Research On Sensor Fault Detection And Separation Algorithm In Mine Pressure Monitoring

With the advancement and development of science and technology,fault diagnosis has been applied universally in both military and civilian applications.This thesis takes the sensor fault of the mine pressure monitoring system as the research background.The fault detection and fault separation methods of sensors in mine pressure detection system are studied in the following four kinds of failures,such as deviation,failure,drift and accuracy decrease.The main work and achievements of this thesis are as follows:Firstly,four kinds of faults,deviation,failure,drift and accuracy,are described respectively.Principal component analysis is used to process and simulate the data based on MATLAB simulation platform.The simulation results show that the traditional principal component analysis does not accurately detect the fault data in the data when detecting the sensor fault of the mine pressure system.However,the contribution plot cannot determine the single variable fault.Secondly,the kernel function is added to improve the detection method.Analyze the influence of different parameters on the statistics of different fault models,and calculate the accuracy of the fault data generated by different parameters,and find the most suitable parameters.The simulation results show that compared with the traditional principal component analysis,the principal component analysis method based on kernel function greatly improves the probability of detecting fault data.Principal component analysis based on kernel function can well solve the shortcomings of failure of nonlinear data failure in traditional principal component analysis.Thirdly,a fault detection method based on Granger causality test is proposed.The number of failures detected by the kernel principal component is extracted.The stability of each variable is tested.Under the premise of variable stability,variable is used to perform Granger causality tests with another variable.Fault separation is completed by comparison.The simulation results show that the combination of kernel function and Granger causality test can detect the fault data and can isolate the fault variables.The above method is tested by the actual fault data of the 32213 working face of the Ru Ji Gou coal mine.The results are successfully detected and the fault variable is separated,which shows the effectiveness of the method.