Fault Detection Method For Fused Magnesium Industrial Process Based On Semi-supervised Learning And Collaborative Modeling

With the rapid development of the modern economy and the continuous expansion of industrial models,the complexity of industrial processes continues to increase.The need for product quality is getting higher and higher.Production safety issues have received increasing attention.The detection of complex industrial processes has become the focus of modern industrial development.Data-driven process detection technology has been widely applied to the industrial field.Based on the previous work,this paper develops a semi-supervised kernel linear discriminant analysis algorithm based on multi-manifold,a manifold dimensionality reduction algorithm based on collaborative modeling and a semi-supervised linear regression algorithm based on collaborative modeling.The main research work is as follows:Considering the difficulty in obtaining the sampled samples in the supervised method,in order to exploit the potential information of unmarked data,this paper proposes a semi-supervised kernel linear discriminant analysis algorithm based on multi-manifold.In the absence of labeled sample,the method can use the unlabled sample to guide the learning process.Furthermore,compared to traditional manifold-based methods,this method add the label information of the sample into the construction of the manifold.Not only can the local structural information of the data be kept unchanged,but also the overall information of the sample can be fully considered,and a more detailed manifold structure can be constructed.Due to the particularity of the fused magnesia process,the traditional physical variables that can be detected in this process are very limited.If only using current variables for modeling,the information extraction will be insufficient.However,if only using the image data for modeling,the problem of manually selecting features will be encountered in feature extraction.Therefore,it is necessary to combine the image data and current data into a model.This paper proposes a manifold dimensionality reduction algorithm based on collaborative modeling,which can effectively combine image data and current data for dimensionality reduction.The fourth chapter is an unsupervised method that cannot classify the data directly.In order to solve this problem,Chapter 5 adds regression functions into the objective function of Chapter 4.Compared with the previous chapter algorithm,the algorithm in this chapter can effectively use the label information in the sample to classify the sample directly.Finally,the proposed algorithm is applied to the simulation of fused magnesium process,and their fault detection performance is tested.The feasibility and effectiveness of the methods are verified by comparison with some traditional methods.