Detecção acústica de vazamento de gás em plataforma de petróleo offshore

Abstract

The automatic gas leak detection systems currently installed on offshore oil and gas platforms are still insufficient to achieve an appropriate autonomous detection rate. Scenarios where losses of containment occur and are only later discovered by the workers’ operational rounds, through the human senses (sight, smell or hearing), which exposes people, the environment and facilities to the risk of accidents. The present work proposes the use of the concept of Sound Event Detection to identify gas leaks using acoustic signal, through an algorithm based on Machine Learning. An acoustic binary classification model based on K-Nearest Neighbors (k-NN) was developed and tested, using extraction of three features selected by the Minimum Redundancy and Maximum Relevance (mRMR) method from fifteen possible features in the time domain and frequency. The algorithm obtained Accuracy and Precision results of 100% from tests carried out on sound samples recorded on an offshore platform.