Research On Inspection Method Of Thermal System Based On Visible Light Image

The thermal system in the coke production process is an important part of coke production.Its normal operation not only affects the normal coke production efficiency,but may also lead to serious safety accidents and cause huge economic losses,so the intelligent inspection method of the operational status of the thermal system has become one of the hot issues of widespread concern in enterprises.At present,the thermal system is still commonly inspected manually.This method has high labor cost,low inspection efficiency and poor real-time,and is prone to mis-inspection and omission,and is not conducive to the realization of fine management of coke production.For these reasons,this paper investigates the complex engineering problem of daily automatic inspection of key equipment such as hoists and switches in the thermal system,and conducts a study of an intelligent inspection method based on visible light images.There is an interlocking relationship between the hooks on both sides of the coke tank hoist and the APS,and the main purpose of its status inspection is to identify the hooks and the APS holding brake opening and closing accurately.In this paper,we introduced Rep VGG to replace the YOLOv4 backbone,cropped network model for Rep VGG-YOLOv4,K-Means++ algorithm to re-cluster Rep VGG-YOLOv4 anchor box,and data enhancement of training samples respectively.The model recognition speed is effectively improved and the robustness of the model is enhanced while ensuring recognition accuracy.The hoist inspection system runs on the PC side.The intelligent system runs on an NVIDIA embedded terminal,limited by the inspection environment in the basement of the switch.By Rep VGG-Tiny instead of YOLOv4-Tiny backbone,and training sample data enhancement method,the number and complexity of model parameters are effectively reduced and the model robustness is improved while the recognition accuracy meets the requirements.The software program of the hoist and switch inspection system was written and deployed on PC and NVIDIA embedded terminal respectively,which realized the functions of hoist and switch working condition recognition,status judgment,voice prompting and log saving.After industrial site stage testing and laboratory testing,the test results show that the hoist inspection sub-system status judgement accuracy is 98.5% and the switch inspection sub-system status judgement accuracy is 99.1%.The detection accuracy and speed can meet the requirements of thermal system status inspection,the inspection system and the corresponding process control equipment linkage work,to achieve automatic inspection,automatic execution of hoist operation has laid the foundation.