Fault Detection Method For Fused Magnesium Industrial Process Based On Modified Cluster And Heterogeneous Data

With the development of modern industry and science and technology,the scale of modern industrial processes continues to expand,and the system structure is increasingly complex.At the same time,the continuous development of detection technology and computer technology and its wide application in industrial processes make the data information of the process larger and larger.How to co-process these multi-source heterogeneous data is an urgent problem in the era of big data industry.Moreover,for the metallurgical,chemical and other process industries,the operation process is complicated,the working conditions are changeable,and sometimes it is difficult to determine the occurrence and development of the fault.The original data-based method is difficult to meet the early monitoring of the failure in the industrial process.In view of the above problems,this paper mainly made the following research:(1)A semi-supervised classification fault detection method based on modified clustering is proposed.Mainly for the defects of traditional clustering,through the modified clustering hypothesis,that is,similar data sharing the same category membership feature information,At the same time,consider the neighborhood information of the data,and combine the marked data with the unlabeled data to obtain a fault detection method that is more in line with the actual industrial process.The process monitoring results using this method show that the proposed method has better performance in fault detection than the traditional clustering semi-supervised method.(2)A semi-supervised KPCA with modified clustering is proposed to detect intelligent faults in early faults.It is mainly for the problem that the early industrial faults are difficult to find and judge.At the same time,the data distribution information and the data category label information are considered,and the modified clustering hypothesis is adopted,that is,the hypothesis data has two kinds of category characteristic information at the same time.The semi-supervised KPCA method using modified clustering is improved and is more suitable for the diagnosis of early industrial process faults.The process monitoring results using this method show the accuracy and effectiveness of the method for fault detection of early faults.(3)An abstract data collection window concept is proposed to select data,and then multi-task partitioning in the window to fully process the multi-source data.This makes it possible to unify the physics and chemistry variables of the production process and the big data pool of image sound and video.A multi-source heterogeneous data collaborative multi-task processing fault detection method is proposed,which is a statistical processing framework for multi-source heterogeneous data of large-scale industrial data.Then,through the collection of traditional data and multimedia stream heterogeneous data for unified modeling,multi-source heterogeneous data collaborative multi-task logistic regression processing model is proposed,and then the improved Nesterov’s method is used to project the gradient in Euclidean space.The drop solves the multitasking model.Finally,the video data of the industrial field and the traditional current,voltage and frequency data are collected,and the method is validated to prove the effectiveness of the proposed method.(4)Summarize the work of this paper,and for the early industrial fault data diagnosis method and multi-source heterogeneous data collaborative modeling method for the related issues.