Research On Failure Diagnosis And Recovery Method For Self-validating Sensor

The self-validating sensor is a novel sensor, which can not only collect dataand output the value of test, but also can detect its work state online. The self-validating sensor can achieve self-diagnosis and the self-restore of the data.Compared with the traditional sensor, the output value of the result is moreabundant and optimization. In this way, the stability and reliability of measurementand control system greatly are improved. The core part of self-validating sensor isthe Fault Diagnosis, Isolation and Recovery(FDIR).Once fault occurred, the sensorcan find the position and the reason, isolate the fault, and use the best estimatevalue temporarily to replace the error output value and complete the data recoveryonline, which provide time for replacing the faulty unit, and ensuring the systemrunning stably. The main research work is as follows:1. A fault diagnosis model based on Principal Component Analysis-RelevanceVector Machine (PCA-RVM) is studied and designed. In order to verify thefeasibility of the method, the fault simulation platform is designed throughresearching on the common faults on the sensitive unit and the reason, mechanism,and form once it occurs. The platform simulates the sensor failure effectively andcreates the simulation data, which is used in FDIR.2. For the fault diagnosis unit, this Thesis studies and designs a kind of faultdetection mode based on the Principal Component Analysis (PCA). The intrinsicassociation among different sensitive units is analyzed by using principalcomponent analysis method. By comparing the statistic of SPE with a threshold, thestate of the sensitive units whether is faulted is estimated. Once the failure of thesensitive unit occurs, the method can detect the failure effectively. If multi-faultoccurs in multiple sensitive unit, the PCA algorithm can simultaneously detectmultiple faults of different sensitive units.3. For the Failure isolation and Recovery, the method based on RelevanceVector Machine (RVM) is designed. Compared with the ability for the predictionand the performance of the RBF, Cauchy and Cubic kernel function and the Morlet,Mexican Hat Wavelet kernel function, the combination kernel function is used onthe RVM predictor. Compared with the former kernel function, the ability to resist noise and modeling speed are improved. Compared with the BP neural network andRBF neural network, the model of RVM online fault isolation and restoration cangreatly improves the real-time performance and the accuracy of failure recovery.The testing results show that the model based on PCA-RVM method caneffectively implement the Fault Diagnosis, Isolation and Recovery. Compared withthe traditional method, this method has a great advantage. The design meets therequirements.