Dimensionality Reduction and Pattern Recognition of Flow Regime Using Acoustic Data

Vahabi, Nafiseh; Selviah, David R.

doi:10.1007/978-3-030-01057-7_65

Cited by 1 publication

(1 citation statement)

References 19 publications

(16 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most studies in the field of multi-phase flow classification have mainly focused on modifying the structure and parameters of Artificial Neural Networks [18] to identify the pattern of each flow regime and have not dealt with processing the big data which is produced by their new developed sensors [15], [19]. In addition, they mainly used data that is collected from the flow loop in the laboratory environment.…”

Section: Introductionmentioning

confidence: 99%

Convolutional Neural Networks to Classify Oil, Water and Gas Wells Fluid Using Acoustic Signals

Vahabi

Selviah

2019

2019 IEEE International Symposium on Signal Processing and Information Technology (ISSPIT)

Self Cite

View full text Add to dashboard Cite

Identifying the fluid type and predicting the amount of each fluid in the fluid mixture within the well pipes are important for oil and gas production energy industry and borehole water supply. Therefore automating this process will be very valuable for the oil industry because it maximises the quality and quantity of extracted oil and reduces the cost. The current study contributes to our knowledge by addressing this important issue using machine learning algorithms. The presented paper investigates the classification algorithms that identify the fluid type in oil, water and gas pipes using acoustic signals. The datasets analysed in this study are collected from real oil, water and gas well pipes under the sea where there is no controlled environment and data contains lots of noisy signals due to unpredicted events under the sea. Data is recorded during 24 hours from Distributed Acoustic Sensors which is attached alongside the 3500 m of three well pipes: oil, water and gas. The acoustic dataset are in time-distance domain and are converted to frequency-wave number domain using 2D fast Fourier transform. The outcome of 2D fast Fourier transform is sampled and fed into Artificial Neural Networks and Conventional Neural Networks algorithms to classify each fluid type. Both algorithms are trained on three datasets (oil, gas and water) and tested on another dataset. The result of this study shows Artificial Neural Networks and Conventional Neural Networks algorithms classify the fluid type with the accuracy of 79.5% and 99.3% respectively when applied on the test dataset.

show abstract

Section: Introductionmentioning

confidence: 99%