Research On Sensor Fault Diagnosis Technique Based On Multiple Source Information Fusion

Automation level in the industrial field constantly improves, and the reliability of sensors influences and restricts the normal operation of the monitoring system seriously. But along with the incessant enlargement of industrial scale, the systems become more complex, the traditional sensor fault detection and diagnosis methods already cannot satisfy the requirement of the diagnostic accuracy, therefore, it is very significant to research and develop the sensor fault diagnosis technology.First, this paper introduces the development course of the fault diagnosis technique and its research situation at home and abroad. The basic principle and the advantages and disadvantages of diagnosis method are introduced, from which the information fusion technology is chosen as the main research method and expounding the causes. Basic theory of multi-source information fusion technology is introduced, and discussing some basic concepts of this method, fusion model and involved theoretical algorithm, etc.Second, the gray system theory and the BP neural network being chosen purposefully are studied, detailed analyzing of the similarities and differences of both and their advantages and disadvantages, and pointing out that constituting a gray neural network complete integration model can be better for saluting complex uncertainty problem. Based on the guiding ideology, for example of the power plant boiler drum water level, hierarchical gray neural network fusion model is designed to be suitable for the multi-sensor information fusion. It is used to fusing the main influence factors of the drum water level, and obtaining its high precision fitted value. The simulation results show that the model has fast convergence rate, high data fitting precision, its network weights and threshold values with physical meaning and other characteristics. It is more suitable than simple BP neural network model for dynamic system modeling.Finally, the drum water level sensor fault diagnosis system is constructed based on the water level control system. It detects faults through comparing the difference between output of hierarchical gray neural network fusion model and the actual output of water level sensor and setting threshold comparison. Using the field data having been pretreated, representative sensor faults are respectively analyzed in the MATLAB. The results show that the designed fault diagnosis system of this paper has good diagnosis effect for the typical sensor faults. Once detecting the fault, the fault signal could also be removed prompt by the system, unsteadied by the fusion output of model, which could eliminate the adverse impact of the fault. It is a set of good fault diagnosis system.