Research On Detection Method Of Corrosion Defects On Inner Wall Of Superheater Pipes In Thermal Power Plants

Electricity brings endless convenience to human life.However,the energy structure of our country determines that for a long period of time in the future,coal will still be the main power generation method.Therefore,the safe operation of thermal power plants is an important guarantee for the normal operation of the entire society and deserves our attention.Corrosion of boiler pipes is the main cause of accidents in thermal power plants.The superheater is the most frequently corroded part in the pipeline,so it is necessary for us to carry out effective inspections on the superheater to avoid accidents.However,the traditional corrosion detection methods mostly rely on manual experience for investigation and cannot automatically detect corrosion areas.Based on this background,this paper proposes a deep learning method to detect the inner wall of superheater pipes in thermal power plants to automatically detect the corrosion of the inner wall.This paper takes the corrosion defect of the inner wall of the superheater as the research object.First of all,in response to the problem of too little data,we used traditional methods to amplify the collected corrosion sample images,and combined with the actual situation,we used a generative countermeasure network to expand the corrosion samples according to certain standards to meet the needs of subsequent network training.Secondly,in view of the small pixel ratio of existing Semantic Segmentation methods and insufficient accuracy of edge detail segmentation,this paper proposes an extended DeepLabv3+network model that extends the parallel ASPP module and integrates the complex decoder structure to make the original image more advanced semantics.The information can be fully extracted,and low-level semantic information at different stages is merged in the above sampling process to ensure that the corrosion image can be segmented accurately.Through experimental verification,the accuracy of the proposed network structure on the corrosion image data set reaches 78.83%,which is higher than 69.22%of the classic model DeepLabv3+and 74.56%of the extended DeepLabv3+.Finally,the designed network model is compressed by the method of pruning,and the model scale is reduced by 56%when the accuracy requirement is met.In order to facilitate the inspection of superheater corrosion defects for users of various power plants,we have compiled a corrosion image inspection software system based on actual scenes,and at the same time demonstrated the functions involved in the system.The software system written in this paper can be applied efficiently and quickly on the actual site and has certain practical value.