Fault Diagnosis And Prediction Based On Improved Networked Strong Tracking Filters

Nowadays, networked control systems have become a fashion trend during the development of society due to people's chase of high speed and high efficiency. But,because of the limited network bandwidth, there are some unstable phenomena while data are transported, such as random packet dropouts, time delays, scheduling confusion, and so on. For the networked industrial systems, useful information may be contained in the data, and their loss can cause a big disaster.This paper mainly discusses the fault diagnosis and fault prediction problem for a class of nonlinear multi-sensor dynamic systems with parameter perturbations and network-induced packet dropouts. The data transmitted via the Internet from different sensors may suffer from independent packet dropouts, and then a series of Bernoulli sequences is employed to simulate the multiple data loss rates. A small change is made in the fading factor of the traditional strong tracking filter in order to improve its good robustness against sudden changes in the network environment. Based on the improved networked strong tracking filter(INSTF), a novel system consisting of a healthy model and a real one is constructed and the bias between their estimated states is treated as a residual. A fault can be alarmed through the comparison of a cumulative residual signal and a Monte Carlo simulation method opted threshold. Meanwhile, it can also be isolated by the idea of residual contributing degree. In addition, an INSTF-based multi-step ahead forecast equation is obtained for a equivalent forecast model which is more suitable to be used for prediction after applying the state augmentation method. Also, an alternative approach about whether to trigger the fault warning or not is given. Finally, some simulation studies are carried out on an Internet-based three-tank system to show the effectiveness of the proposed approaches, and two performance indicators are introduced to verify the superiority of INSTF against extended Kalman filter.