Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Data-points: 103

Flow condition: Two-phase (Air and Oil)

Max GVF, %: 60

Description:

The data set was collected at the UK National Flow Laboratory TUV-NEL, using air and a viscous oil, and provides experimental points over a wide flow range and with Gas Volume Fraction (GVF) values up to 60%. The Coriolis meter used is a 50 mm diameter Foxboro CFS-10 flowtube and CFT-51 transmitter, incorporated into the Net Oil and Gas skid arrangement.
The data set consists of ten flow lines, at approximately 0.4 kg/s, 0.6 kg/s, 0.8 kg/s, 1.0 kg/s, 1.2 kg/s, 1.6 kg/s, 2.0 kg/s, 2.4 kg/s, 2.8 kg/s, and 3.2 kg/s, covering an 8:1 turndown ratio.

Download dataset

Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Data-points: 61

Flow condition: Two-phase (Salty Water and Nitrogen)

Max GVF, %: 60

Description:

The data set was collected at the UK National Flow Laboratory TUV-NEL, using (salty) water and nitrogen on a 12 mm Foxboro flowtube, and provides experimental points over a wide flow range and with Gas Volume Fraction (GVF) values up to 60%.

The data set consists of ten flow lines, at approximately 0.1 kg/s, 0.12 kg/s, 0.14 kg/s, 0.19 kg/s, 0.23 kg/s, 0.3 kg/s, 0.37 kg/s, 0.48 kg/s, 0.61 kg/s, and 0.75 kg/s.

Download dataset

Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Data-points: 126

Flow condition: Two-phase (Oil and Nitrogen)

Max GVF, %: 65

Description:

The data set was collected at the UK National Flow Laboratory TUV-NEL, using a conventional (relatively low viscosity) oil and nitrogen on a 50 mm Foxboro flowtube, and provides experimental points over a wide flow range and with Gas Volume Fraction (GVF) values up to 65%.

The data set consists of nine flow lines, at approximately 1.5 kg/s, 1.75 kg/s, 2 kg/s, 2.5 kg/s, 3.2 kg/s, 4 kg/s, 5 kg/s, 6.15 kg/s, and 7.45 kg/s.

Download dataset

Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Data-points: 715

Flow condition: Three-phase (Light Oil, Water, and Nitrogen)

Max GVF, %: 70

Description:

The data set was collected at the UK National Flow Laboratory TUV-NEL, using a light oil, water, and nitrogen. and provides experimental points over a wide flow range and with Gas Volume Fraction (GVF) values up to 70%. The Coriolis meter used is a 75 mm diameter Foxboro CFS-10 flowtube and CFT-51 transmitter, incorporated into the Net Oil and Gas skid arrangement. The data set consists of eleven water cuts, from 100% down to approximately 0% in steps of 10%. For each water cut, a series of flowlines have been recorded, where the maximum flowrates and GVFs are selected so that the pressure drop across the flowtube does not exceed 150 kPa. The datasets have been calibrated so that, for each flowline, the mass flow and density errors have zero error at zero GVF, so that the recorded errors relate to the effects of multiphase flow.

Download dataset

Number of the analyzed dataset: 1

Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Description:

This dataset was analyzed in the paper (see the link below) in FMI. Four data sets have been modelled, as follows:
1. The full benchmark data set, with 103 data points.
2. All flow lines but excluding every other GVF point, yielding 53 data points.
3. Excluding every other flow line, yielding 51 data points.
4. Excluding every other flow line and every other GVF point for the remaining flow lines, yielding 27 points.

Four datasets, the paper and exemplary code are provided.

The paper

Download analysis

Number of the analyzed dataset: 1

Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Description:

To analyze dataset 1, a support vector machine method was used. It was compared with the results that were given by neural networks in the analysis of 1 of this dataset. The exemplary code is provided.

Download analysis

Number of the analyzed dataset: 4

Author:
Manus P. Henry
manus.henry@eng.ox.ac.uk

Description:

Three ensemble methods were applied to data set 4: Random Forest, Extra Randomized Trees, Gradient Boosting. The data was split into training and test sets of 75 and 25 percent, respectively. Parameter grids for each method are provided in the source code. The attached notebook contains the result of the selection of optimal parameters for Gradient Boosting of decision trees, which gave the best results.

Download analysis