Estimation of Fluid Flow Rate and Mixture Composition



One of the key problems in oil production is to determine flow rate of oil/gas mixture in the pipeline. Currently available methods rely on flow meters that are expensive, require installation of dedicated tank separators, and allow to measure the flow rate from a single well at a time for a short period of time. We provided a proof-of-concept using accelerometers with acoustic range bandwidth to estimate flow rates. The key idea is that flow of mixtures inside the pipe generates acoustic range vibrations that can be sensed by an accelerometer magnetically attached to the pipe , and analytic methods can be used to estimate the flow rate from the accelerometer signal. Our approach uses inexpensive sensors that can be installed on pipes in oil wells and provide continuous estimate of flow rates for fluids extracted from the wells.

To simulate oil production conditions, we designed an experiment to collect data at two university laboratories. A closed-circuit pipeline was assembled, the flow rate and oil/air mixture composition was controlled using an oil pump and air compressor. First, we estimated autoregressive models for each condition. For evaluation, we calculated a normalized log-likelihood score using the AR coefficients. For 63 test conditions, AUC was 0.9676 across 5 folds. Second, we used a sliding window approach to convert the sensor time series to a sequence of cepstral coefficient vectors for each condition. These cepstral sequences were used as predictors in Hidden Markov Model with an average diagonal accuracy of 92.5% across 5 folds.


Knowledge of basic Machine learning principles and acoustic signals will be a plus. However this poster presentation is prepared, keeping in mind that people with no or little background in the subject will also be able to understand the problem, its solution and impact.

Content URLs:

Below link is of my publication of same work in Springer:

Please see below link for draft of poster:

Speaker Info:

Pradyumn is Senior Data Scientist with Deloitte Consulting. He has more than 4 years of experience in Consumer, Oil & Gas, Technology and Insurance industries. His predictive modeling work has addressed various business problems such as Predicting Oil Flow Rate, Customer Acquisition & Satisfaction , Pro-active Service Recovery, Sales Quota Deployment, Recommendation Algorithms, Insurance Risk Assessment, Medical Billing Integrity Detection.

Id: 1612
Section: Data Science, Machine Learning and AI
Type: Poster
Target Audience: Intermediate
Last Updated: