Research On Fault Detection And Diagnosis Methods And Their Application On Networked Systems

With the development of science and technology, the system has become intelligent, automatic and complex. In the long course of their work, the system will inevitably lead to failure due to the wear or aging of the equipment. Once a fault occurs, it will cause some impact on the production process, and even property damage and casualties. Therefore, the timely and effective fault detection and diagnosis of equipment is very important. At the same time, with the popularity of computer technology and network, network control systems are increasingly appearing in the production process. Due to the limited carrying capacity and communication bandwidth of the networks, and the impact of environment blocking and interference and other factors, there are inevitably exist some problems in the network system, such as, the time delay, packet loss and so on, which further enhances the difficulty of fault detection and diagnosis for the networked system. In this context, this paper considering these questions above, focusing on fault detection and diagnosis technology and its application in the networked system, the main contents and innovations are as follows:(1) The fault detection method of weighted square sum of residuals and norm-bounded are proposed. First, for the random delay and multi-packet loss problem of data communications in the network, an optimal linear filter is designed at the sense of linear minimum variance with the method of innovation analysis, and analyse the steady-state characteristics of the filter. Second, the fault detection methods of weighted square sum of residuals and norm-bounded are proposed based on the filter designed, and the performance of these two methods is compared through simulations.(2) The fault diagnosis method based on BP neural network with an adaptive network structure and high classification accuracy is proposed. First, for the problem that the structure of the network is difficult to determine, the method of neurons number determination in the hidden layer based on the bounded AIC criteria is proposed, which provides an effective way for the optimization of neural network structure. Second, for the problem that the classification accuracy of the BP neural network is not so high in the case of small sample, the AdaboostM2 algorithm which can improve the learning ability of any system is used to optimize the BP neural network, and the BP_Adaboost strong classifier is designed. The measures to improve the probability of misclassification sample learning are also given, so that the accuracy of fault diagnosis improved greatly. Finally, the practicality and effectiveness of the method are verified through practical examples.(3) The fault diagnosis method based on consistent strength evidence combination rule is proposed. First, for the problem that the classical DS fuse rule cannot handle conflict evidence, the concept of the consistent strength is presented, and then the combined rule of evidence based on the consistent strength is also presented. The computational complexity of this rule is analysed detailed, and its effectiveness is verified through numerical examples. Finally, the diagnosis model based on the consistent strength evidence combination rule and its implementation steps is given. The practical examples show that the correct fault diagnosis result is obtained after several fault feature fusion.(4) The method proposed above is used to solve the problem of fault detection and diagnosis of mobile robots. For the fault characteristics of the sensor subsystem and the drive subsystem of the mobile robots, the fault diagnosis method of mobile robots based on the innovation cloud features and neural network is proposed. The implementation process of this method is introduced detailed and performed the example tests. In view of the problem that the diagnosis accuracy of the neural network is not high for some types of fault, the secondary fusion fault diagnosis method based on evidence theory is proposed, which use the diagnosis results of a neural network as the evident body, and then fused by evidence theory. The practical example shows that this strategy effectively improves the accuracy of diagnostic results.