Research On Prediction Of Converter Endpoint Based On Image Processing

At present,converter steelmaking is still the most important and most efficient smelting method among many steelmaking methods in my country.The entire process is a complex process with high temperature sealing and lack of internal information.The problem that the converter has always been difficult to solve is the accurate control of the end point prediction.Among them,oxygen blowing is an important part of converter steelmaking.Oxygen blowing can reduce the carbon content,increase the furnace temperature to remove impurity elements,and accurately control the amount of oxygen can improve the end point hit rate to a certain extent.With the continuous development of image processing technology,it can be proposed to extract effective flame information to predict the end point by the flame of converter steelmaking.In the process of steelmaking,as the blowing progresses,various elements and temperatures in the molten steel must be reflected in the flame,and the blowing is established by extracting the flame color characteristics,boundary complexity characteristics,and texture characteristics.Model,which is the key to predicting the end point based on image processing.The flame changes very rapidly during the combustion process,so there are certain difficulties in extracting features.In order to accurately predict the blowing end point,a series of studies are carried out in this paper.The specific research contents are as follows:(1)In the process of making steel,oxygen blowing is an important link.A reasonable amount of oxygen blowing can effectively improve the hit rate of the end point.This paper introduces deep learning,uses deep belief network,establishes the oxygen blowing model,and improves the deep belief network,on the one hand,it effectively improves the end point hit rate,on the other hand,it can improve the production efficiency of the steel plant;(2)In view of the current production status of converter steelmaking and blowing status recognition in a domestic steel mill that is still completed by human experience,which leads to a low recognition rate of blowing status,a deep convolutional neural network algorithm based on ResNet is proposed,which can effectively identify converter blowing Compared with the existing algorithm,the recognition rate has been greatly improved,which lays the foundation for subsequent endpoint prediction;(3)The flame images collected on site are disturbed to some extent,so effective preprocessing must be performed first,including image de-drying and segmentation.In this paper,the domain average method and median filtering are compared,and the average filtering with better effect is used.At the same time,the threshold segmentation and FCM segmentation method are compared,and the FCM segmentation method with better effect is selected.For the flame color feature extraction of the image,we calculate the third-order moment as the chromaticity feature;for the boundary feature extraction,we calculate the line invariant moment;for the texture feature,we use the gray difference statistical method to calculate the contrast and angle direction.Four characteristic values of order moment,average value and entropy are used as texture features;(4)For the flame chromaticity features,boundary features and texture complexity features that have been extracted,the three image features are used as inputs,and the least square support vector machine network model is used to establish the relationship between the steelmaking flame image features and the blowing data For the optimization of parameters,particle swarm optimization is used to establish the best prediction model.In the process of converter blowing,the flame at the furnace mouth changes rapidly and complicatedly as the blowing proceeds.Combining with the basic automation conditions on site,this article aims to extract various characteristics of the flame by collecting images of the flame at the furnace and combining the blowing at the site.Refining data,establishing the relationship model between the flame image features and blowing data,can effectively predict the blowing data,so as to achieve the end point control of the converter.