Research On Fault Diagnosis Method Based On Randomized Fisher Discriminant Analysis

Modern industrial production processes and equipment structures are becoming increasingly complex,larger in scale,and more automated.Once a fault occurs,it will seriously interfere with the normal operation of the production process,which may not only lead to a decline in product quality and stagnation of the production process,but also lead to unpredictable catastrophic accidents.Therefore,fault diagnosis technology is of great significance for ensuring the safety of the production process and equipment.In traditional fault diagnosis methods,data driven methods do not rely on process knowledge.Utilizing collected operational data related to operational status,exploring the information within the data through machine learning and statistical analysis,establishing fault classification and prediction models,and achieving timely diagnosis of fault phenomena are currently more practical fault diagnosis techniques.Due to the problems of high dimensionality,non Gaussian,and non-linearity in most industrial data,resulting in failure classification results that do not meet the expected requirements,and using kernel techniques to process nonlinear data will face the problem of increasing the computational complexity of the kernel matrix with the increase of training samples.Therefore,based on Fisher discriminant analysis,this paper introduces a random Fourier feature mapping method,A fault diagnosis model based on random Fisher discriminant analysis is proposed for the fault diagnosis of rolling bearings,and a fault diagnosis model based on random local Fisher discriminant analysis is proposed for the fault diagnosis of blast furnace iron-making processes.The effectiveness of the two fault diagnosis designs is verified through actual data sets.The main contents of this study are as follows:(1)A rolling bearing fault diagnosis method based on random Fisher discriminant analysis is proposed to solve nonlinear problems in bearing data.Firstly,12 representative time-domain features are extracted from the collected bearing vibration signals,and the extracted time-domain feature data is mapped to a high-dimensional space using random feature mapping method.Then,Fisher discriminant analysis is used to process linear problems to extract discriminant features from the data.Finally,Bayesian inference is used to identify the category of vibration signals collected for bearing fault diagnosis.In order to verify the effectiveness of the model for fault diagnosis,two widely used bearing datasets,CWRU and PU,were used to conduct model training experiments on the diagnostic model proposed in this study.The experimental results show that the fault diagnosis model proposed in this study can effectively classify the types of bearing faults,and has better computational speed and classification accuracy compared to kernel Fisher discriminant analysis.(2)A method of blast furnace fault diagnosis based on locally retained random Fisher discriminant analysis is proposed to solve the nonlinear problem of data in blast furnace iron-making process.Firstly,select the data variables that reflect the condition of the blast furnace,and use a nonlinear mapping function to map the raw space blast furnace data to a random Fourier feature space.Compared to the high-dimensional space induced by kernels,this space dimension is smaller,greatly reducing computational costs.In addition,referring to the strategy of local preserving projection,preserving the local structure of the data,reconstructing the divergence matrix,and finally using Bayesian inference strategy for fault classification.This method utilizes local and global information to better represent the manifold structure in the data and better solve the nonlinear problem of the data.The experimental results show that the proposed method can not only effectively classify blast furnace fault data,but also greatly reduce the calculation and storage costs.