Fault Detection Based On Local Tangential Space Alignment Under Data Missing Condition

The With the rise of intelligent manufacturing and Industry 4.0 technology,equipment in various fields is becoming more and more complicated,precise and comprehensive,operating conditions and working environments are becoming more and more complex and changeable,and the probability of equipment failure is gradually increasing.For automatic system,monitoring its status and timely and accurate diagnosis and detection is of great significance.In the actual system,due to interference factors such as network transmission,data samples collected will be missing,which makes fault detection difficult.In this paper,the Local Tangent Space Alignment(LTSA)method is used to construct the fault detection model in the data situation,and combined with the missing data processing scheme to carry out the fault detection of the data missing system.Traditional data-driven fault diagnosis methods have many limitations,such as data must satisfy integrity,linearity and gaussian distribution.To solve these problems,this paper proposes several effective fault detection schemes based on the local tangential space arrangement method,which mainly include:(1)Data missing preprocessing method based on Extreme Learning Machine(ELM).Because the local tangential space permutation model is sensitive to the sample with missing data,a limit learning machine algorithm is proposed to fill the missing data.First look for sampling instant,with the missing value to contain the missing value of input variables as model,missing variables as model output,according to the properties of related variables between training extreme learning machine model,and will,in turn,update training model containing the missing value fill complete the sampling times,due to the incomplete cannot solve online sample as local tangent space alignment method input problem.(2)LTSA modeling and fault detection considering category information under data missing condition.In view of the nonlinearity and non-Gaussian characteristics of actual industrial process Data,a detection model of Discriminant Improved Local Tangent Space Alignment-Support Vector Data Description(DILTSA-SVDD)was proposed.It mainly solves the problem that traditional LTSA algorithm is difficult to deal with new data samples and can well retain the fault category information of the system.(3)LTSA modeling and fault detection considering global information under the condition of missing data.Due to some fault information embodied in the global information,only consider the LTSA category information while method can better for nonlinear,non-gaussian data feature extraction,but in the process of dimension reduction could damage the external shape of the original data set,global information based on the consideration of LTSA method,combining improved support vector is described in this paper,a new fault detection scheme,In other words,Global-Local Structure Tangent Space Alignment-Compound Support Vector Data Description(GLTSA-CSVDD)solves this problem and carries out fault detection simulation in TE process.The result shows that GLTSA-CSVDD method has higher detection accuracy.Finally,a fault detection and monitoring platform based on MATLAB-GUI is developed for the proposed detection scheme.