Research On Fault Diagnosis Bayesian Network Model For Cement Rotary Kiln

Calcination is the most important aspects of the whole cement production line.Calcining process is completed in the cement rotary kiln. The working status ofcement rotary kiln cement directly affects the quality and output of cement products.The system of cement rotary kiln is complex, and the defaults of the cement rotary kilnare diverse and vague. With the restriction of equipments, artificial methods have beenunable to meet the requirements of accurate fault diagnostic. Based on this, accordingto the working principle of the cement kiln system, the model of fault diagnosis forcement rotary kiln is designed based on Bayesian Network in this paper. Then, thecomposition of the kiln fault diagnosis system is described. The fault diagnosis systemis consist of data module, learning module, and inference module. After learning andreasoning for the model, the precise fault diagnosis for the cement kiln system can beachieved.First of all, study the working principle of cement rotary kiln and the mechanismsof reactions, analysis the property of the variables and the interaction between them.After enough collected, these datasets must be screened and quantization processedthe data according to the technical requirements. Then these datasets used forBayesian Network learning and diagnostic can be obtained.Secondly, in order to solve the difficult problem of modeling for the cementrotary kiln, two improved methods named Dataset Speed Correct algorithm andSimplified Greedy algorithm for Bayesian network structure learning named areproposed. The DSC is a new algorithm that totally dependents on datasets. Simulationresults show that the execution speed of Dataset Speed Correct algorithm is higher,and the Dataset Speed Correct algorithm is more accurate. Therefore, the DatasetSpeed Correct algorithm is selected to learn the structure of Bayesian Network for thefault diagnosis of the cement rotary kiln. The process of fault diagnostics modeling isintroduced according to the Dataset Speed Correct algorithm in detail.At last, the detail description of process of parameter learning and diagnostic reasoning buy using of maximum likelihood estimation algorithm and variableelimination method. According to the curve of the accuracy obtained by dataexperiments, it can be seen that the accuracy of the diagnostic model is high.