Intelligent monitoring method for leakage defect in key facilities of LNG receiving station based on infrared thermal imaging

Infrared thermography monitoring of LNG receiving stations has the characteristics of large size of key facilities and complex site conditions, which put forward higher requirements on data cleaning, identification and location of cold leakage defects of infrared thermography. Using infrared thermal imaging technology to monitor key facilities of LNG receiving stations can characterize the correspondence between equipment operation status and surface temperature, and at the same time convey the current operation information or fault situation of key facilities of LNG receiving stations, which is important for early leakage monitoring and early warning of LNG site facilities. This paper proposed an intelligent monitoring method for leakage defects of key facilities in LNG receiving stations, which integrated data cleaning, leakage defects monitoring and intelligent identification, in order to address the problems that easily occur in the application of infrared thermal imaging monitoring technology in liquefied natural gas (LNG) receiving stations. Firstly, a cleaning method of infrared thermal imaging monitoring data based on the combination of Histogram of Oriented Gradients (HOG) and Support Vector Machine (SVM) was established, which can accurately identify the video frames of foreign objects intruding into the field of view of the lens and mark them as abnormal, reduce the interference of abnormal objects to the monitoring process, and reduce the cleaning accuracy. The cleaning accuracy rate is over 95%. Further, for the problems that the abnormal data of key facilities in LNG receiving stations are very few, which leads to misjudgment and untimely abnormality identification, and the infrared monitoring is easily affected by the surrounding environment, an abnormality monitoring method based on convolutional neural network was proposed. After comparison and analysis, the method proposed in this paper can remove the limitation of boundary setting, effectively identify the scenes where personnel enter the monitoring screen to different degrees, and identify the abnormality of another facility in the same category more accurately by learning the abnormality of a facility in the same category. That is the convolutional neural network can well identify the case of one insulation defect by learning the normal scene and the scene containing two insulation defects in advance. The storage tank is selected as the research object, and a specific convolutional neural network is constructed to identify the abnormal moments of the storage tank by training the historical data and then. The advantage is that it has good learning among different individuals of the same kind of facilities and the recognition accuracy is up to 99%.