Research On Incipient Fault Detection Of Industrial Process Based On Fractional Fourier Transform

With the development of science and technology and the growth of social demand,process industry has become an important basis for the development of national economy.However,due to the increasing complexity of modern industrial production systems,large-scale industrial systems are more prone to major safety accidents.To ensure the industrial production process can run stably for a long time,it is important to detect incipient faults in time.These incipient faults usually have characteristics of slow change speed,not obvious mutation characteristics and being easy to be submerged by noise,and are not easy to be detected by conventional fault detection methods.At present,research on detection of incipient faults is just beginning.Therefore,research of incipient faults detection has important theoretical significance and application value.The work of this thesis is as follows:(1)In view of the characteristics of incipient faults,which are not obvious in amplitude characteristics and easy to be submerged by interference and noise,an incipient-fault detection method with fractional Fourier transform based on data eigenvector scaling is proposed.Firstly,feature vector of incipient fault data is scaled to enlarge the variance of fault data and highlight main features of incipient fault.Secondly,fractional Fourier transform is carried out on the incipient-fault data scaled by eigenvector,and incipient fault is transformed from time-domain where the feature is not obvious to fractional domain where amplitude is the most obvious.Finally,in fractional domain,principal component analysis method is used to reduce dimension of incipient-fault data,extract main feature information,and complete detection.The validity of proposed method is verified by simulation experiments with data sets generated by Tennessee chemical process and penicillin fermentation process.(2)Aiming at the non-linear problem existing in industrial process,a kernel principal component analysis based on moving window for incipient-fault detection method is proposed.To begin with,by using moving window technology,errors generated by sampling data in window are accumulated to enlarge features of incipient-fault data as far as possible,so that it can be detected easily.Then,fractional Fourier transform is applied to incipient fault data processed by moving window to make amplitude characteristics of incipient fault more obvious and filter part of high-frequency interference signals at the same time.Finally in the light of non-linear characteristics of chemical data,kernel principal component analysis is utilized to map linearly inseparable data to linearly separable high dimensional space for completing detection in high dimensional space.The proposed method is applied to fault detection of Tennessee chemical and penicillin fermentation process,and it is verified that the method is effective in detecting incipient faults.Finally,incipient detection method proposed in this thesis is summarized,and problems existing in relevant research and future research direction are pointed out.