Drift Elimination Method Of Electronic Nose Signal Of White Spirit And A Preliminary Study On Drift Phenomenon Under The Condition Of No Load

The electronic nose(e-nose)is considered to be the most promising analytical instruments for volatile sample testing.In recent years,e-nose is widely used in the field of food testing.However,the drift phenomenon of sensor makes e-nose can not be long-term robust detection of food and other samples in the actual application.Therefore,it is important to eliminate or reduce drift effects to achieve long-term robust testing of e-nose.In the process of long-term identification of six kinds of Chinese white spirit samples,the drift phenomenon is very obvious.The identification model constructed by a priori data could not predict the late sample data effectively and accurately.Therefore,for the long-term monitoring of six kinds of white spirit,how to reduce the impact of drift and achieve long-term stable detection of e-nose is the work that must be studied.In this paper,a recursive drift correction method based on wavelet packet transform was proposed for the long-term robust detection of six kinds of white spirit samples from the perspective of signal transformation.The method mainly included constructing the correction function of the wavelet packet decomposition coefficient and determining the sample test time window(SMTW)to construct the detection model.The drift signal of e-nose of six kinds of white spirit samples were well corrected by means of the correction function of the constructed wavelet packet decomposition coefficient.By comparing the differences between the different models constructed under different SMTW,the SMTW was determined to be six months.After recursively correcting the sample data in SMTW,a recursive Fisher discriminant analysis(FDA)identification model was established,and the recursive FDA model could accurately identify the sample data for two months after SMTW.Therefore,through the SMTW of six months moving forward,FDA models were also constantly recursively corrected,and could be real-time for the next two months of the correct identification of the sample data,the minimum accuracy of the identification was up to 96.5%.Thus,the long-term robust identification of six kinds of white spirit samples is realized by the above method.Finally,considering the drift is an inherent behavior of gas sensor,this paper discusses the phenomenon of e-nose drift from the no sample(no load)condition,and gives the feasible method of drift elimination under no load condition in order to lay the foundation for the drift elimination under no load condition.