Soft Matter and Complex Systems Seminar

An introduction into the use of Machine Learning techniques with raw pipeline data, giving examples into both Classification and Regression problems in industry. Description of models followed by an evaluation of results using two different types of industrial flow data: univariate time series and multi-input numerical data. The first project investigates time series classification to detect flow regimes from the change in liquid holdup over time. The flow regimes of multiphase flow can have huge effects on systems, the project focuses specifically on classifying Slug Flow which can cause large pressure drops as well as damage to components over time. The second project investigates sensor data from a multiphase flow meter. This data is typically fed into physical models to make decisions for the downstream processes from the extraction site. These flow models are limited in accuracy and rely on analytical solutions; this project looks to see if real data can be used to predict important parameters more accurately. Are data driven solutions better for industry? In what ways could Machine Learning help to improve other fluid systems.