Research On Self-validating Methods For Metal Oxide Semiconductor Gas Sensor Arrays

Self-validating sensor is a new generation sensor,which can effectively achieve abnormal condition monitoring and measurement quality evaluation of sensor itself by using a series of self-validating methods,such as fault detection and isolation,fault identification and measurement uncertainty evaluation,in order to significantly improve the reliability of sensor measurement process.Metal-oxide semiconductor(MOS)gas sensor array is a widely used information acquisition device in machine olfactory systems.However,due to the gas sensitive element's own characteristics,gas sensors will inevitably occur failures and interferences that result in a decline in the reliability of MOS gas sensor array,and then affect the overall system performance.Therefore,it is necessary to implement abnormal condition monitoring and measurement quality evaluation of MOS gas sensor arrays,in order to improve the credibility of the gas detection and analysis results of machine olfaction system.For that reason,this paper researches on the corresponding self-validating methods of MOS gas sensor arrays.The main contents of this dissertation are as follows:The traditional fault detection and isolation methods based on multivariate statistical analysis for MOS gas sensor array have low incipient fault detection rate.Aiming at this problem,a fault detection and isolation method of MOS gas sensor array based on sparse non-negative matrix factorization(SNMF)is proposed.The proposed method utilizes the local feature extraction ability of non-negative matrix decomposition to represent the monitoring signals through low rank approximation,and on this basis,a new C2 detecting statistic based on coefficient vectors clustering is designed to improve the sensitivity of incipient fault.To isolate the faulty sensor further,a self-adaptive multiple variables reconstruction(SMVR)fault isolation method based on squared prediction error(SPE)statistic of SNMF is proposed.This fault isolation method takes SPE statistic of fault reconstruction signal as judgement criterion,solving the smearing problem in fault isolation process,and improving the accuracy of fault isolation.The experimental results show that the proposed method can simultaneously maintain low rate of false positives,significantly improve the incipient fault detection rate,and effectively improve the accuracy of the fault isolation.It meets the requirements for abnormal condition monitoring of MOS gas sensor array itself.Sensor fault identification is an important segment to implement self-validation of MOS gas sensor arrays.This paper investigates the problem that low fault identification accuracy of fault identification methods based on time and frequency analysis caused by instability of fault signal decomposition results and poor separability of fault feature extraction.A sensor fault type identification method based on ensemble empirical mode decomposition–fast sample entropy(EEMD–FSampEn)feature extraction coupled with sparse representation classifier(SRC)is proposed.This method uses the analytical ability of EEMD adaptively decomposing non-linear and non-stationary signals to a series of intrinsic mode functions(IMFs)containing fault characteristics,and solves the instable problem of decomposed results caused by poor adaptivity of wavelet packet decomposition and mode mixing of EMD.Taking advantage of the differences of complexity of the intrinsic mode functions from different fault types,the fault feature vectors are extracted by the description capability of signal complexity by using sample entropy.Moreover,in order to improve the computational efficiency of sample entropy,a fast sample entropy based on Kd tree is adopted and it can reduce the computational complexity of sample entropy.Over-complete dictionary of SRC is constituted by training samples from different fault types,and the sensor fault type identification is determined by reconstruction error minimum between test sample and its reconstructed signal in different fault modes.The experimental results show that,compared with the existing methods,the fault feature extracted by proposed method has better separability and effectively improve fault identification accuracy.This method achieves the expected objective for abnormal condition monitoring of MOS gas sensor array.Measurement uncertainty is an important index for evaluating measurement quality in self-validating sensor.To effectively achieve measurement quality evaluation of MOS gas sensor array,a sensor measurement quality evaluation method based on process capability index(Cpk)is proposed.The proposed method can effectively evaluate probability distribution function of measured values by using grey bootstrap method(GBM)under the condition of small sample and realize uncertainty evaluation of dynamic measurement condition.The problem that traditional GUM and Monte Carlo methods are not suitable for uncertainty evaluation of dynamic measurement condition is solved.On this basis,process capability index is taken as evaluation criterion to implement measurement quality evaluation online.To solve the problem that estimated low accuracy of validated measurement value caused by the partial failure of MOS gas sensor array,a reliability evaluation model of MOS gas sensor array outputs is established to distinguish between gas response and its own fault by using grey forecasting model GM(1,1)and the correlation of multiple sensor outputs.On the basis of accurately identifying fault state,the validated measurement values are evaluated.The experimental results show that the proposed method can effectively achieve the measurement quality evaluation of MOS gas sensor array,and obtain excellent the estimated accuracy of validated measurement values of MOS gas sensor array.To verify the feasibility and effectiveness of the proposed self-validating methods of MOS gas sensor array,a machine olfactory system based on the self-validating MOS gas sensor array was designed and implemented.All gas sensors mounted on the MOS gas sensor array are calibrated and tested before using,and on-line detection and identification of binary mixture of methane and carbon monoxide gas are realized by prevalent gas analysis methods based on pattern recognition.The effectiveness of proposed key selfvalidating methods used to realize the abnormal condition monitoring and measurement quality evaluation of MOS gas sensor array is verified and the feasibility of improving the reliability of machine olfactory system is demonstrated.