Nonlinear System Fault Diagnosis Based On Fuzzy Neural Network

In the process of automation manufacture in the modern industry, fault diagnosishas become a key technology which can improve productive efficiency and ensurequality. At present, the fault diagnosis for the linear system has many researches, andthe diagnosis theory development is more mature. However, because of the factors ofthe interference of the external environment, the frame change of the internal system,the real system presents nonlinearity at varying degrees. The traditional nonlinearsystem fault diagnosis which based on the observer did not fully take the interferenceof noise, disturbance and the uncertainty of modeling into consideration, which madeit has limitation in the practical application as well as bring difficulty in designingfault threshold, and easy to result in the wrong alarm and missing alarm. A faultdiagnosis method based on fuzzy neural network observer is discussed in order tosolving the problem of the difficulty of modeling.Firstly, the development and current situation of the fault diagnosis areintroduced, and the emphasis work of the research of this article is proposed. Thenfuzzy logic, neural networks and the design of the adaptive observer are introduced indetail in this article, especially focused on the mixed type pi-sigma nerve network andRBF neural network, and made the case study simulation for the approaching abilityof the fuzzy network. Using the property of the fuzzy network of approaching anycontinuous function with any accuracy designed the adaptive observer which is basedon the fuzzy network, provided the stable constraint for the observer, and applied it inthe fault diagnosis. The result of the simulation proved that the designed observer hasgood robustness for the uncertainty of the system model, and can detect the faultoccurs timely. At last, also did research for the nonlinear system that has input andoutput failure at the same time, and made the real example simulation study for thesuggested algorithm, proved the effectiveness of the algorithm.