Application of SOM Network in Ship Pipeline Leakage Detecting

Abstract

At present, ship pipeline leakage has become a great hidden risk of safe navigation and environmental pollution, but piping detecting technology mostly focuses on long-distance oil and gas pipeline, and does a little on the complicated pipeline system, for example, ship pipeline system. The frequently-used leakage detecting of negative pressure wave method, because the frequent adjustable pump or reset valve of ship pipeline system will also produce the negative pressure wave, may easily fail to report or even misreport. In order to monitor ship pipeline leakage effectively and greatly reduce fault alarm rate and missing alarm rate, SOM network (self-organizing feature map neural network) had been used to identify leakage from different working conditions. At first, the waveform characteristics of pressure and flow signals were analyzed by kurtosis calculating to obtain condition eigenvectors. From data sampling in terms of pipe working conditions, learning samples were obtained. Accordingly, the nonlinear mapping between SOM neural network inputs and outputs were well established via training. Afterwards, ship piping leakage was detected based on input eigenve