| Electric-fused magnesium has the properties of high melting point,compact structure,high crushing strength,strong corrosion resistance,stable chemical performance.It's the strategic raw materials of the metallurgy,scientific research,national defense,spaceflight areas,it has a wide prospect of application and enjoy huge potential for development.In our country,the smelting progress of high grade magnesium is smelted and recrystallized via arc furnace.In melting process of electric-fused magnesium furnace,semi-molten is an abnormal condition,if the abnormal condition is not handled in time,it will cause huge economic losses.The leakage of high temperature molten liquid will also jeopardize the safety of workers on the spot.At present,the early warning of the semi-molten condition of the electric-fused magnesium furnace mainly depends on the uninterrupted inspection of the workers on the spot.The workers observe the morphology of flame brightness,burning furnace wall area,mars regional characteristics in the production process via 'Artificial watch fire',and according to the experience of the prejudgment to predict the occurrence of semi-molten condition.The main problems of manual inspection are:1)the bad production environment(light,heat,dust and water fog),high risk,and workers for inspection at night,high labor intensity,not suitable for on-site inspection long time;2)the accuracy of discriminant is related to the operators'working experience,which means it is easy for missed inspections and false inspections.Therefore,enterprises need a kind of intelligent recognition technology of semi-molten condition,which can continue to work online,have a stable and accurate warning when the semi-molten condition has just occurred,reducing the labor intensity of workers,maintaining the safety of the workers,which is also good for improving the quality of the electric-fused magnesium products and increasing the economic benefit.In view of the existing problems in the visualization monitoring process of electric-fused magnesium furnace and the development of related advanced technologies,this paper puts forward a kind of electric-fused magnesium furnace perceptive technology of semi-molten working conditions based on visible light RGB image and infrared thermal image of deep learning.Using industrial camera and thermal infrared imager to obtain videos and images of the melting process of electric-fused magnesium furnace,adopt deep learning technology to extract feature and analysis magnesium furnace images in real-time type.Combing with the experience of the workers in the scene,this paper has set up a detection and recognition model of video and image information based on the semi-molten condition,which realized the on-line identification of the semi-molten condition of the electric-fused magnesium furnace.This project is not interfered by harsh environment such as strong light,high temperature and dust and so on,it also supports the recognition of the semi-molten condition of the electric-fused magnesium furnace with multi-channel real-time video and the historical video stream.Using deep learning method to solve the current industry's low detection efficiency and low accuracy problems of the area of furnace fire,semi-molten area of furnace wall,areas of outer mars.Thus providing a high accuracy of working condition recognition function in the industrial production of electric-fused magnesium,which has a positive significance in improving product' s quality,reducing the labor intensity of workers,ensuring the safe operation of the scene.Main works of this paper include:1.According to the characteristics of electric-fused magnesium's product environment,a lot of research have been made on the choices of sensors,the final choice is the Point Gray FL3-GE-50S5C-/M-C industrial camera and Optris 1450 infrared thermal imager which as the image acquisition equipment to set up the hardware platform,and the internal parameters were calibrated.According to the current monitoring and identification methods in the industry,combine the complementarity of the visible light information and the infrared thermal image information in different environments,the switch of the camera is carried out according to the different environment,so as to ensure the melting of video and image of the electric melting magnesium furnace with better acquisition effect.The experimental results show that this method can achieve good recognition effect in both dark environment(in the process of start and stop working condition of electric-fused magnesium furnace)or in the presence of strong light and dust.2.In order to solve the problem that the training sample is less in the semi-molten condition,the image samples generation technique is used to generate the training samples in the semi-molten condition.The application of Deep Convolutional Generative Adversarial Networks(DCGAN)method combined with TensorFlow deep learning frame has generated images of the semi-molten condition,so as to increases the image samples of semi-molten condition,this paper used this method to expand the training samples,and get complete and balanced training samples,which played a crucial role in the next step's detection,classification model.3.This technology combined the YOLO algorithm of the key regions detection with Darknet deep learning framework,and realized a real-time detection in the electric-fused magnesium furnace's melting industrial environment,which includes the flame area of furnace mouth,semi-molten area of furnace wall,outer mars area;and combined the AlexNet network model which is based on deep convolutional neural network with the Caffe deep learning framework to realize the classification of the working condition based on the single frame of the image,and calculated the accuracy of the working condition's classification results.In this process,the ordinary visible and infrared images are applied to the semi-molten condition of the electric-fused magnesium furnace in the automatic recognition process,and provided a real-time monitoring and identification process to display and switch infrared video and visible video,so as to analyze and evaluate the production process of electric-fused magnesium.4.The upper computer monitor terminal software is designed and developed,which is used to display the results of detection and recognition progress. |
PDF Full Text Request