A Study On Burning State Recognition Of Clinker Based On Semi-supervised Independent Component Analysis And Hidden Markov Model

Rotary kiln is widely used in metallurgy,cement and chemical fields,for the physical or chemical treatment of materials.In the process of kiln clinker sintering,the kiln rotates continuously,the process of complex combustion,convection of flue gas and materials is carried out,and the key process parameters related to clinker quality index are difficult to be measured online,leading to the current rotary kiln Clinker sintering process is still in the"artificial observation" stage.The manual mode of operation subject to subjective factors,there will be low product quality,poor equipment operation and high energy consumption and other issues.The burning video images of the clinker burning zone contain a lot of temperature field information and clinker sintering information,which provides the possibility to study the burning state of the rotary kiln clinker based on the burning image.However,due to the rotation of the kiln and the interference of the dust and smoke,the flame image of the firing zone is not high and the regions are strongly coupled,and it is difficult to obtain the distinguishing features and robustness based on the static image method.Combining the latest research results of image processing and machine learning,it is of great significance to studies the identification method of clinker sintering condition of rotary kiln based on clinker firing flame video image.In order to solve the problems of rotary kiln clinker sintering condition identification,this article relies on the National Natural Science Foundation of China project "Based on image and process data fusion of rotary kiln product quality parameter prediction modeling",with alumina rotary kiln as the research object,carry out a study on the identification of sintering status of clinker based on flame video image processing and machine learning.The main work is as follows:1.The existing image-based rotary kiln clinker sintering condition identification work is more supervised or unsupervised image feature extraction method.The supervised image feature extraction method requires the input data to be labeled,while marking a large amount of data is very time-consuming and labor-intensive.Unsupervised algorithm is mainly used to study the low-dimensional structure of high-dimensional data,mining low-dimensional features,while lower ability to learn the characteristics of class distinction.Aiming at these problems,this thesis proposes a method of flame video images feature extraction based on semi-supervised independent component analysis(SS-RICA).Independent component analysis is used to extract the features of statistical independence,marginal discriminant analysis method of framework is used to enhance the separability of learned features.The model is compared with the PCA,MFA and RICA feature extraction algorithms on 10%,?30%and 70%percentage labeled sample sets respectively.The model is evaluated by linear support vector machine classifier.The experimental results show that the SS-RICA feature extraction method have achieved the highest recognition precision using training and the test sample set.2.Rotary kiln firing zone flame video contains more clinker sintering information,In this thesis,a continuous hidden Markov model(cHMM)is proposed for the identification of sinter:ing conditions of rotary kiln clinker based on the burning video image.Based on the flame video image data of three sintering conditions,the Gaussian Mixture Model(GMM)is used to estimate the observed sequence,and Baum-Welch algorithm to learn the HMM model structure,?1??2,?3under three sintering conditions,finally,according to Bayesian decision-making theory to establish the final pattern classifier.The HMM classifier and non-linear SVM classifier are tested on the 10%,30%,and 70%percentage labeled sample sets,and the HMM classifier has the highest recognition precision in the training and test sample set.3.In order to verify the algorithm proposed in this thesis and explore the transformation of laboratory research results to industrial results,this thesis designed and developed a rotary kiln clinker sintering conditions identification experimental system.The experiment system adopts the browser/server mode,the front-end browser interface is HTML,PHP and Javascript as the development language,and the back-end algorithm is realized by MATLAB executable file.The function of system introduction,model offline training and model online application are designed respectively.Among them,the model off-line training section design training data selection and model parameter input function,to achieve the data pretreatment,feature extraction model training,classifier training and the results show that the four main function modules.The on-line application section utilizes off-line training models to identify the sintering conditions in real time for the input data.The development of the experimental system provides a way to transfer the research results in the laboratory environment to the industrial products.