Study On Condition Monitoring Method Of Crude Oil Storage Tank

Energy is the foundation of modern industrialization,and oil is the most widely used energy worldwide.With the increasing demand for energy in the country,more and more large oil storage tanks have been established.The increase in the number of storage tanks alleviates the pressure on energy reserves,but it also brings huge risks.It is necessary to measure and monitor its parameters and status.Realizing the monitoring of the oil level state of the oil storage tank can effectively reduce the risk of oil spill pollution and improve the storage efficiency of the oil storage tank.The early monitoring of the fire state of the oil tank can be effectively extinguished in the early stage of the fire and reduce the accident loss.Monitoring the number and distribution of oil storage tanks can provide scientific and technological support for battlefield environment analysis and natural resource supervision.In this thesis,the method of monitoring the liquid level,fire state,quantity and distribution of oil storage tanks is studied.The work is as follows :(1)The oil in the storage tank is in a high temperature and high pressure environment,and the stored liquid is flammable and explosive.The detection of the liquid level will seriously affect the service life of the sensor and have the risk of explosion.Aiming at this problem,this thesis uses a non-destructive testing technology to realize the monitoring of the oil level of the oil tank.In this method,the infrared thermal imager is installed near the oil tank to photograph the outer surface of the oil tank,and the infrared image of the oil tank is collected by non-contact.The multi-interface information of the oil tank is extracted by image processing technology to realize the measurement of the liquid level state.The experimental results show that the method can realize the monitoring of oil-water interface and oil-gas interface height of oil storage tank without contact and risk.(2)The early flame target features of oil tank fire are difficult to extract and easy to lose,and the target features are easily submerged by the background.Aiming at this problem,based on the idea of polarization self-attention and positive and negative sample balance,this thesis proposes an early flame target detection algorithm for oil tank to realize the monitoring of oil tank fire state.The algorithm introduces polarization self-attention to weight the image features to improve the saliency and recognition of the flame target.The cross-stage local connection residual module is designed to improve the feature expression ability of the network,thereby improving the feature extraction ability of the small flame target,and the Ghost convolution lightweight network structure is introduced into the module.Combined with the Focal Loss function,the positive and negative samples are balanced to promote the network to learn the flame feature information.The experimental results show that the algorithm has certain advantages in the early flame target detection task of oil tank fire.(3)In remote sensing images,the oil tank texture features are diverse,the scale is different,the surrounding background is complex,and there are a large number of small oil tank targets,which restrict the detection performance of oil tank targets.Aiming at this problem,this thesis proposes an oil tank target detection algorithm based on enhanced multi-level feature fusion and spatial attention to realize the monitoring of the number and distribution of oil tanks.The algorithm adds two multi-level feature fusion paths on the basis of PAN module to strengthen the interaction of information between deep network and shallow network.The cascade residual module is designed to improve the ability of the network to learn high-level semantic information.Introduce spatial attention to reduce the interference of background debris and other factors on the tank target.The experimental results show that the algorithm can achieve better detection results in the oil tank target detection task.