Research On Fault Diagnosis Method Of Gas Sensor Based On Data-driven

Nowadays,sensor has been widely applied in various fields,whose reliability is increasingly paid attention.The research of fault diagnosis technique provides important guarantee for sensor being applied.Self-validating sensor is a new-type detector,which not only can output observed value,but also conduct on-line detection of its operation capacity and condition.Its research contents include intelligent detecting units such as measure theory,fault detect and digital communication technology.The focused research is based on fault detection technology of self-validating gas sensor of data driven,including fault feature extraction algorithm and fault classification algorithm.The selenographic miniature ecosphere is used as an experimental platform,examining the effectiveness of the proposed method.The paper's primary work as follows:(1)The paper researches the framework design of sensor fault-diagnosis system and basic theory,including characteristic analysis of sensor's sensing unit(technical index,operating principle,performance feature and so on),fault type of sensor,wavelet analysis,Ensemble Empirical Mode Decomposition(EEMD)and Support Vector Machine(SVM).The paper makes a detailed introduction of overall framework design of system and fundamental principles of the algorithm used in this topic.Meanwhile,the paper also gives further description of design content of the framework.(2)Based on the wavelet analysis of signal,and aimed at different data type of sensor,the paper studies fault diagnosis method of gas sensor which can computing relative wavelet energy.The paper makes a specific introduction of the method of obtaining relative wavelet energy through wavelet analysis and structural feature vector,which can further input SVM to reach the recognition of different types of fault signal.The paper makes a detailed introduction of the process of fault pattern recognition of SVM,gives specific multi-classification and identification method,and carry out verification for above-said theoretical method taking fault diagnosis of gas sensor for example.(3)Based on the EEMD decomposition of signal,and aimed at the problem of wavelet analysis parameter difficult to determine,the paper studies fault diagnosis method of gas sensor of joint information entropy and variance.The paper firstlymakes a detailed introduction of the concept of information entropy as well as feature analysis of information entropy,and discusses the limitations of the original signal extraction feature entropy.The paper further combines EEMD with such feathers as information entropy and variance to screen out IMF component decomposed and calculate eigenvalue of various fault types.At the same time,by entering the feature into SVM to achieve the separation of the fault and differentiate specific fault types,the essay gives a fault diagnosis process of gas sensor based on EEMD,and carries out verification for above-said theory by means of experimental taking gas sensor for example.(4)Aimed at the research on fault diagnosis of gas sensor,the essay designs experimental platform of miniature ecosphere.The essay gives a specific description of system experiment platform and software and hardware design,and puts forward the fault data collection method of gas sensor;then respectively conducts wavelet transform and EEMD decomposition of gas sensor signals to exact the significant relative wavelet energy characteristics and information entropy and variance characteristics that gas sensor is equipped with,after screening,construct the feature vector.Finally,SVM is used to classify and identify different fault types.It is can be seen through experiment that above-said method based on data driven both can achieve fault diagnosis of gas sensor.