Research On Small Sample And Small Fault Diagnosis Method Based On PCA

Generally speaking, the premise of fault diagnosis is to detect failure firstly, then find the location and reason of the failure using some methods and tools, and then complete fault reparation, which insures the safe operation of the system.With the advancement of the modern science and technology, the production scale of modern industry is larger and larger, and the complexity of the system is higher and higher, in addition, some industrial production keep the attitude that time is money, so the longer and longer time of system running continuously is necessary. Once part failure of the system is caused by some uncertain factors, the product quality and system performance will be ill affected, and even the system breakdown caused by the system operation failure will be appear, The safe operation of the system must be guaranteed, otherwise, it will bring great harm to personal safety and the environment, The safe operation of system is closely related to the fault diagnosis. There are various forms of fault diagnosis technology. For example, the method based on the model has a good application in fault diagnosis research, but the performance is greatly dependent on the accuracy of the system mechanism model. Actually, mechanism model with high accuracy is too difficult to establish. It’s not necessary for the fault diagnosis method based on data driven to establish accurate mechanism model, and the only one thing to do is to analyze the process data, As a result, the method attracts wide attention. The principal component analysis is an important branch of fault diagnosis methods based on the data driven.In order to solve the fault diagnosis problem and the tiny fault diagnosis problem under the small sample data in linear time invariant system, adopt a PCA based on Bayesian spatial decomposition were estimated combined CS thoughts and ideas on a variable weighting system failure analysis and research, the main research content is as follows:1) The actual process for the shortcomings in the system is sometimes influenced by the environment can only get a small sample of the fault data, and traditional PCA method when dealing with small sample fault diagnosis performance is not ideal, according to the PCA geometric sense, combined with CS decomposition Bayesian estimation ideological space proposed feature subspace estimation method. Firstly, statistical thresholds is set failure mode feature subspace matrix library is established depending on the fault feature subspace different rotation angles with small sample the feature subspace is extracted through the feature subspace estimation method, the fault diagnosis is implemented by the similarity matrix. The simulation results show that the subspace estimation method based on small samples for fault diagnosis have better diagnostic performance.2) Taking the dimension of the same system into account, the location of the different variables, and the variable is not the same degree of importance, which introduced the idea of variable weights, a new method based on variable weighted principal component analysis. Firstly, the use of offline data to establish the normal operation of the main element model; secondly to determine the importance of each variable based on a priori information analysis system; then, under the criteria of energy conservation systems, giving appropriate weight of each variable system; finally the online data sampled each sensor gives corresponding weight of the system on-line fault detection. Theoretical analysis and simulation results show that the method based on feature subspace estimation method has a good performance on the fault diagnosis with small samples.