Research On Self-Validating Pressure Sensor

Self-validating sensor is a new generation sensor, which not only can output the measurement value, but also can evaluate its performance and status on line. It can implement self-diagnosis and self-recovery of fault, output more and optimized information. It can improve the reliability of the measurement and control system distinctly. Until now, there are many self-validating prototypes have been developed, including self-validating temperature sensor, flow meter and dissolved oxygen Sensor, but there is no self-validating pressure sensor has been developed. Our research is supported by National Natural Science Foundation of China. The aim is to research theory and approaches of self-validating sensor technology and develop a prototype of self-validating pressure sensor. This dissertation designs the transducer structure, researches sensor fault detection and diagnosis algorithm, calculating method of self-validating parameters and develops a prototype of self-validating pressure sensor. It resolves some key technology in self-validating pressure sensor. The main contributions of this dissertation are as follows:1. The structure parameters for the circular-flat diaphragm elastic body of self-validating transducer are designed. The available area, inherent frequency, working band, linearity of the elastic body are determined by finite element analysis. After that, a multi-displacement of strain gauges on the elastic body is established and the transducer is manufactured.2. To implement fault self-detection, based on the redundancy design of the sensor transducer, consistency checking is used to detect the fault of main strain gauge. The problem of finding the high consistency group is transmitted to the maximum clique problem in graph theory, and solved using a new tow step searching method which can find all maximum clique efficiently. When the consistency checking is failed, LS-SVM predictor is used to sensor fault detection and short period data recovery, which has higher prediction and recovery accuracy, consumes less time than neural network predictors.3. To implement fault self-diagnosis, based on the fault modes analysis of the strain gauge pressure sensor, aiming at the non-stationary property of the fault signal and the small sample property of the fault diagnosis, the fault diagnosis method based on wavelet packet (WPT) decomposition, empirical mode decomposition feature extraction and hierarchical support vector machine (H-SVM) multi-classifier is researched. The wavelet function and decomposition levels of WPT are selected according to feature evaluation, energy leakage and real-time ability of the algorithm. The structure of the H-SVM multi-classifier is determined using k-means clustering method. The designed diagnosis method meets the robust request of fault diagnosis to signal amplitude and fault occurring time, and resolves the small sample problem in fault diagnosis.4. To resolve the parameter optimization problem of SVM, the niche genetic algorithm based on sharing function is researched. Using this method, the optimized parameters of LS-SVM predictor is selected with least prediction error, the optimized parameters of SVM is selected with highest identification ratio and simplest structure.5. To calculating self-validating parameters, a fusion method of the high consistency data is proposed to calculate the validated measurement value, which fully considers the influence of measurement accuracy and consistency degree of sensor on the measurement result. A validated uncertainty calculating method based on reliability mode and Bayesian theory is proposed, integrating all self-validating information into the validated uncertainty.6. The self-validating sensor hardware platform based on double DSP system is implemented. The fault detection, diagnosis and self-validating parameters calculating algorithm are implemented on this platform and experiments are done, including sensor calibration and testing. The experiments evaluate the reliability and the real-time ability of the algorithm and the self-validating function of the sensor.