Fault Diagnosis Of A Class Of Nonlinear Dynamic Systems Based On Depth Feature Engineering

With the rapid development of modern engineering systems,more and more systems are developing in the direction of large-scale,intelligent and dynamic development.The characteristics of engineering systems such as nonlinear and dynamic complexity increase the probability of system accidents,and the loss of financial and material resources caused by system failures is immeasurable.Therefore,in order to ensure the safe and stable operation of nonlinear dynamic systems,It is of far-reaching significance to strengthen the research of fault diagnosis technology for nonlinear dynamic systems.In this thesis,two nonlinear dynamic systems,the CAFe superconducting linear accelerator control system and the Tennessee Eastman process,are studied.Based on intelligent research methods such as algorithm optimization and neural networks,the in-depth feature information of system fault data is extracted,and ultimately the fault diagnosis of nonlinear dynamic systems is realized.The main innovation content of this thesis is arranged as follows:(1)Aiming at the shortcomings of manta ray foraging optimization(MRFO),which is prone to fall into local optimization and slow convergence in the later stage,an intelligent improved manta ray foraging optimization(IMRFO)was proposed.Using two hybrid strategies to optimize and improve MRFO.Firstly,the initial population is updated by using a sine chaotic mapping strategy to ensure uniform distribution of the initial manta ray population and avoid falling into local optima;Then,the optimal population position is updated through Cauchy mutation strategy to accelerate the later convergence speed of the algorithm and improve the optimization accuracy;Finally,10 test functions are used to evaluate the performance of IMRFO,and 5 classic optimization algorithms are used to verify the effectiveness and superiority of IMRFO.(2)Aiming at the problems of CAFe superconducting linear accelerator system mainly using artificial experience for fault diagnosis,such as simple model,difficulty in extracting deep feature information from data,and low diagnostic accuracy,a fault diagnosis method for CAFe superconducting linear accelerator based on IMRFO-SCN is proposed.A stochastic configuration network(SCN)is selected for fault diagnosis of the CAFe superconducting linear accelerator system.Because the randomness of the regularization parameter of SCN and the proportional factor of input weight and deviation will lead to poor network performance,easy to overfit,poor generalization ability and other problems,hence the IMRFO is used to optimize the parameters of SCN.Finally,to verify the effectiveness of IMRFO optimize SCN network fault diagnosis model,some typical nonlinear dynamic system fault diagnosis neural network models are used for comparative experiments,Through experiments,it has been proven that the SCN network optimized by IMRFO has a significant improvement in the fault diagnosis accuracy of the CAFe superconducting linear accelerator system,proving the effectiveness of this method.(3)Due to the strong correlation between the system operating state at the current moment of the TE process and the operating state at the previous moment,and the inability of traditional fault diagnosis methods to extract deep feature information,resulting in low diagnostic accuracy,a TE process fault diagnosis method based on IMRFO-LSTM is proposed.Firstly,convolutional neural networks(CNN)are used to reduce the dimensionality of TE process fault data and extract features.Aiming at the problem that long short-term memory(LSTM)has limited feature extraction ability for long-term associated fault data,"peephole" is added to improve the LSTM unit structure,and the attention mechanism is added to "focus" the feature of data information in the process of network training.The ability of LSTM to extract the deep features of long-term correlation data is enhanced.Then,IMRFO is used to optimize the weight and threshold of the network,which speeds up the training speed of the network,improves the fault diagnosis and classification accuracy of the network model,and effectively improves the network performance.Finally,experiments are conducted to verify the effectiveness of the proposed network model for TE process fault diagnosis.Experiments show that the network model can have a good diagnostic effect on the fault data of TE process,a nonlinear dynamic system.