Research On Fault Diagnosis Based On Semi-supervised Learning

With the development of computer technology and automation technology, the complexity and the scale of the process control system are steadily rising, if the system is running in improper operation, or due to some unexpected circumstances, the possibility of system failure is also growing. This will result in incalculable economic loss and physical security vulnerabilities. At the same time, with the development of automatic control theory, machine learning techniques and data mining technology, fault diagnosis technology has got a rapid development. How to use it to effectively improve the security and reliability of the system, which has been the research hot for the industrial and academic fields.From the viewpoint of the combination of semi-supervised learning and manifold learning, the main research of this paper is the fault diagnosis process based on semi-supervised locally linear embedding algorithm, and the work and contribution of this paper is as follows:(1) describes the status quo of the fault diagnosis, and draws a detailed analysis from the both directions of model-based and data-driven. This paper mainly focus on fault-based diagnostic techniques based on data-driven, and presents a new classification method according to the characteristics of the data itself, which includes supervised learning methods, unsupervised learning methods and semi-supervised learning methods.(2) studies the semi-supervised learning and manifold learning theory, which focuses on the semi-supervised classification algorithms and a variety of non-linear dimensionality reduction in algorithms manifold learning, and draws a comparison on Swiss roll data sets with the linear dimension reduction algorithms. (3) proposes a fault diagnosis method based on semi-supervised locally linear embedding algorithm. We get low-dimensional feature space using semi-supervised locally linear embedding algorithm to the data collected for feature extraction. And we achieve the purpose of fault diagnosis in the low-dimensional feature space for fault pattern classification. We use this fault diagnosis method on fault simulation data sets, and simulation results show that the effect of fault classification algorithm is obvious, which have a certain practical value.(4) The proposed method is used in TE model and BSM1 sewage treatment model. Using the model to generate fault data sets, we verify the proposed fault diagnosis method. Experimental results show that the method can effectively identify the faults generated by the models and to improve fault diagnosis performance.