Detecção e localização de vazamentos em tubulações utilizando sistemas acústicos e redes neurais / Leak detection and location in pipelines through acoustic method and neural networks

AUTOR(ES)
FONTE

IBICT - Instituto Brasileiro de Informação em Ciência e Tecnologia

DATA DE PUBLICAÇÃO

29/02/2012

RESUMO

Pipeline networks are complex systems of ducts used nowadays for gas and chemical products transporting through long distances. They frequently cross highly populated regions, water supplies or natural reserves. Even small leaks in pipelines can lead to great losses of products and serious damages to the environment before it could be detected. With the purpose to track these leaks, this work developed a technique to detection of leaks in pipelines, of rigid or flexible nature, based on acoustic method and on analysis of pressure transients generated by leak occurrence, in order to localization and determination the magnitude of leaks by using neural artificial networks. The methodologies proposed are notated for not having impacts on the environment. Variations of pressure transients and the noise generated by leakage will be detected and analyzed in a 60m-galvanized iron pipe and in a 100m-flexible pipe operating with continuous flow of gas (air) under various operating conditions. Leakages were provoked in many positions along the two pipes and used hole of distinct magnitudes. The pressure transients and the audible noise was captured by the pressure transducer and the microphone, respective, both installed inside the pressure vessel connected to a data acquisition system at a computer. The signal generated by the microphone was amplified and also passed through a bank of band pass filters being transformed into three signals with independent amplitude, each one with a band of specific frequency of 1 kHz, 5 kHz and 9 kHz. The data acquisition software was written in C language to read and process all data. The experimental results showed that it is possible to detect leaks in pipelines, for all holes, based on a pressure transient and on acoustic methods. The dynamics of these data in time is used as input to the neural model to location and determine of the leaks magnitude, simultaneously. The method chosen for training the neural networks was the Levenberg-Marquardt with Bayesian Regularization. The results of neural models indicated successfully in the same time the location and the magnitude of the leaks.

ASSUNTO(S)

gas - vazamento detectores de vazamento sinal-ruido (acustica) redes neurais (computação) gas leak detectors signal to noise (acoustic) neural networks (computer)

ACESSO AO ARTIGO

Documentos Relacionados