Testing Different Ways to analyse Data from Well Sensors Using Neural Networks and Image Processing

Abstract

This thesis focuses on evaluation and testing different approaches of image recognition, neural networks and data analytics with application to the analysis of data collected from sensors installed in wells operated in the oil and gas industry. Simple image recognition algorithms are compared with a newly implemented approach for feature extraction. Two different neural networks are also described and implemented, to compare against the image recognition. Image recognition algorithms had a limited amount of success due to the sample size of images, while neural networks and feature extraction are viable methods to analyse and classify pressure transients.