Fault Detection Method Based On Modified Kernel Locality Preserving Projections

With the continuous promotion of automation level of industrial production equipment,modern industrial processes tend to be more integrated,complex and intelligent.How to eliminate the fault in production process accurately and timely is crucial to ensure production safety.In recent years,with the rapid development of sensor technology and real-time data storage technology,a large number of process data in industrial production can be saved.Therefore,how to fully extract the data features in a large number of historical industrial process data becomes a key step in fault detection.Considering the characteristics of high-dimensional,nonlinear and Gaussian data in real industrial process,the traditional single feature extraction method can no longer meet the requirements of extracting rich data features.To improve the traditional feature extraction method,the main contents include the following aspects:(1)Since the kernel locality preserving projection is added to the kernel function on the basis of locality preserving projection,the method can better deal with nonlinear data while retaining the characteristic of locality preserving projection.In addition,a nonlinear approximate projection matrix can be obtained by kernel locality preserving projection,which the low-dimensional representation of test sample can be obtained quickly,and the out-of-sample problem is solved well.However,the kernel locality preserving projection can only extract the local neighbor structure of data and cannot obtain the comprehensive feature information of original data.Therefore,it is considered to combine kernel principal component analysis with kernel locality preserving projection.At the same time,considering the sample data may affect the rate of fault detection,the characteristic.vectors obtained from kernel principal component analysis are scaled down,which make the data inside the volatility changes more smoothly,and then combining the kernel principal component analysis to the kernel locality preserving projection,and the modified kernel locality preserving projection method is proposed.Taking TE process as the research background,the proposed method is verified by making use of simulation experiments for TE process data.And the simulation results verified the effectiveness of the proposed method.(2)Because the traditional T2 statistics and SPE statistics are measured in different methods,they cannot be combined into a single statistic for fault detection,which will lead to different fault detection rates.In order to avoid the shortcomings of T2 statistics and SPE statistics,the support vector data description method is used to establish the fault detection model,and the distance statistics and radius control limit are used as the basis for fault detection.Combined with modified kernel locality preserving projection method,a fault detection method based on modified kernel locality preserving projection support vectors data description is proposed.The validity of the proposed method is also verified by the data of TE process.