Recognition And Detection Of Defects In Boiler Water Wall Tube Based On Machine Vision

Defect recognition and inspection based on machine vision is a research hotspot in industrial inspection.The automatic,high-efficiency and intelligent defect identification and detection of water wall tubes of power plant boiler units are of great significance to reduce the non-stop time of the boiler units and avoid boiler safety accidents.The existing boiler water wall tube detection technology has the problems of dangerous operation,long detection cycle,high false detection rate and high detection cost.This paper presents a method for the recognition and detection of boiler water wall image defects based on machine vision,and explores a new method for automatic,efficient and intelligent recognition and detection of boiler water wall tube image defects.This paper uses deep learning image classification algorithm to realize automatic classification and recognition of boiler water wall tube defects.Constructed the boiler water wall tube defect classification and recognition data set,and used the transfer learning method to train the classification and recognition model,realized the boiler water wall defect classification and recognition model based on VGG16,Goog Le Net and Res Net101,and obtained the classification and recognition of the boiler water wall tube image defect Optimize the model.Through comparative analysis of experiments,the Res Net101 network is more fully capable of extracting and learning the characteristics of boiler water wall tube defects.It has a good classification and recognition effect on the boiler water wall tube image.The accuracy of the Res Net101 model on the test set is 97.5%,and the accuracy is 96.5.%,the recall rate was 94.3%,and the F1-score was 94.1%.In this paper,the goal detection algorithm of deep learning is used to realize the automatic detection of the defects of the boiler water wall tube.Constructed the boiler water wall tube defect target detection data set,used the transfer learning method to train the target detection model,realized the boiler water wall defect target detection model based on Faster-RCNN,SSD and YOLOV5,and obtained the boiler water wall tube video defect target Optimized model for testing.Through experimental comparison and analysis,YOLOV5 network has sufficient feature extraction and learning of the defect target feature of boiler water wall tube and has fewer model parameters.The defect target effect on the boiler water wall tube video is good.The YOLOV5 model on the test set IOU=0.5.The accuracy of m AP reaches 97.8%,and the inference speed of 3840×2160 high-definition images reaches 45.4FPS.This paper develops and implements a boiler water wall tube defect recognition and detection system and completes its engineering deployment.The Lib Torch method is used to deploy the defect recognition and defect target detection model to the algorithm server,and the Tornado server framework is used to realize the data interaction between the Web and the server.Deploy the entire system using Docker technology.Through the above three aspects of research,the boiler water wall tube identification and detection system based on machine vision is realized.By uploading the unmanned aerial vehicle to be identified on the web page to collect images,the defect classification and recognition results are returned in real time and rendered on the web page.By uploading the video collected by the drone to be detected on the web page,the defect target detection result is returned in real time and rendered on the web page.It provides an automated,efficient and intelligent detection method for the defect detection of power plant boiler water wall tubes.