Research On Fault Monitoring Method Of Grinding Classification Process Based On Transfer Learning

With the rapid development of computer technology and instrument technology,a large amount of process data has been reserved.The process monitoring and fault diagnosis methods based on data driven have been widely researched and applied,the traditional machine learning methods require training data and test data to meet the same distribution conditions.In the actual grinding-classification process,when the production granularity,ore properties or equipment parameters changes,it may lead to the data of the old production process and the new production process cannot meet the conditions of the identical distribution.Therefore,the original monitoring model cannot accurately monitor and diagnose faults in the new production process.Transfer learning can make use of the production experience and expert knowledge to guide the new process to establish the monitoring model,which can effectively improve the efficiency of learning and reduce the waste of knowledge.In this paper,combined with the fault monitoring of ore grinding-classification production process,the process monitoring method based on transfer learning is studied.The main contributions are listed as follows:(1)Firstly,the process monitoring methods are summarized,and the research status of data-driven process monitoring methods is analyzed.Through the description of the advantages and disadvantages of the statistical process monitoring method and the traditional machine learning method,and then propose the overall research strategy of this thesis.(2)In the large-scale modern mineral processing production,usually including a series of similar ore grinding-classification process,the series of ore grinding-classification process according to the requirements of production indicators or equipment maintenance to switch and combination.The process of running a new operation is called a new process,and it is difficult to establish the monitoring model for the new process contains a small amount of production experience and historical process data.This article introduce the transfer learning method,and establish a model based on feature transfer learning.According to the experience and expert knowledge of the old process,the definition mode of labeled data and state label matrix is described.The use of the collaborative matrix factorization to extract the common latent factors between process data and make a reliable bridge,density-sensitive distance measurements are used to calculate the similarity between samples.By fully considering the local and global consistency of the data,it effectively alleviates the occurrence of negative transfer.By solving the objective function to get the state label of the monitoring data,and then judge the running state of the process.It is proved that the method is practical and effective by using the production data of the grinding classification process.(3)In the actual grinding-classification production process,since affected by factors of the production equipment failure,sensor sensitivity failure,human error and other reasons,the production data contains noise,which will seriously affect the monitoring results.In view of this situation,this paper proposes a new anti-noise performance process monitoring method based on transfer learning.The method uses non-negative matrix factorization to remove the noise in the data,and obtain the accurate common latent factors.The state label of data samples in the new process is obtained by further optimizing calculation,through which the running state of the process can be judged.The validity and practicability of the method is verified by the simulation analysis with the actual process data.