Research On Diagnosis Method Of Incipient Fault In Industrial Process

Dying to their small amplitude and easy to be masked by noise,the incipient faults in the industrial process have always been the hotspots and difficulties for scholars.Especially in recent years,modern industrial processes have the characteristics of large scale,large number of variables,and tend to be streamlined.Moreover,the industrial production environment is harsh,often accompanied by various noises and disturbances,further increasing the difficulty of incipient fault diagnosis.The data-driven fault diagnosis method has been welcomed by more and more experts and scholars because it does not rely on accurate mathematical models.Based on background of the Tennessee Eastman process,this thesis introduces mutual information theory on the basis of traditional methods,improves the principal component analysis method,and improves the detection rate of incipient faults.And a two-stage support vector machine fault diagnosis method based on mixed features is proposed to diagnose the fault type of the faulted samples.It is mainly divided into the following two parts:(1)This thesis deeply studies the reasons why the traditional principal component analysis method has low detection rate of incipient faults.Based on principal component analysis model,the columns of loading matrix in principal component analysis are reordered by mutual information between different statistic component matrices and training data.Then,the new principal component subspace is selected according to the largest mutual information.The selected principal component subspace can maximally reflect fault characteristics into a new statistical index.Besides,a detection index based on the sliding average control chart statistic is proposed,which eliminates the effect of noise by proper averaging of the most recent samples,greatly improving the ability of fault detection.(2)Aiming at the problem that traditional classification methods have low ability for incipient fault diagnosis,this thesis proposes a two-stage support vector machine fault diagnosis method based on mixed features.The main ideas of this method are as follows:using the statistic component feature mentioned above as the input of the first stage support vector machine,the fault can be locked into a certain subclass,and the selection range of the type can be reduced.Then,the local rotating space statistical feature and the global slow feature of the fault sample are combined to form a mixed feature,which can fully reflect the local and global effective fault information of the fault.The reliefF feature fusion method is used to reduce the dimension of the mixed feature,and the feature vector after the dimension reduction is used as the input of the second stage subclass support vector machine model to determine the specific type of fault.A large number of simulation experiments were carried out on the incipient fault diagnosis method proposed above.The experimental results show that the proposed principal component analysis method based on mutual information and the two-stage support vector machine fault diagnosis method based on mixed features can greatly improve the diagnosis ability of incipient faults.