Research On Working Condition Identification In The Production Process Of Rotary Kiln

The accurate identification of working condition in the production process of the rotary kiln,especially the sintering belt,has an impact on the final quality of the product.The accurate identification of sintering belt is mainly based on the flame image of sintering belt,which contains rich information of sintering condition of clinker and temperature field of sintering belt.However,due to the influence of coal dust in the kiln,there is a strong coupling between the areas of interest in the firing zone flame image and the boundary is blurred.At the same time,some complex noise interference will have a certain impact on the process parameters that need to be controlled during the whole roasting process.As a result,many production enterprises still observe the sintering process of rotary kilns through the “artificial fire watching” method.Based on this background,in this paper we studies the recognition of the combustion state of the firing zone image and the prediction of product quality through key process parameter information.Firstly,the convolution neural network(CNN)in machine learning theory is used to classify and identify the combustion state of the firing band flame image.The principle of each layer of CNN was briefly described,and the traditional LeNet-5 neural network was changed.The image of firing zone was identified by the modified network and the experimental results were analyzed by using the Tensorboard visualization platform in machine learning.Secondly,the knowledge and methods of multi-information fusion technology are briefly described.Based on this method,during the synchronous cycle,a data was set by the result of CNN’s recognition of the firing flame image combine with the data of the temperature of the kiln head,firing zone,kiln tail and the negative pressure of kiln tail that we collect,according to application the random forest algorithm,a kind of method of classification and recognition for rotary kiln products was proposed based on RF.The product quality classification and recognition model of rotary kiln with multiple information fusion was built by using Python language to write programs for experiments.Finally,analyzing the experimental results through confounding thevisualization of the matrix.The ROC curve was used as an index to compare with the SVM classification algorithm.The experimental results show that the accuracy of this algorithm is 81%,and the performance is superior to SVM.This realizes the network information control system of multi-information fusion by human and machine.