Research On Recognition Method Of Safety Hazards Based On Machine Vision And Its Application In Oil And Gas Stations

Oil and gas stations refer to various well sites and functional sites used for oil and gas collection and transportation.The supporting facilities and process equipment of oil and gas stations are relatively concentrated and generally have independent functions.They are the key link in coordinating the upstream and downstream of the oil and gas industry,and are also an important part of the modern energy system.Timely and accurately discovering potential safety hazards in oil and gas stations and making corresponding decisions can reduce the occurrence of accidents from the source,which is of great significance for the safe and stable operation of oil and gas stations.However,the equipment in the oil and gas station is interconnected,and the working conditions and environment change rapidly.In this situation,the traditional methods based on manual inspection and monitoring lack continuity and intelligence.The existing detection and recognition methods based on images and videos lack adaptability in dynamic changing environment and working conditions.Meanwhile,these methods have poor detection performance for multi-scale and small targets.Therefore,this paper proposes the method for detecting and identifying potential safety hazards in oil and gas stations based on machine vision,and carries out corresponding research from four aspects: abnormal data cleaning,adaptive segmentation of potential safety hazards,intelligent identification,and fault traceability.The main work is as follows:(1)Aiming at the problem that foreign objects will occlude the subject and affect the detection and recognition results after entering the field of view of the camera,a rapid identification and cleaning method of abnormal data based on the similarity of boundary features is proposed.The mechanism to detect the crossing-boundary behavior of objects is established based on the similarity and periodicity of the gray value changes on the four sides of the image.It can automatically detect the direction and timing of foreign objects entering and exiting the boundary and analyze the magnitude of their potential impact.Compared with global analysis and trajectory tracking methods,this method reduces the amount of basic computation by 98.8%,reduces redundant computation,and achieves 100% cleaning integrity.The cleaned data reduces the interference of abnormal objects intrusion on the results,and lays the foundation for subsequent detection and identification.(2)Given the problem that it is difficult to detect potential safety hazards caused by changes in the environment and working conditions of oil and gas stations,an adaptive detection and segmentation method for safety hazards in oil and gas stations is proposed based on the visual saliency,and infrared thermal imaging data preprocessing is carried out.An adaptive superpixel segmentation algorithm for typical potential safety hazards is built to cope with the dynamic changes in the foreground and background.The proposed method realizes the detection and segmentation of potential safety hazard areas in two stages,and doesn’t need network training and prior knowledge.It shows higher recall and precision performance than traditional methods in the MSRA1000 data set and potential safety hazards detection of the LNG storage tank insulation layer.(3)In order to solve the problem of insufficient accuracy and generalization of traditional target detection and segmentation models when faced with multi-scale targets and unbalanced samples,an intelligent identification method for equipment safety hazards based on improved Mask R-CNN is proposed.By establishing a two-column feature pyramid network structure,the multi-scale fusion features of the target are extracted from two path directions,and the semantic and location features are preserved.At the same time,a random undersampling mechanism is introduced to solve the problem of class imbalance of samples.Compared with the model before the improvement,the accuracy of the proposed method in the identification of hidden dangers in the insulation layer of the storage tank and the identification of the leakage at the output end of the plunger pump has increased from 87.5% and 96.67% to 93.75%and 100%,which can be used for instance segmentation of potential safety hazards in oil and gas stations with prior knowledge.(4)The existing fault traceability methods cannot directly deal with multi-source heterogeneous information including images and ignore the influence between devices,resulting in poor accuracy of the traceability results and path inference results.Taking advantage of the clear direction of material transfer flow in oil and gas station equipment,combined with the Multilevel Flow Models and the logic gate structure of the fault tree,a fault traceability analysis method that integrates image information,operarion parameters and sensor data is proposed.At first,the system is decomposed into three levels of energy,matter,and information,and the corresponding equipment and components are represented by specific graphics.After that,by establishing the state mapping relationship between different fault types and functional nodes,the fault signal is correlated with the detection results based on the machine vision method.Finally,the corresponding reasoning rules are established to realize the visualization and intelligent calculation of the fault causal path.After faults and abnormalities in the oil and gas station equipment are detected,the method can be used to analyze the root fault,visualize the fault propagation path,and provide support for control and decisionmaking.The proposed method correctly inferred the fault propagation path in the case of the BOG recondensation system of the LNG receiving station,while the FTA method produced three wrong paths.