Study On Leak Detection Method For Long-distance Liquid Pipeline

Long-distance pipeline, one of the most important transportation means, has been widely utilized around the world. However, leak in pipeline often occurs owing to different reasons. To reduce the economic losses and environmental destruction, it is necessary to study the leak detection methods. Traditional methods are not suitable for the long-distance pipeline due to their expensive price, difficult maintenance and influence on normal running, etc. That is the motivation of this thesis. The research in this thesis emphasizes on the leak identification and the leak point location according to the characteristics of the flow and the pressure at inlet and outlet of the pipeline.In this thesis, the method of characteristics is used to calculate the governing equations including the leak. It is noticed that the variation of the hydraulic parameters under the leak situation is different from those caused by opening and closing the valve at inlet and outlet. This finding provides the foundation to develop the leak identification methods in theory.According to the change law of hydraulic parameters at inlet and outlet of the pipeline, two methods for leak identification have been presented. One is a new abrupt change method, which is based on innovation theory. This new method is deduced by combining the extended sequential probability test and the innovation theory. The innovation model of the hydraulic parameters is established by prediction method of nonlinear time series based on BP neural network. Thus, the on-line leak identification can be fulfilled by monitoring the innovation series given by the innovation model. The other is the semi-fuzzy clustering method. This method is obtained through limiting the membership degree of Chen's fuzzy clustering method. This method is used to classify the hydraulic parameters at inlet and outlet of the pipeline and to detect the leak in pipeline. The probability of missing alarm of these two methods is 3.4% through the leak identification of experiments and actual tests.Two methods for locating leak point have been developeded. One is a pressure grade method based on forcasting theory of BP neural network. The other is iterative method based on the inverse transient method and pressure gradient method. In this technique, decision variables are friction factors and objective function minimizes the difference of water heads between caculated valves by waterhammer equations and measured valves at measured sites. The mean location error of two methods is respectively 12% and 4% based on experimental tests. Software for leak detection system, composed of interface program, leak monitoring program and leak point location program, was developed. The software has been installed to the oil pipeline in Pinghu oil field.To verify the developed methods, a water pipe with a length of 48.775m and a diameter of 53mm has been set up in the hydraulics laboratory of Dalian University of Technology. The change law of the hydraulic parameters at inlet and outlet has been validated under the situation of leak and valve opening and closing. The software for leak detection system is also used to identify the situation of leak and valve opening and closing respectively in the experimental pipeline.