Belt Anomaly Detection Algorithm And Application Based On Machine Vision

Machine vision technology has an important application in the current industrial inspection field.Many aspects of thermal power plants have been gradually moving toward automation and intelligence,with belt coal transport systems being an essential tool in the power generation process.In practical applications,machine vision systems can interact with the belt in a variety of ways and provide timely alarm information.Machine vision can help detect and analyze the status of the belt to prevent accidents and improve operational efficiency and equipment life.Coal belt anomaly detection system combined with multiple anomaly detection algorithms can reduce cost and improve data analysis,making it a technology and system worthy of research and promotion.In this thesis,the belt runout problem,leakage and foreign matter problems are firstly studied and analyzed.The causes of belt abnormalities and their detection are analyzed in detail;in view of this,this thesis proposes an algorithm for belt runout based on image processing.After the algorithm is processed by steps such as video priming,frame extraction,grayscale transformation,edge detection,straight line detection,line segment screening,etc.,the runout can be judged against the predetermined belt benchmark,and the original video image can be marked and warned.As far as the edge detection algorithm is concerned,various existing operators were experimentally studied,analyzed and tested against each other,and the Canny operator with the best overall performance was selected as the belt edge detection algorithm.The results show that the proposed belt runout detection algorithm has high accuracy and robustness.Since the edge extraction algorithm is affected by the surrounding environment,especially when the outdoor light pollution is heavy,this effect is not stable enough;due to the implementation of automatic threshold setting,it is better to avoid the influence of light,weather and others in the algorithm in this study;meanwhile,the speckle detection algorithm is applied in the experiment,which smoothly meets the requirements of the power plant party for the detection system function.Finally,deep learning target detection algorithm is used for automatic detection of belt foreign objects.A belt foreign object data set is constructed,as well as a target detection network based on YOLOv5.The multiple foreign object detection was reduced to the question of whether there is a foreign object,that is,except for the coal block,which is uniformly owned by the foreign object;when the test set IOU=0.5,the model m AP accuracy was 98.7%.Through the analysis and processing of the test data,it is proved that the above algorithm can meet the requirements of power plant operation standards and can realize various functions such as automatic alarm and online monitoring.Based on the proposed algorithm,this thesis completes the design and implementation of the belt abnormality detection and early warning system according to the actual requirements of the project and in accordance with the software engineering process,principles and methods.The proposed algorithm is used in the belt abnormality detection in order to rectify the abnormal situation of the belt in a timely manner,and can provide real-time warning for the detected abnormality,this system reduces the subsequent hidden danger brought by the belt abnormality.