Z-ENG: Estimating Crude Oil Properties with Nuclear Magnetic Resonance
In petrochemistry, regarding raw materials, a diversity of crude oil products is available on the market. Different crude oil products have different compositions, meaning that these products consist of different raw materials. Refinery structures are typically optimised for a certain brand of crude oil, for a particular composition of components. Purchasing the wrong type of oil can lead to problems in the production and may damage the refinery structure. For example, too much sulphur can damage the catalyzer. Lighter crudes have less amount of atmospheric residue and higher amount of light distillation products (Naphtha, ….) which can cause operation limitations if the refinery structure is not flexible enough. Generally speaking, the wrong composition leads to the loss of profit by, e.g., downtime, increased energy consumption.
Due to the reasons above, it is important to acquire reliable information about the oil to be purchased. At the moment, raw oil products come with an assay. The assay describes the essential properties of such products. An example of an assay can be found at: https://corporate.exxonmobil.com/crude-oils/crude-trading/assays-available-for-download
At the moment, the component concentrations presented in the assay are estimated with the help of gas chromatography. This technique is reliable but its application also involves relatively high costs. The goal of this project is to involve Nuclear Magnetic Resonance (NMR) to provide a more cost efficient solution while still providing high quality estimations. The estimations are calculated with machine learning (probably deep learning) techniques.
The student will work on machine learning algorithms, which involves
- data acquisition (measurement),
- preprocessing the data (cleaning, formatting),
- exploratory data analysis,
- algorithm development.