Research On Fault Diagnosis Method Of Gas Sensor Array Based On Data Drive

With the development of artificial intelligence and information technology,human access to information is developing in the field of bionics.For the machine olfaction system,the key to sensing and acquiring the gas data information to be recognized is the gas sensor array,so the reliability of the detection result plays an extremely important role in the overall performance evaluation of the entire olfaction sensing system.This article mainly studies the fault diagnosis method of gas sensor array.By analyzing the fault causes and fault signal manifestations of Metal Oxide Semiconductor(MOS)gas sensor arrays widely used in machine olfactory systems,focusing on data-driven gas sensor array fault detection,fault isolation and fault pattern recognition method.Through the data collection of the machine olfactory system based on the MOS gas sensor array,the experimental data is obtained for simulation analysis,and the effectiveness of the proposed fault diagnosis method is verified.The main research work of the paper is:(1)Aiming at the problem that the traditional data-driven fault detection method has low detection accuracy for microfaults and mixed faults,a fault detection method based on serial principal component analysis(SPCA)for gas sensor arrays is proposed.The new SPCA fault detection model uses secondary modeling to make microfaults and mixed faults stand out,making them easier to detect.It can be seen from the simulation experiment results that,compared with the fault detection methods of principal component analysis(PCA)and kernel principal component analysis(KPCA),the proposed method has higher Fault detection accuracy and lower false detection rate.(2)To achieve multi-fault isolation of gas sensor array,based on the previous stage of fault detection algorithm,a multi-fault isolation method based on SPCA for reconstruction contribution gas sensor array is proposed.Using the reconstructed value of SPE statistic of SPCA algorithm as variable contribution,combined with fault direction set theory and iterative thought,the number of fault sensors is adaptively determined.To achieve multi-fault isolation,the delay effect of traditional fault isolation method based on contribution plots is improved.Experimental results show that when three gas sensors fail at the same time,the proposed method still has more than 95% accuracy of fault isolation.(3)Fault pattern recognition is one of the important links in the fault diagnosis process of gas sensor arrays.Aiming at the problem that the fault features extracted from the existing fault pattern recognition methods have poor ability to distinguish different types of faults and lead to low fault recognition accuracy,a fault feature extraction method based on composite multi-scale weighted permutation entropy and Fisher discrimination is proposed and combined with Bagging(Bootstrap aggregating,guided aggregation algorithm)integrated learner to realize the failure mode recognition of gas sensor.This method uses composite multi-scale analysis to expand the dimension of the fault signal,and obtains the weighted arrangement entropy of the signal sequence fragments at different scales as the fault signal feature sample set.Compared with the approximate entropy,sample entropy,and permutation entropy,the weighted permutation entropy not only analyzes the complexity of the time series but also integrates the amplitude information of the sequence.So it can better describe the signal characteristics.Then use Fisher discrimination to reduce the dimensionality of the fault signal feature sample set,simplify the subsequent classifier training process,and improve the accuracy of fault pattern recognition.Finally,use the Bagging-based integrated learning method to classify the fault feature samples.This method obtains a better classification result by integrating multiple decision tree classification results.Experimental results show that compared with the existing fault recognition methods,the proposed new feature extraction method makes different types of fault features more separable and improves the accuracy of fault recognition.At the same time,the Bagging-based integrated learning method has better classification accuracy than the traditional single classifier.The proposed pattern recognition method has a fault recognition accuracy rate of over 97% for gas sensor.