cases, but as they get more complicated, the computations become unwieldy. Flow experimentation is a tedious process and is only able to extract specific information for observable cases. In an effort to address these issues, Professor John Eaton of Stanford University shared his work using magnetic resonance imaging (MRI) to understand complex fluid flows. He spoke at the University of Wisconsin-Madison on Friday, Nov. 3, 2017, as part of the Midwest Mechanics Seminar series.
Analytical models are only applicable to basic cases of fluid flow while complex flows require extensive high fidelity simulations, costing both time and money. Up to now, flow experimentation has been a rudimentary alternative to computations, normally yielding inference rather than large numerical data sets. However, Eaton’s work is changing this. His group is using MRI to study fluid flows in 3D models. The imaging detects particle movements (among other things) and can give full field data on the flow. Additionally, since it uses the same MRI machines that are in the medical field, the experiments are relatively cheap, quick, easy, and physical. The best part is that their technique extracts data in bulk, with such parameters as velocity, temperature, concentration, and more.
The one downside, which can also arguably be an upside, is that their technique produces a lot of data. In order to process all this data, Eaton’s group employs a machine learning code trained specifically on this data. This method allows for a computer to determine what is best from the experiments, compare that data to a computational model, and determine improvements to that predictive model. The group was able to apply this process to the notoriously difficult turbine coolant mixing problem. A turbine often runs at extreme temperatures and needs to be cooled via injected liquid, but the position of the coolant outlets sometimes does not work. Their MRI based studies were able to observe a displacement of the cooling fluid from where it needed to be, which was not normally predicted by other models. This provided new insight into the phenomenon and allows for the incorrect models to be updated and better turbines to be made. In the future, they plan to make better model-updating algorithms and they hope to increase the utilization of their technique.