Fault Diagnosis Method Based On Data Driven For Industrial Process

It’s inevitable to occur failure in the use of modern industrial production equipment. If the failure can not be timely and accurately detected and ruled out, it may lead to paralysis of entire system and lots of loss and damage. Because a large number of tracking and machine data are generated and collected by the production equipments, the method based on data driven has been widely used in the fault diagnosis of industrial process. This method doesn’t rely on accurate mathematical models of systems, the control algorithms based on a large amount of data can be used to make it high efficient and convenient.Considering the complexity and linearity of data, fault diagnosis technology based on data driven will be constantly improved in the basis of common methods. The major research work of this paper is shown as follows:1) In order to overcome low accuracy of traditional nonlinear fault diagnosis method, Diffusion Maps and K Nearest Neighbor (DM-KNN) based fault detection method is researched. The manifold structure of data is fully considered and the global characteristics of nonlinear data are retained; in term of the advantage that KNN rule can overcome residual information missing, KNN fault detection theory is applied into low dimensional manifold feature space to find potential faults. The diagnosis result of this method is more reliable and accurate. Under the research background of TE process, simulation data is used to verify typical fault. The results show that the method has high precision.2) To realize automation and high accuracy of industrial process fault diagnosis, a data driven method based on Diffusion Maps (DM) and Hidden Markov Model (HMM) is proposed. The intelligent fault diagnosis thought with feature extraction and classification decision is adopted, data storages are reduced by DM and the low-dimensional eigenvectors are regarded as the observation sequence of HMM for fault pattern recognition to improve the quickness and accuracy of diagnosis. Finally, the effective of this novel method is proved by TE process simulator.