Research On Fault Detection Methods Based On Fuzzy Clustering And Heterogeneous Data

In the actual production process of fused MgO,due to the complexity of the structure and composition of the equipment and the large and complex of scale of production,data generated during the production process is explosive growth.And monitoring of the production is extremely difficult.Due to the inaccurate regulation of the material electrode,it eventually leads to problems such as low product yield,poor quality,and even major accident.For the excessive number of data collected during the production process of fused magnesium oxide,the feature dimension is too high and the data has a nonlinear relationship and the large correlation.This paper proposes a fault diagnosis method based on clustering hypothesis to improve the monitoring of the production process.And this paper proposes a fault diagnosis method for multi-source industrial heterogeneous data based on manifold learning;the process automation for fused MgO is extremely low,and the process is prone to fault and difficult to monitor and improves some problems existing in this method.In view of the above problems,this paper mainly made the following research:(1)For the production process of fused MgO,the chemical information is difficult to collect and monitoring this process is difficult.The fault is easy to occur.This paper proposes an algorithm for industrial data fault detection based on semi-supervised kernel principal component analysis.This algorithm uses the kernel principal component analysis(KPCA)that relies on the principal component for fault detection to eliminate redundant features.And then combines the clustering hypothesis to modify the semi-supervised KPCA that ignores the local information between data.The experiments show that the fault identification of industrial data based on semi-supervised kernel principal component analysis is not only accurate and fast,but also can better guide actual production.(2)There are many sources of raw data collected during the process of fused MgO.The feature dimensions are too high and there are redundant features.In this paper,a multi-source industrial heterogeneous data fault detection algorithm based on manifold learning is proposed.The algorithm takes into account the variables of the production process physical and the image video.Thereby,the heterogeneity of multiple information sources is realized and the relationship between the data sources is found to achieve the purpose of joint modeling so that the subsequent fault detection and diagnosis process only needs to construct a model on a part of the features.And the problem of dimension disaster will be greatly alleviated.On the other hand,in these industrial fault detection and diagnosis algorithm,which generally use high-dimensional features,utilizing the features obtained from the selection of the original features helps to reduce the computational cost,save storage space,and reduce the chance of over-fitting.(3)The data collected in the industrial production process of fused MgO is less than the label and the manual marking of the data requires more manpower and material resources.This paper proposes a semi-supervised multi-source based on manifold learning.The detection algorithm for industrial heterogeneous data improves the problem that the original algorithm needs to mark each data of the data set.It only needs to mark a small amount of data and uses the loss term and the offset to solve the inaccuracy of the feature extraction caused by the reduction of the label information.The data after feature extraction restores the original data features as much as possible.