Study Of Fault Detection Based On Global And Local Structure Feature Extraction

With the gradual increase in the degree of industrial process automation, real-time detection of the fault and the safe operation of the process is increasingly important. In modern chemical process, numerous observations can be well collected and stored, it has become one of the research hotspots in industrial process monitoring which only depends on the process data for the feature extraction, process modeling and monitoring. At present, the main method of the fault detection is global and local structure feature extraction method, such as principal component analysis(PCA) and neighborhood preserving embedding(NPE). In order to solve some problems of the traditional fault detection algorithm, some new methods are proposed as follows:(1) A fault monitoring method based on distributed independent component analysisprincipal component analysis(ICA-PCA) model is proposed, which is suitable for complex industrial process that can’t be divided into several sub-blocks through an automatic way and has non-Gaussian information. Firstly, an initial PCA decomposition is carried out upon the whole process variables. By constructing sub-blocks through different directions of PCA principal components, the original feature space can be automatically divided into several sub- feature spaces. In addition, a two step ICA-PCA information extraction is carried on upon all sub-blocks in order to extract both Gaussian and non-Gaussian information, and the new statistics and their statistic limits were built. Finally, simulation of TE process shows that the proposed fault detection model is efficient.(2) To handle the problem of industrial data is not completely in line with the normal distribution and the nonlocal properties exist, a new fault detection algorithm based on sparsity preserving projections(SPP) is proposed. Unlike many existing techniques such as local preserving projection(LPP) and neighborhood preserving embedding(NPE), where local neighborhood information is preserved during the dimensionality reduction procedure, SPP aims to preserve the sparse reconstructive relationship of the data, which is achieved by minimizing a 1l regularization-related objective function, and the projection vector is calculated. Next, by means of the projection vectors, the original variable space is divided into the feature space and residual space, and hotelling’s 2T and squared predication error are constructed to monitor the variations among the two spaces. Finally, a case study demonstrates that the proposed method has good detection effect.(3) Since conventional methods may preserving the local or global data structure during feature extraction, a fault detection method called local and global preserving embedding(LGPE) is put forward to tackle the problem. The LGPE algorithm uses the objective function of neighborhood preserving emedding(NPE) to embed the local data structure, a nd designs a new target function to constrain the relative position betwee n the sample and its nonlocal area. Then, the local and global structural features of the original data are preserved by the dual optimization problem, and the statistics and statist ics are constructed. Finally, the feasibility and practicability of the proposed algorithm are verified by simulation experiments.(4) In view of the limitation of the fault detection algorithm for shallow structures, which are not good at the representation of complex functions and the generalization of complex problems, in this chapter, a fault detection algorithm based on denoising Auto-Encoder which uses the advantages of deep learning algorithm is proposed. Firstly, 1l paradigm is introduced to avoid overfitting, and the noise is added to improve the robustness, then fault detection model based on denoising Auto-Encoder is constructed. Finally, the simulation experiments in the TE process show that the proposed fault detection a lgorithm is robust and effective to extract the essential features of the original data.