Research On Particle Size Temperature Distribution Of Sinter Based On Image Recognition

The characteristic parameters of sinter at the interface between the sintering system and the cooling system are the key parameters for optimizing the sintering process and improving the efficiency of sintering waste heat recovery.Although many studies have been carried out on sintering process optimization and waste heat recovery,due to the lack of the particle size and temperature parameters of the sinter at the interface,the above studies are slightly insufficient in practical applications.Therefore,this paper takes the characteristic parameters of sinter as the research object,and uses the method of combining image recognition and deep learning to identify the sinter particle size and temperature parameters.The research content is as follows:(1)Establishing a high-temperature image data set of sinter.In order to systematically study the relationship between sinter temperature and image data,heat the sinter in the laboratory and collect high-temperature images.Through the analysis of the relationship between visible light imaging principle and temperature,a method for the brightness of visible light image to fit the temperature is proposed,and the sintering is established.Mine high temperature image data set.(2)Aiming at the problems of inaccurate temperature measurement such as inaccessible temperature measurement at the sintering site,and inaccurate temperature measurement due to the influence of high temperature flue gas in infrared temperature measurement.In this paper,combined with the BP neural network,the visible light image recognition method is used to identify the sinter temperature.The study found that building a BP neural network with a hidden layer of 3 and a single layer of neurons has the best temperature prediction effect,with a prediction error of only 2.02%.(3)Establishing a data set of sinter particle size images.The method of labeling the sinter particle size as the category is proposed.The image of the sinter particle size is collected in the laboratory,and the image is enhanced by preprocessing methods such as image rotation and translation.Use labelimg software to label the sinter,obtain the location and particle size category information of the sinter,and establish the sinter image data set.(4)In view of the slow speed of traditional image segmentation algorithms and poor generalization ability,the YOLOv3 target detection algorithm is used to image the sinter particle size,and the sinter with different particle sizes is divided into different types of targets for training.The algorithm is optimized by improving the learning strategy and loss function,which improves the accuracy of model particle size detection,and realizes the detection of sinter particle size distribution and location distribution.The trained model is applied to the detection of piled sinter,which realizes the measurement of sinter particle size distribution and position distribution.