Video Process Monitoring Of Fused Magnesia Furnace Based On Improved Non-negative Matrix Factorization Algorithm

With the rapid development of modern industry and science technology,especially the wide application of computer technology and industrial control equipment,the structure of processing industry system becomes more and more complex.Meanwhile,automation gets a high level which shows the large-scale,continuous and smart features.There will be huge casualties and property losses,as soon as a fault happens in the industrial field of the complex huge system.As a result,industrial safety problem becomes more and more popular.In order to detect the fault in furnace eruption process,this thesis focus on furnace eruption process of fused magnesium furnace as background,based on the FDA relative information entropy non-negative matrix factorization algorithm.Based on the analysis of the traditional non-negative matrix factorization algorithm,this thesis improves a new non-negative matrix factorization algorithm based on relative information entropy and FDA-constrained.The new algorithm makes the two algorithms complement each other and takes advantage of the excellent properties of non-negative matrix factorization in matrix decomposition dimensionality reduction,and makes up for the shortcomings of the FDA over-relying on the mean value in the classification process,resulting in unsatisfactory classification results.Through the simulation experiment of the production process of the fused magnesia furnace,the most original data of the video data in the production process of the fused magnesia furnace is further processed.The video data is divided into four angles and taken as:0 degree?45 degree?90 degree?135 degree.Converting each frame of video data into pictures and extracting their five sets of contrast,correlation,energy,entropy,and inverse disparity,a 20-variable data is formed as a training sample and a test sample in the simulation.Offline modeling of the system.Finally,the test sample is monitored and classified to get the effective result of the algorithm.Then,we improves the optimization method based on our algorithm:the relative information entropy nonnegative matrix factorization algorithm based on FDA maximum margin criterion.The core of the algorithm has been improved FDA,known as the maximum margin criterion.Taking the FDA-based maximum margin criterion as a new constraint on the new non-negative matrix factorization algorithm of relative information entropy,a new objective function is obtained.The same as fused magnesia furnace video data as the input data,the simulation results obtained show that the algorithm is equally valid.