Deep Learning Fault Detection Algorithm For Belt Conveyor With Multi-source Heterogeneous Data Fusion

Belt conveyor is the key equipment during the production operation of coal processing plant.With the strong continuity and high degree of automation in the production process of coal processing plant,once the equipment fails,it may lead to a major production accident if not handled in time.This unplanned downtime leads to a certain impact on the productivity and economic efficiency of the enterprise,which not only affects the continuous production of the enterprise,but also may cause major safety accidents.Therefore,the establishment of a perfect fault diagnosis system is an important guarantee to achieve production safety.This paper combines the operation of belt conveyors and proposes a technical solution to monitor belt conveyors based on multi-source data fusion to solve the problems of belt runout and abnormal operation of rollers by acquiring the video data when the belt is running and the sensor data when the rollers are running.The main research contents of this paper are as follows:(1)Design and implement a belt conveyor health status detection platform.An intelligent wireless sensor based on Lo Ra wireless transmission technology is proposed to collect data such as vibration and temperature of belt conveyor rollers in real time,and send the collected data to PC and cloud platform for real-time analysis through the communication port of upper computer to monitor and discover the abnormal operation of rollers in time.(2)A belt offset detection method based on the combination of deep learning and image processing is proposed.Firstly,the HD camera captures the image information of the belt running in real time,and uses the deep learning YOLOv5 target detection method to obtain the position of the rollers in the screen,and determines the left and right maximum limit boundaries of the belt displacement according to the spatial logic relationship in the image where the rollers are located.Then,the belt image is subjected to Canny operation using Open CV to find the edge segments,and the belt edge segments in the image are Hough transformed to label the belt edges by polynomial regression.Finally,the belt edge line segment and the edge line segment of the carrier roller are detected in real-time relative distance,and then the real-time belt offset is obtained to complete the detection of belt runout of belt conveyor.(3)Data fusion algorithm based on DDNet of dendritic neural network is proposed.The information of vibration,temperature and distance data of belt conveyor rollers during operation acquired by intelligent wireless sensors and the information after image processing are fused with multi-source data to complete the diagnosis of abnormal vibration of belt conveyor rollers and the positioning of belt motion deflection.The final processed data can be transferred to the cloud platform via 4G network for storage and visualization.Practical partial applications and some simulation experimental results show that the algorithm based on deep learning and machine vision for belt runout faults can accurately predict belt operation with an accuracy of up to 90% or more.The data fusion algorithm based on dendritic neural network DDNet can classify and locate the abnormal fault of the rollers with prediction accuracy up to 98%.The multi-source heterogeneous data fusion belt conveyor health status detection platform of coal processing plant can effectively deploy and implement the algorithms studied in this paper,and the platform can assist staff in equipment maintenance and diagnosis,thus improving the production efficiency of coal processing plant and making its production safety effectively guaranteed.