METHOD FOR MULTI-INFORMATION FUSION OF GAS SENSITIVITY AND CHROMATOGRAPHY AND ON-SITE DETECTION AND ANALYSIS OF FLAVOR SUBSTANCES BASED ON ELECTRONIC NOSE INSTRUMENT

Abstract

Provided is a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument. The electronic nose instrument includes a gas sensor array module (I), a capillary gas chromatographic column module (II), an automatic headspace sampling module (III), a computer control and data analysis module (IV), an automatic lifter (V) for headspace sampling, a large-volume headspace vapor generation device (VI) and two auxiliary gas sources (VII-1, VII-2). The electronic nose instrument detects a large number of odorous samples to establish a big odor data. On this basis, the normalization fusion preprocessing is done, and the cascade machine learning model realizes both an on-site recognition of many foods, condiments, fragrances and flavors, and petroleum waxes and a real-time quantitative prediction of their odor quality grades and many key component concentrations.

Claims

1. A method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument, wherein the electronic nose instrument comprises a gas sensor array module I, a capillary gas chromatographic column module II, an automatic headspace sampling module III, a computer control and data analysis module IV, an automatic lifter V for headspace sampling, a large-volume headspace vapor generation device VI and two auxiliary gas sources VII-1 and VII-2, which are configured to implement an on-site real-time detection and intelligent analysis of such flavor substances as foods, condiments, fragrances and flavors, and petroleum waxes; wherein, the gas sensor array module I comprises a gas sensor array I-1, an annular working chamber I-2 for installing the gas sensor array I-1, a thermal insulation layer I-3, a partition plate I-4, a fan I-5 and a resistance heating element I-6, and is located in a middle right part of the electronic nose instrument; the capillary gas chromatographic column module II comprises a capillary gas chromatographic column II-1, a detector II-2, an amplifier II-3, a recorder II-4, a thermal insulation layer II-5, a fan II-6, a resistance heating wire II-7 and an inlet port II-8, and is located in a right upper part of the electronic nose instrument; the automatic headspace sampling module III comprises a first micro vacuum pump III-1, a first flowmeter III-2, a first throttle valve III-3, a first two-position two-port electromagnetic valve III-4, a second two-position two-port electromagnetic valve III-5, a two-position three-port electromagnetic valve III-6, a second micro vacuum pump III-7, a third two-position two-port electromagnetic valve III-8, a fourth two-position two-port electromagnetic valve III-9, a side-hole sampling needle III-10, a first pressure relief valve III-11, a first purifier III-12, a second throttle valve III-13, a second pressure relief valve III-14, a second purifier III-15, a third throttle valve III-16, a second flowmeter III-17, a fourth throttle valve III-18 and a fifth throttle valve III-19, and is located in a right lower part of the electronic nose instrument; main constructional units of the computer control and data analysis module IV comprise an A/D data acquisition card IV-1, a driving and control circuit board IV-2, a computer mainboard IV-3, a 4-path precision DC stabilized power supply IV-4, a WIFI board card IV-5 and a display IV-6, and is located in a left side of the electronic nose instrument; main constructional units of the automatic lifter V for headspace sampling comprise a support disc V-1, a step motor V-2, a screw mechanism V-3 and a gear transmission mechanism V-4, and is located in a right front lower part of the electronic nose instrument; main constructional units of the large-volume headspace vapor generation device VI comprise a thermal insulation layer VI-1, a resistance heating wire VI-2, a heat conduction sleeve VI-3, a temperature sensor VI-4, a tested sample VI-5, a 250 ml glass sample bottle VI-6, a silicone rubber sealing sheet VI-7 and a cup cover VI-8; one electronic nose instrument is provided with 4-6 large-volume headspace vapor generation devices VI, and the role of the large-volume headspace vapor generation device VI is to make 10 ml-30 ml tested sample within the 250 ml glass sample bottle VI-6 at a constant temperature of 40-80±0.1° C. for about 30 min in a test site, and generate 220 ml-240 ml headspace vapor; the automatic lifter V is employed to make the large-volume headspace vapor generation device VI up 20 mm within 3 s, so that the side-hole sampling needle III-10 fixed under a gas inlet port of the annular working chamber I-2 penetrates through a silicone rubber sealing sheet VI-7 and contacts with headspace vapor in the 250 ml glass sample bottle VI-6; and a gas sampling period of a headspace vapor for the tested sample VI-5 by the electronic nose instrument is T=300-600 s, and T=480 s by default; in a gas sampling period T, setting the sampling time of a tested headspace vapor of the capillary gas chromatographic column module II to be earlier than that of the gas sensor array module I; in a case of T=480 s, setting the default headspace vapor sampling time of the capillary gas chromatographic column module II to be 1 s earlier than that of the module I; setting the default ratios of a flow rate, a sampling duration and an accumulated sampling volume of the gas sensor array module I to the capillary gas chromatographic column module II for a tested odor sample to be 1,000:6 ml/min, 60:1 s and 1,000:0.1 ml (theoretical value) in order; and performing, by the computer control and data analysis module IV, a sensitive information selection and analysis operation on the gas sensor array module I and the capillary gas chromatographic column module II simultaneously; in the gas sampling period T, pumping, by the first micro vacuum pump III-1 and the second micro vacuum pump III-7, the headspace vapor into the gas sensor array module I and the capillary gas chromatographic column module II, respectively, so that the gas sensor array I-1 and the capillary gas chromatographic column II-1 generate a sensitive response, respectively; obtaining, by the electronic nose instrument, 1 group of gas sensor response curves and 1 gas chromatogram to serve as an analog signal of gas sensitivity and gas chromatography obtained by perceiving an odorous sample by the electronic nose instrument; in the gas sampling period T, extracting, by the computer control and data analysis module IV, 48 response information variables from a plurality of response curves of the gas sensor array module I, selecting 21 feature information variables from a finite-duration semi-separated chromatogram of the capillary gas chromatographic column module II, and therefore obtaining, by the electronic nose instrument, a 69-dimensional response vector x(τ)∈R.sup.69, which is referred to as a pattern hereinafter; saving the response vector in a corresponding data file of a hard disk in the computer mainboard IV-3; and sending the pattern data to a cloud terminal and many specified fixed/mobile terminals through a WIFI routing module; on-site detecting, by the electronic nose instrument, various flavor substances such as foods, condiments, fragrances and flavors, and petroleum waxes for a long time over many months and years to form an big odor data X, and establishing, by a part of data of the big odor data X, a corresponding relation between a gas sensitivity/gas chromatography response and an odor type, an intensity grade and main component concentrations of the flavor substances; and in a learning stage, learning, by a cascade machine learning model of the computer control and data analysis module IV, a normalized pre-processing big odor data X offline to determine the structure and parameters of the cascade machine learning model, and learning a gas sensitivity/gas chromatography recent response online to finely tune the parameters of the cascade machine learning model; in a decision-making stage, online determining, by the cascade machine learning model, types of various foods, condiments, fragrances and flavors, and petroleum waxes according to a gas sensitivity/gas chromatography current response vector x(τ), and quantitatively predicting an intensity grade and main component concentration values of odors.

2. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein the gas sensor array I-1 and the annular working chamber I-2 are located in a thermostatic working room of 55±0.1° C.; in the gas sampling period T, the gas sensor array module I goes through: (i) a headspace sampling stage of the capillary gas chromatographic column module II for a tested odor, with a default duration of 1 s, and a default flow rate of 6 ml/min; (ii) a headspace sampling stage of the gas sensor array module I for the tested odor, with a duration of 60 s, and a flow rate of 1,000 ml/min; (iii) a transition stage, with a duration 4 s, a flow rate of 1,000 ml/min for the ambient air; (iv) a flushing stage of the ambient air, namely a rough recovery stage of the gas sensor array, with a duration of T−110 s; (v) an accurate dry air calibration stage, with a duration of 40 s; and (vi) a balance stage, i.e., a silent stage without gas flow, with a duration of 5 s, in order; where the “transition stage” realizes the transformation from the tested headspace vapor to the ambient air, and the ambient air is used for a rough recovery of the gas sensor array I-1, a flushing of the annular working chamber I-2 and the inner walls of the related gas pipelines, and a removal of the accumulated heat generated by the gas sensor array.

3. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein an [1 s, 61 s] interval of the gas sampling period T is the headspace sampling stage of the gas sensor array module I for a tested odor; in this stage, setting the second two-position two- port electromagnetic valve III-5 to be on, and setting the first two-position two-port electromagnetic valve III-4, the third two-position two-port electromagnetic valve III-8 and the fourth two-position two-port electromagnetic valve III-9 to be off, whether the two-position three-port electromagnetic valve III-6 is on or off has not effect at the moment; under a suction action of the first micro vacuum pump III-1, a headspace vapor generated by the tested sample VI-5 flows through, at a flow rate of 1,000 ml/min, the side-hole sampling needle III-10, the annular working chamber I-2 and the internal gas sensor array I-1, the second two-position two-port electromagnetic valve III-5, the first throttle valve III-3 and the first flowmeter III-2 in order, and finally is discharge to outdoor for 60 s; and during this stage, therefore, generating, by the gas sensor array 1-1, a sensitive response for the tested odor, and saving the response data in a temporary file of the computer mainboard IV-3.

4. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein a [T−45 s, T−5 s] interval of the gas sampling period T is the dry air calibration stage of the gas sensor array module I; in this stage, setting the fourth two-position two-port electromagnetic valve III-9 to be on, setting the second two-position two-port electromagnetic valve III-5 to be off; whether the other electromagnetic valves are on or off is irrelevant; and making the dry air in the dry air bottle VII-2 flows through, at a flow rate of 1,000 ml/min, the first pressure relief valve III-11, the first purifier III-12, the second throttle valve III-13, the fourth two-position two-port electromagnetic valve III-9, the gas sensor array I-1 and the side-hole sampling needle III-10 inside the annular working chamber I-2, and finally discharge the dry air to outdoor for 40 s; and during this stage, making the gas sensor array I-1 accurately recover to a reference state under the role of the dry air.

5. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein a [65 s, T−45 s] interval of the gas sampling period T is the flushing stage of the ambient air or the rough recovery stage of the gas sensor array; in this stage, setting the second two- position two-port electromagnetic valve III-5 and the first two-position two-port electromagnetic valve III-4 to be on, and setting the second two-position two-port electromagnetic valve III-4 to be off, whether the third two-position two-port electromagnetic valve III-8 and two-position three-port electromagnetic valve III-6 are on or off is irrelevant at the moment; under the suction action of the first micro vacuum pump III-1, making the ambient air flow through, at a flow rate of 6,500 ml/min, the side-hole sampling needle III-10, the annular working chamber I-2 and the gas sensor array I-1, the second two-position two-port electromagnetic valve III-5, the first two-position two-port electromagnetic valve III-4 and the first flowmeter III-2 in order, and finally discharge the ambient air to outdoor for T−110 s; and during this stage, washing away the residual odor molecules on the inner walls of the related gas pipelines, taking away the accumulated heat generated by the long-term work of the gas sensor array, and making the gas sensor array I-1 roughly recover to a reference state under the role of the ambient air.

6. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein a commercially available capillary gas chromatographic column II-1 is in a size of length×inner diameter×film thickness=L×ϕd×δ=30 m×ϕ0.53 mm×0.25 μm, and is located in an thermostatic box with a constant temperature of 250-300±0.1° C.; in the gas sampling period T, making the capillary gas chromatographic column module II undergo the following three stages: (i) a headspace sampling stage for a tested odor, with a duration of 1 s (by default), (ii) a gas chromatographic separation stage of T−11 s (by default), and (iii) a discharging, cleaning and purging stage of 10 s, wherein a duration of the headspace sampling stage for the tested odor is 0.5 s˜1.0 s, and the default value is 1 s, the range of sampling flow rate is 1.5 ml/min˜15 ml/min, and 6 ml/min by default; using H.sub.2 in the auxiliary gas source VII-1 as both a carrier gas and a fuel gas, and using the dry air in the auxiliary gas source VII-2 as a combustion-supporting gas; a [0, 1 s] interval of the gas sampling period T is the headspace sampling stage of the capillary gas chromatographic column module II for the tested odor; in this stage, setting the two-position three-port electromagnetic valve III-6 to be at “1”, setting the third two-position two-port electromagnetic valve III-8 to be on, and setting the second two-position two-port electromagnetic valve III-5 and the fourth two-position two-port electromagnetic valve III-9 to be off, whether the first two-position two-port electromagnetic valve III-4 is on or off irrelevant at the moment; under the suction action of the second micro vacuum pump III-7, making the headspace vapor of the tested sample VI-5 flow through, at a flow rate of 1.5 ml/min-15 ml/min, the third two-position two-port electromagnetic valve III-8, the two-position three-port electromagnetic valve III-6 and the fourth throttle valve III-18 in order, and mixing with the carrier gas H.sub.2 at the inlet port II-8, whereby flowing into the capillary gas chromatographic column II-1 for 0.1 s-1.5 s; in a case of the sampling flow rate of 6 ml/min and the duration of 1 s, making a cumulative sampling volume of the tested odor be 0.1 ml, and thus meeting an optimal sampling volume requirement of the capillary gas chromatographic column.

7. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein a [1 s, Ts] interval of the gas sampling period T is the chromatographic separation, discharging, cleaning and purging stage of the capillary gas chromatographic column module II for T−1 s; in this stage, setting the two-position three-port electromagnetic valve III-6 to be at “2”, whether the other two-position two-port electromagnetic valves are on or off is of little significance at this stage; due to the pushing action of carrier gas H.sub.2, separating, the tested odor in the capillary gas chromatographic column II-1, generating, by the detector II-2, a sensitive response, amplifying, through the amplifier II-3, the sensitive response, recording, by the recorder II-4, a sensitive response in an interval of [0, 470 s] or a duration of 470 s, and saving the recorded sensitive response in a temporary file of the computer mainboard IV-3; and not recording the sensitive data in an interval of [T−10 s, T] or a duration of 10 s in the discharging, cleaning and purging stage.

8. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein the gas sensor array module I and the capillary gas chromatographic column module II simultaneously enter an information selection and analysis operation region for a last 10 s of the gas sampling period T; selecting, by the computer control and analysis module IV, 3 pieces of sensitive information from the i.sup.th gas sensor in a [1 s,61 s] headspace sampling time stage of the gas sampling period T, namely, from a voltage response curve having a 60 s duration and recorded in a temporary file simultaneously, to meet a triangular stability principle and thus improve the qualitative and quantitative capability of the gas sensor array, wherein the 3 pieces of sensitive information comprises a steady-state maximum response v.sub.gs_i(τ), a corresponding peak time value t.sub.gs_i(τ), and an area A.sub.gs_i(τ) under the 60 s voltage response curve; obtaining, by the computer control and data analysis module IV, 16*3=48 sensitive variables in total from 16 response curves of the gas sensor array I-1 with 16 gas sensor elements; selecting, by the computer control and data analysis module IV, 21 sensitive variables from a semi-separated chromatogram of the capillary gas chromatographic column II-1 in an interval of [0, T−10 s] or a duration of T−10 s; where the 21 sensitive variables comprise the first 10 maximum chromatographic peaks h.sub.gc_i(τ), the 10 corresponding retention time values t.sub.gc_i(τ), and an area A.sub.gc(τ) under a chromatogram curve; in a case where the number q of chromatographic peaks of the semi-separated chromatogram with a duration of T−10 s is less than 10, selecting, by the computer control and data analysis module IV, the first q<10 maximum chromatographic peaks h.sub.gc_i(τ), the 10 corresponding retention time values t.sub.gc_i(τ), and 1 area A.sub.gc(τ) under the chromatogram curve from the semi-separated chromatogram, and performing a zero-padded operation for those not enough chromatographic peaks and retention time values; wherein the obtained chromatographic sensitive information is X.sub.gc(τ)={(h.sub.gc_1(τ), h.sub.gc_2(τ), . . . , h.sub.gc_q(τ), 0, . . . , 0; (t.sub.gc_1(τ), t.sub.gc2(τ), . . . , t.sub.gc_q(τ), 0, . . . , 0; A.sub.gc(τ)}; and in the gas sampling period T, fusing, by the computer control and data analysis module IV, through a normalized pre-processing, 48 sensitive variables extracted from the 16 response curves of the gas sensor array I-1 and 21 sensitive variables extracted from the semi-separated chromatogram of the capillary gas chromatographic column II-1 to obtain a sensitive vector x(τ)∈R.sup.69 with m=48+21=69 dimensions, saving the sensitive vector x(τ)∈R.sup.69 in a specified file of a hard disk of the computer mainboard IV-3; and using as a numerical basis of doing a qualitative and quantitative analysis on foods, condiments, fragrances and flavors, and petroleum waxes by the electronic nose instrument.

9. The method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument of claim 1, wherein the computer control and data analysis module IV employs a modular cascade neural network model to perform (i) identification and (ii) sensory quality indicator score and main component quantitative prediction for the foods, condiments, fragrances and flavors, and petroleum waxes; wherein (i) a first level of the modular cascade neural network model comprises n(n−1)/2 single-output neural networks in parallel to form n vote recognition groups, and the n vote recognition groups are used for identifying n foods, condiments, fragrances and flavors, and petroleum waxes, comprising brands, producing places, authenticity and fragrance types, one for one; (ii) a second level of the modular cascade neural network model comprises nxq single-output neural networks in parallel; q single-output neural networks form one group and are used for quantitatively predicting q indices comprising odor intensity grades and main component concentrations of n foods, condiments, fragrances and flavors, and petroleum waxes; in a learning stage of the first level (i) of the modular cascade neural network model, making the big odor data, namely a training set X, be done an one-to-one decomposition based on brands, production places, authenticity and odor types to form n(n−1)/2 binary-class training subsets; then, learning the n(n−1)/2 subsets by using n(n−1)/2 single-output neural networks in the first level of the modular cascade neural network model with an error back-propagation algorithm, each for one; making all the single-output neural networks be single-hidden-layer in structure, that is, the number of input nodes is m=69, the number of hidden nodes is s.sub.1=8, and the number of output nodes is 1; making the n(n−1)/2 single-output neural networks form n vote recognition groups, and making each single-output neural network takes and only take part into two vote recognition groups among; in a learning stage of the second level (ii) of the modular cascade neural network model, making the big odor data, namely the training set X, be done the one-to-one decomposition based on brands, production places, authenticity and odor types again, to form n×r single-output regression training subsets; fitting, by n×q single-output neural networks in the second level, to fit the multi-input single-output nonlinear curves for the q indices, each for one, wherein the q indices comprise a sensitive vector x.sub.p of gas sensitivity and gas chromatography, and making all the single-output neural networks be single-hidden-layer in structure, i.e., the number of input nodes is m=69, the number of hidden nodes is s.sub.2=5, and the number of output nodes is 1; and identifying, by the modular cascade neural network model, n kinds of foods, condiments, fragrances and flavors, and petroleum waxes by using a majority vote decision-making rule; a decision-making rule for identifying an unknown pattern x is that x belongs to a brand, a production place, a authenticity and a type of the foods, condiments, fragrances and flavors, and petroleum waxes represented by a vote recognition group with the most votes; on the premise that the vote number of a vote recognition group Ω.sub.j is the max in the first level of the cascade model, and predicting an intensity grade and main component concentration values of x, by a quantitation prediction group Λ.sub.j in the second level of the cascade model corresponding to the vote recognition group Ω.sub.j.

10. The method for performing an on-site detection and analysis of flavor substances by using multi-information fusion technology of gas sensitivity and gas chromatography of an electronic nose instrument of any one of claims 1 to 9, wherein performing, by the electronic nose instrument, the real-time on-site detection and prediction of various foods, condiments, fragrances and flavors, and petroleum waxes comprises the following steps: (1) Power-on: setting a preheating time length of the electronic nose instrument to be 30 min; and setting a “gas sampling period T” value to be a default value T=8 min in the screen menu; setting the first two-position two-port electromagnetic valve III-4 and the second two-position two-port electromagnetic valve III-5 to be on, setting the fourth two-position two-port electromagnetic valve III-9 to be off, and making the ambient air flow through, at a flow rate of 6,500 ml/min, the side-hole sampling needle III-10, the gas sensor array I-1 inside the annular working chamber I-2, the second two-position two-port electromagnetic valve III-5, the first two-position two-port electromagnetic valve III-4 and the first flowmeter III-2 in order, and finally discharging to outdoor; and making an internal temperature of the annular working chamber I-2 of the gas sensor array reach a constant temperature of 55±0.1° C.; setting the two-position three-port electromagnetic valve III-6 to be at “2”, under the pushing action of the carrier gas H.sub.2, making the capillary gas chromatographic column II-1 gradually recover to a reference state, and making the internal temperature in a chromatographic column box reach a constant temperature of 250-300±0.1° C.; preparation and temperature constant of a tested sample; firstly pipetting by an operator, 10-30 ml tested sample VI-5 into the glass sample bottle VI-6, then placing the glass sample bottle VI-6 into the heat conduction sleeve VI-3 of the large-volume headspace vapor generation device VI, covering with the silicone rubber sealing sheet VI-7, and screwing the cup cover VI-8; pressing an confirmation key to start a constant-temperature time, starting to heat the tested sample VI-5, rising the temperature to 45-80±0.1° C. from the room temperature within 8 min, and accurately keeping the temperature constant for 20-30 min; (2) starting the gas sampling period T of the k.sup.th tested sample VI-5, and taking T=8 min as an example below; (2.1) the gas sensor array module I: (2.1a) the headspace sampling stage: in an 1 s-61 s interval of the gas sampling period T, setting the second two-position two-port electromagnetic valve III-5 to be on, setting the first two-position two-port electromagnetic valve III-4, the third two-position two-port electromagnetic valve III-8 and the fourth two-position two-port electromagnetic valve III-9 to be all off, and under the suction action of the first micro vacuum pump III-1, making the headspace vapor of the tested sample VI-5 flow through, at a theoretical flow rate of 1,000 ml/min, the side-hole sampling needle III-10, the annular working chamber I-2 plus the gas sensor array I-1, the second two-position two-port electromagnetic valve III-5, the first throttle valve III-3 and the first flowmeter III-2 in order, and finally, discharging the used headspace vapor to outdoor for 60 s, whereby generating, by the gas sensor array I-1, a sensitive response for the tested odor, and saving a response data in a temporary file of the computer mainboard IV-3 for 60 s, and making the initial 0 s-1 s to be the headspace sampling stage of the capillary gas chromatographic column II; (2.1b) the transition stage: in a 61 s-65 s interval of the gas sampling period T, setting the second two-position two-port electromagnetic valve III-5 to be on, and setting the first two-position two-port electromagnetic valve III-4, the third two-position two-port electromagnetic valve III-8 and the fourth two-position two-port electromagnetic valve III-9 to be off; making, by the automatic lifter V, the large-volume headspace vapor generation device VI descend by 20 mm within 3 s; keeping the first micro vacuum pump III-1 to suck at a flow rate of 1,000 ml/min, and along with the descending of the large-volume headspace vapor generation device VI, and making the gas flowing inside the annular working chamber I-2 gradually transition to the ambient air from the headspace vapor of the tested sample VI-5; (2.1c) the rough recovery stage: in a 65 s-435 s interval of the gas sampling period T, setting the first two-position two-port electromagnetic valve III-4 and the second two-position two-port electromagnetic valve III-5 to be on, setting the fourth two-position two-port electromagnetic valve III-9 to be off, and under the suction action of the first micro vacuum pump III-1, making the ambient air flow through, at a flow rate of 6,500 ml/min, the side-hole sampling needle III-10, the gas sensor array I-1 inside the annular working chamber I-2, the second two-position two- port electromagnetic valve III-5, the first two-position two-port electromagnetic valve III-4 and the first flowmeter III-2 in order, and finally discharging the ambient air to outdoor for 370 s, whereby making the gas sensor array I-1 roughly recover to a reference state under the role of the ambient air; (2.1d) the accurate calibration stage: in a 435 s-475 s interval of the gas sampling period T, setting the fourth two-position two-port electromagnetic valve III-9 to be on, setting the second two-position two-port electromagnetic valve III-5 to be off, and making the dry air in the dry air bottle VII-2 flow through, at a flow rate of 1,000 ml/min, the first pressure reducing valve III-11, the first purifier III-12, the second throttle valve III-13, the fourth two-position two-port electromagnetic valve III-9, the gas sensor array I-1 inside the annular working chamber I-2, the side-hole sampling needle III-10 in order, and finally discharging the dry air to indoor for 40 s, whereby making the gas sensor array I-1 accurately recover to a reference state; in the reference state, placing, by the operator, a tested sample on the support disc V-1 of the automatic lifter V through the large-volume headspace vapor generation device VI in a precise thermostatic process, and preparing next headspace sampling period; and (2.1e) the balance stage, in a 475 s-480 s interval of the gas sampling period T, setting all the two-position two-port electromagnetic valves to be off, and making no gas flow inside the annular working chamber I-2 of the gas sensor array for 5 s; from the 475.sup.th second, making, by the automatic lifter V, the large-volume headspace vapor generation device VI up 20 mm within 3 s; (2.2) the capillary gas chromatographic column II module: (2.2a) the headspace sampling stage: in a 0 s-1 s interval of the gas sampling period T, setting the two-position three-port electromagnetic valve III-6 to be at “1”, setting the third two-position two-port electromagnetic valve III-8 to be on, and setting the second two-position two-port electromagnetic valve III-5 and the fourth two-position two-port electromagnetic valve III-9 to be off; under the suction action of the second micro vacuum pump III-7, making the headspace vapor of the tested sample VI-5 flow through, at a default flow rate of 6 ml/min, the third two-position two-port electromagnetic valve III-8, the two-position three-port electromagnetic valve III-6 and the fourth throttle valve III-18 in order, mixing with the carrier gas H.sub.2 at the inlet port II-8 and then flowing into the capillary gas chromatographic column II-1 for 1.0 s, wherein an accumulated sampling volume of the tested headspace vapor is 0.1 ml by default; (2.2b) the gas chromatographic separation stage: in an [1 s, 480 s] interval of the gas sampling period T, setting the two-position three-port electromagnetic valve III-6 to be at “2”, and under the pushing action of the carrier gas H.sub.2, separating, the tested headspace vapor, in the 30 m capillary gas chromatographic column II-1; generating, by the detector II-2, a sensitive response, amplifying, by the amplifier II-3, the sensitive response, recording, by the recorder II-4, the sensitive response with a [0, 470 s] interval or a 470 s duration to form a semi-separated chromatogram, and saving the semi-separated chromatogram in the temporary file of the computer mainboard IV-3; (2.3) the information selection and analysis stage: in a 470 s-480 s interval of the gas sampling period T, selecting, by the computer control and data analysis module IV, 3 pieces of sensitive information, i.e., a steady-state peak value v.sub.gs_i(τ), a corresponding peak time value t.sub.gs_i(τ), and an area A.sub.gs_i(τ) under the curve from the voltage response curve of the i.sup.th gas sensor which is recorded in the [1 s, 60 s] time stage; selecting 21 sensitive response variables, i.e., the first 10 max chromatographic peak values h.sub.gc_i(τ), the 10 corresponding retention time values t.sub.gc_i(τ), and one area A.sub.gc(τ) under a chromatogram curve from the semi-separated chromatogram which is recorded in a [0 s, 470 s] stage, and saving them in the temporary file of the computer mainboard IV-3; whereby, obtaining, by the computer control and data analysis module IV, one sensitive vector x(τ)∈R.sup.69 with 69 dimensions from the sensitive information of the gas sensor array module I and the capillary chromatographic column module II for the tested sample VI-5; then, performing, by the cascade machine learning model, the type identification of a tested odor and the quantitative prediction of its whole intensity as well as main component concentrations s according to the sensitive vector x(τ), displaying, by the monitor, a detection and prediction result, and then transmitting the result to a central control room and a plurality of fixed/mobile terminals through the Internet; and (3) repeat the step (2), realizing, by the electronic nose instrument, the real-time on-site detection and identification of foods, condiments, fragrances and flavors, and petroleum waxes samples and quantitative prediction of their intensity and main component concentration index values.

Description

BRIEF DESCRIPTION OF DRAWINGS

[0073] FIG. 1 is a schematic diagram of working principle of the electronic nose instrument, i.e., a schematic diagram of headspace sampling state of the gas sensor array module, in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0074] FIG. 2 is a schematic diagram of working principle of the electronic nose instrument, i.e., a schematic diagram of headspace sampling state of the capillary gas chromatographic column, in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0075] FIG. 3 is a schematic diagram of working principle of the electronic nose instrument, i.e., a schematic diagram of a rough recovery state of a gas sensor array and a separation state of a capillary gas chromatographic column, in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0076] FIG. 4 is a schematic structural diagram of a gas sensor array module and a capillary gas chromatographic column module in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0077] FIG. 5 is a schematic diagram of the automatic lifter for headspace sampling in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0078] FIG. 6 is a schematic diagram of the headspace vapor generation device in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0079] FIG. 7 is a schematic diagram of changes in gas sampling time points, flow rates, and sensitive responses of a capillary chromatographic column module and a gas sensor array module in a gas sampling period T=480 s in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0080] FIG. 8 is a schematic diagram of multi-information selection of a semi-separated chromatogram in a gas sampling period T=480 s in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0081] FIG. 9 is a schematic diagram of feature selection of two semi-separated chromatograms in a gas sampling period T=480 s in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0082] FIG. 10 is a schematic diagram of multi-feature selection of a gas sensor response curve in a gas sampling period T=480 s in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

[0083] FIG. 11 is a schematic diagram of an offline learning and online decision-making process in a machine learning model in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure.

DETAILED DESCRIPTION

[0084] The present disclosure will be further described in details by combining with the accompanying drawings.

[0085] FIG. 1 is the schematic diagram of working principle of the electronic nose instrument in a method for multi-information fusion of gas sensitivity and chromatography and on-site detection and analysis of flavor substances using an electronic nose instrument provided in the present disclosure. Main constructional units of the electronic nose instrument include a gas sensor array module I, a capillary gas chromatographic column module II, an automatic headspace sampling module III, a computer control and data analysis module IV, an automatic lifter V, a large-volume headspace vapor generation device VI, a hydrogen bottle VII-1 and a dry air bottle VII-2. FIG. 1 shows a headspace sampling state of the gas sensor array module I. Hydrogen H.sub.2 is used both as the carrier gas for the capillary gas chromatographic column module II and as the fuel gas for a Hydrogen flame ionization detector (FID). The dry air is used not only as the combustion-supporting gas of FID in the capillary chromatographic column module II, but also as the calibration gas (not combusted) in the gas sensor array module I.

[0086] FIG. 2 is the schematic diagram of working principle of the electronic nose instrument in the headspace sampling state of the capillary gas chromatographic column, in the present disclosure.

[0087] Main constructional units of the gas sensor array module I include the gas sensor array I-1, the annular working chamber I-2, the thermal insulation layer I-3, the partition plate I-4, the fan I-5 and the resistance heating element I-6, and is located in the middle right part of the electronic nose instrument. Main constructional units of the capillary gas chromatographic column module II include the capillary gas chromatographic column II-1, the detector II-2, the amplifier II-3, the recorder II-4, a thermal insulation layer II-5, the fan II-6, the resistance heating wire II-7 and the inlet port II-8, and is located in the right upper part of the electronic nose instrument. The gas sensor array module I and the capillary gas chromatographic column module II are used for converting the chemical and physical information of an odor into the electric signals online.

[0088] The constructional units of the automatic headspace sampling module III include the first micro vacuum pump III-1, the first flowmeter III-2, the first throttle valve III-3, the first two-position two-port electromagnetic valve III-4, the second two-position two-port electromagnetic valve III-5, the two-position three-port electromagnetic valve III-6, a second micro vacuum pump III-7, the third two-position two-port electromagnetic valve III-8, the fourth two-position two-port electromagnetic valve III-9, the side-hole sampling needle III-10, the first pressure relief valve III-11, the first purifier III-12, the second throttle valve III-13, the second pressure relief valve III-14 and the second purifier III-15, the third throttle valve III-16, the second flowmeter III-17, the fourth throttle valve III-18 and the fifth throttle valve III-19, and is located in the right lower part of the electronic nose instrument.

[0089] Main constructional units of the computer control and data analysis module IV include an A/D data acquisition card IV-1, a driving and control circuit board IV-2, a computer mainboard IV-3, a 4-path precision DC stabilized power supply IV-4, a WIFI board card IV-5 and a displayer IV-6, and is located in the left side of the electronic nose instrument. The role of the WIFI board card IV-5 is to real-time transmit the sensitive information of the gas sensor array module I and the capillary gas chromatographic column module II to multiple specified fixed/mobile terminals.

[0090] Main constructional units of the automatic lifter V include the support disc V-1, the step motor V-2, the screw mechanism V-3 and the gear transmission mechanism V-4, and is located in the right front lower part of the electronic nose instrument. Main constructional units of the large-volume headspace vapor generation device VI include the thermal insulation layer VI-1, the resistance heating wire VI-2, the heat conduction sleeve VI-3, the temperature sensor VI-4, the tested sample VI-5, the 250 ml glass sample bottle VI-6, the silicone rubber sealing sheet VI-7 and the cup cover VI-8. One electronic nose instrument is equipped with 4-6 large-volume headspace vapor generation devices VI. The role of the large-volume headspace vapor generation device VI is to make 10 ml-30 ml tested sample within the 250 ml glass sample bottle VI-6 at the constant temperature of 40-80±0.1° C. for about 30 min in a certain test site and generate 220 ml-240 ml headspace vapor. The role of the automatic lifter V is to make the headspace vapor generation device VI up 20 mm within 3 s, in order the side-hole sampling needle III-10 fixed under the inlet port of the annular working chamber I-2 penetrate through the silicone rubber sealing sheet VI-7 on 250 ml glass sample bottle VI-6 and thus contact with the headspace vapor in the glass sample bottle VI-6.

[0091] FIG. 3 is the schematic diagram of working principle of an electronic nose instrument, i.e., the schematic diagram of the rough recovery state of the gas sensor array and the separation state of the capillary gas chromatographic column. FIG. 4 is the schematic diagram of the gas sensor array module I and the capillary gas chromatographic column module II, which may be easily replaced as needed. At this moment, under the action of the automatic lifter V, the large-volume headspace vapor generation device VI descends 20 mm to the original position along with the support disc V-1 and then is taken away by an operator, and the preparation is made for replacing a new headspace vapor generation device and detecting a new sample.

[0092] FIG. 5 gives the detailed structure of the automatic lifter V. The ratio of gear tooth numbers in the gear transmission mechanism V-4 is 17:73, and the gear module is 1 mm. The step motor V-2 drives the screw of the screw mechanism V-3 to ascend through the gear transmission mechanism V-4, so that the large-volume headspace vapor generation device VI placed on the support disc V-1 ascends.

[0093] FIG. 6 is the schematic diagram of the mutual positional relationship and the detailed structure of the large-volume headspace vapor generation device VI. When testing, the operator takes 10-30 ml liquid or solid tested sample VI-5 to place in the 250 ml glass sample bottle VI-6, covers the bottle mouth with the silicone rubber seal sheet VI-7, and tightens the cup cover VI-8. Under the heating action of the resistance heating wire VI-2 which is controlled by the computer control and data analysis module IV, the tested sample VI-5 is kept at the constant temperature of 40-80±0.1° C. about 30 min to ensure the consistency of multiple tests.

[0094] FIG. 7 are the schematic diagrams of changes about the gas sampling time lengths, the flow rates and the responses of a specified gas sensor given by the gas sensor array module I and the capillary gas chromatographic column module II in the electronic nose instrument in the gas sampling period T=480 s. The gas sampling period may be adjusted in the range of T=5 min and T=10 min. FIG. 7 only uses the default gas sampling period T=480 s as an example. The adjustable time length is mainly the flushing stage of the ambient air or the rough recovery stage of the gas sensor array module I, and the separation stage of the capillary gas chromatographic column module II. For the gas sensor array module I and the capillary gas chromatographic column module II, the information selection and analysis are performed simultaneously in the last lOs stage of the gas sampling period T.

[0095] FIG. 7(a) shows the change of cyclical gas sampling cases for the capillary chromatographic column module II, which includes such 3 stage: (i) the tested gas sampling stage, (ii) the tested gas separation stage, and (iii) the chromatographic column discharging stage. The tested gas sampling stage (i) is at a beginning stage of the gas sampling period with a sampling duration of 0.5 s˜1.0 s, and is by default. The range of the sampling flow rates is 1.5 ml/min˜15 ml/min, 6 ml/min by default.

[0096] Referring to FIG. 7, and in combination with FIG. 2, Table 2 shows the operation parameters and the on-off states of the related electromagnetic valves of the capillary gas chromatographic column module II in the gas sampling period T=480 s. In the tested gas sampling stage (i), the two-position three-port electromagnetic valve III-6 is located at “1”, the third two-position two-port electromagnetic valve III-8 is on, and the second two-position two-port electromagnetic valve III-5 and the fourth two-position two-port electromagnetic valve III-9 are off, and the on-off status of the first two-position two-port electromagnetic valve III-4 is irrelevant to at this stage. Under the suction action of the second micro vacuum pump III-7, the headspace vapor of the tested sample VI-5 flows through, at a flow rate of 6 ml/min, the third two-position two-port electromagnetic valve III-8, the two-position three-port electromagnetic valve III-6 and the fourth throttle valve III-18 in order, and then mixes with the carrier gas H.sub.2 at the inlet port II-8, flows into the capillary gas chromatographic column II-1 accordingly for 1.0 s. If the sampling flow rate is 6 ml/min and the duration is 1 s, then the accumulated sampling volume for a tested odor is 0.1 ml, which meets the requirement of the optimal sampling volume of the capillary gas chromatographic column. In the tested gas separation stage (ii) and the chromatographic column discharging stage (iii), since the two-position three-port electromagnetic valve III-6 is located at “2”, and whether the other two-position two-port electromagnetic valves are on or off is not critical. During this period, under the pushing action of the carrier gas H.sub.2, the tested odor is separated in the capillary gas chromatographic column II-1.

[0097] Referring to FIG. 6 and in combination with FIG. 1 and FIG. 3, Table 3 shows the operating parameters and the on-off status of the related electromagnetic valve for the gas sensor array module I in the gas sampling period T.

[0098] Several main working states of the gas sensor array module I are described below in details by taking the gas sampling period T=480 s as an example.

TABLE-US-00002 TABLE 2 Operation parameters and on-off status of the related electromagnetic valves for the capillary gas chromatographic column module II in the gas sampling period T = 300 s-600 s (480 s by default) 2-position 2-position 2-position 2-position 2-position Initial Flow 3-port 2-port 2-port 2-port 2-port Duration time- rate Gas valve valve valve valve valve Stage Description (s) point (s) (ml/min) type III-6 III-8 III-4 III-5 III-9 (i) Gas sampling 0.5-1.5 0 1.5~15  Tested odor “1” On Irrelevant Off Off (ii) Chromatographic 289-589 1 30~50 Tested odor + “2” Irrelevant Irrelevant Irrelevant Irrelevant separation H.sub.2 (iii) Chromatographic 10 290-590 30~50 H.sub.2 “2” Irrelevant Irrelevant Irrelevant Irrelevant column discharging

TABLE-US-00003 TABLE 3 Operation parameters and on-off status of the electromagnetic valves for the gas sensor array module I in the gas sampling period T = 300 s-600 s (480 s by default) 2-position 2-position 2-position 2-position 2-position Flow 2-port 2-port 2-port 2-port 2-port Duration Initial rate Gas valve valve valve valve valve Stage Description (s) moment (s) (ml/min) type III-5 III-4 III-6 III-8 III-9 (i) Headspace 60 1 1,000 Tested odor On Off Irrelevant Off Off sampling (ii) Transition  4 61 1,000 Purified air On Off Irrelevant On Off (iii) Rough recovery 175-475 65 6,500 Purified air On On Irrelevant Irrelevant Off (iv) Clean air 40 255-555 1,000 Clean air Off Irrelevant Irrelevant Irrelevant On calibration (v) Balance  5 295-595 0 — Off Irrelevant Irrelevant Off Off

[0099] In the headspace sampling stage (i) for the tested odor, namely, the [1 s, 61 s] time stage with a duration of 60 s in the gas sampling period T, the second two-position two-port electromagnetic valve III-5 is on, the first two-position two-port electromagnetic valve III-4, the third two-position two-port electromagnetic valve III-8 and fourth two-position two-port electromagnetic valve III-9 are all off, and the on-off status of the two-position three-port electromagnetic valve III-6 has not effect. Under the suction action of the first micro vacuum pump III-1, headspace vapor of the tested sample flows through, at a flow rate of 1,000 ml/min, the side-hole sampling needle III-10, the gas sensor array I-1 inside the annular working chamber I-2, the second two-position two-port electromagnetic valve III-5, the first throttle valve III-3 and the first flowmeter III-2 in order, and finally is discharged to outdoor for 60 s. During this stage, the gas sensor array I-1 generates a sensitive response to the tested odor.

[0100] In the rough recovery stage of the gas sensor array, namely, the clean ambient air flushing stage (iii), and the second two-position two-port electromagnetic valve III-5 and the first two-position two-port electromagnetic valve III-4 are on, the fourth two-position two-port electromagnetic valve III-9 is off, and the on-off status of the third two-position two-port electromagnetic valve III-8 and the two-position three-port electromagnetic valve III-6 has not effect at this stage. The ambient air flows through, at a flow rate of 6,500 ml/min, the side-hole sampling needle III-10, the gas sensor array I-1 inside the annular working chamber I-2, the second two-position two-port electromagnetic valve III-5, the first two-position two-port electromagnetic valve III-4 and the first flowmeter III-2 in order, and finally is discharged to outdoor for 370 s. During this stage, the residual odor molecules on an inner walls of the relevant gas pipelines are washed away, the accumulated heat by the gas sensor array is taken away, and the gas sensor array I-1 is roughly recovered to the reference state under the action of the ambient air.

[0101] In the accurate dry air calibration stage (iv), namely the time stage of from 435.sup.th sec to the 475.sup.th sec of the gas sampling period T, the fourth two-position two-port electromagnetic valve III-9 is on and the second two-position two-port electromagnetic valve III-5 is off, and the on-off status of the other electromagnetic valves is irrelevant. The dry air in the dry air bottle VII-2 flows through, at the flow rate of 1,000 ml/min, the first pressure reducing valve III-11, the first purifier III-12, the second throttle valve III-13, the fourth two-position two-port electromagnetic valve III-9, the gas sensor array I-1 and the side-hole sampling needle III-10 inside the annular working chamber I-2 in order, and finally is discharged to outdoor for 40 s. During this stage, the gas sensor array I-1 is accurately recovered to the reference state under the role of the dry air.

[0102] According to FIGS. 7(a) and 7(d), in the last 10 s stage of the gas sampling period T, the gas sensor array module I and the capillary gas chromatographic column module II enter the information selection and analysis operation region simultaneously for 10 s.

[0103] FIG. 8 is the schematic diagram of information selection of the semi-separated chromatogram in the gas sampling period T=480 s. In the information selection and analysis region of 10 s in the gas sampling period T, the computer control and data analysis module IV sequentially selects 21 feature variables: 10 pairs of {peak height h.sub.gcj, retention time point t.sub.gcj} (j=1, 2, . . . , 10) and one under-curve area A.sub.gc from the semi-separated chromatogram with the appointed 460 s time length, which are the basic sensitive information of the capillary chromatographic column module II to the tested odor and are recorded as x.sub.gc={(h.sub.gc1, h.sub.gc2, . . . , h.sub.gc10); (t.sub.gc1, t.sub.gc2, . . . , t.sub.gc10); A.sub.gc}.

[0104] FIG. 9 is the schematic diagram of feature selection of two semi-separated chromatograms in the gas sampling period T=480 s. The semi-separated chromatogram in FIG. 9(a) has only 8 chromatographic peaks, or only 8 peaks h.sub.gci(i=1, 2, . . . , 8) and the corresponding 8 retention time values t.sub.gci(i=1, 2, . . . , 8) and plus the under-curve area A.sub.gc are obtained from the semi-separated chromatogram. Our practice is to fill ‘0’s elements for the insufficient chromatographic peaks and the corresponding retention time values. Through doing so, the final chromatographic sensitive information is x.sub.gc={(h.sub.gc1, h.sub.gc2, . . . , h.sub.gc8, 0, 0); (t.sub.gc1, t.sub.gc2, . . . , t.sub.gc8, 0, 0); A.sub.gc} according to FIG. 9(a). The semi-separated chromatogram in FIG. 9(b) has more than 10 chromatographic peaks, and thus the top 10 maximum chromatographic peaks are selected from them.

[0105] FIG. 10 is the schematic diagram of multi-feature selection of response curves of gas sensors in the gas sampling period T=480 s. Three illustrations in FIG. 10 respectively show the response curves of 3 gas sensors, TGS822, TGS826 and TGS832, for a petroleum wax sample, a 2,000 ppm ethylene gas and a 5,000 ppm ethanol vapor. The steady-state maximum voltage response values in FIGS. 10(b) and 10(c) are equal, or v.sub.b=v.sub.c. According to the conventional feature selection method of the single steady-state maximum value from a single voltage response curve, the electronic nose instrument cannot distinguish the 2,000 ppm ethylene gas and the 5,000 ppm ethanol vapor at that time. After careful observation, it is found that FIG. 10(b) and FIG. 10(c) show such a case, called ‘Case 1’: although the voltage response steady-state maximum response values in the two diagrams are equal, the corresponding peak time values and the under-curve area are not equal. Similarly, Case 2 shows that the corresponding peak time values are equal to one another, but their peak values and the under-curve areas are not equal. Case 3 shows that the under-curve areas are equal to one another, but their peak values and the corresponding peak time values are not equal.

TABLE-US-00004 TABLE 4 Comparison of the main operation parameters between the gas sensor array module I and the capillary gas chromatographic column module II (taking the gas sampling period T = 8 min as an example) Information Headspace Starting Flow Chamber selection and Meaning of Module duration time-point rate Sampling Fuel temperature analysis time information size Module name (s) (s) (ml/min) manner Carrier gas (° C.) length (s) components (mm) capillary gas 0.5-1.5 0 1.5-15 Automatic H.sub.2 H.sub.2 200-250 470-480 10 peak heights, 300 × chromatographic 10 retention time-points, 300 × column one under-curve area 120 Gas sensor array 30~60 1 1,000 Automatic None None 55 470-480 16 peak voltages, 300 × 16 peak time-points, 300 × 16 under-curve areas 100

[0106] According to FIG. 10, the present disclosure proposes the following viewpoint: three pieces of information, i.e., a steady-state maximum voltage response v.sub.gsi, a corresponding peak time value t.sub.gsi from the starting moment of the headspace sampling, and an area A.sub.gsi under the curve in the 60 s headspace sampling time stage, are selected simultaneously from the response curve of the gas sensor i. If the gas sensor array is composed of 16 sensitive elements, then in the lOs information selection and analysis region of the gas sampling period T, the computer control and data analysis module IV sequentially selects 3*16=48 feature values from 16 response curves as the basic sensitive information of the gas sensor array module I to the tested odor, which is recorded as x.sub.gs={(v.sub.gs1, v.sub.gs2, . . . , v.sub.gs16); (t.sub.gs1, t.sub.gs2, . . . , t.sub.gs16); (A.sub.gc1, A.sub.gc2, . . . , A.sub.gc16)}.

[0107] In the 10 s information selection and analysis region of the gas sampling period T, the computer control and data analysis module IV fuses the sensitive information of the gas sensor array module I and the capillary chromatographic column module II in different time stages to the tested odor, and then performs the normalization pretreatment to obtain a sensitive information vector of the electronic nose instrument to the tested sample, i.e., x=x.sub.gs+x.sub.gc={(v.sub.gs1, v.sub.gs2, . . . , v.sub.gs16); (t.sub.gs1, t.sub.gs2, . . . , t.sub.gs16); (A.sub.gc1, A.sub.gc2, . . . , A.sub.gc16); (h.sub.gc1, h.sub.gc2, . . . , h.sub.gc10); (t.sub.gc1, t.sub.gc2, . . . , t.sub.gc10); A.sub.gc}∈R.sup.69. The sensitive vector x∈R.sup.69 is the basis of online identification of types and quantitative prediction of main components by the electronic nose instrument for foods, condiments, fragrances and flavors, and petroleum waxes.

[0108] Table 4 shows the comparison of the main operation parameters between the gas sensor array module I and the capillary gas chromatographic column module II by taking the gas sampling period T=8 min as an example. In comparison with the 60 s sampling duration and the 1,000 ml/min flow rate of the former, the corresponding terms of the latter are only 1 s and 6 ml/min. According to the above data, the gas sampling volume of the gas sensor array module I is 1,000 ml, but that of the capillary gas chromatographic column module II is only 0.1 ml, or the difference between the two is 10,000 times.

[0109] In the present disclosure, the semi-separated chromatogram is regarded as a part of the sensitive information or pattern of the electronic nose instrument, and thus the big odor data is established by combining with the sensitive information of the gas sensor array. Furthermore, the recognition, qualitative analysis and main component quantitative prediction of an undermined odor are realized by means of the artificial intelligence or machine learning method. FIG. 11 is the schematic diagram of offline learning and online decision-making process of the modular cascade machine learning model adopted in the present disclosure.

[0110] In the offline learning stage of the modular cascade machine learning model, the primary task is to establish the big odor data, including the online sensitive data of the gas sensor array module I and the capillary gas chromatographic column module II for a large number of foods, condiments, fragrances and flavors, and petroleum waxes, the offline measuring data of the conventional instruments such as gas chromatography/mass spectrometry; the label data and the sensory evaluation data for the known types and constituents of odors.

[0111] Next, the sensitive data of both the gas sensor array and the capillary gas chromatographic column are fused, including the normalization and dimensionality reduction preprocessing. In order to reduce the analysis difficulty of the big odor data, the present disclosure employs the “divide and conquer” strategy, (i), a complex multi-type recognition problem is decomposed into multiple simpler two-odor recognition problems, that is, one n-class problem is decomposed into n(n−1)/2 binary-class problems {X.sub.j, X.sub.k} and then solved by n(n−1)/2 single-output machine learning models {ω.sub.jk}.sub.n, where j, k=1, 2 . . . , n and j≠k, one by one; and (ii) a complex multi-component estimation problem is decomposed into multiple simpler single-component quantitative prediction problems, one by one, that is, a q-curve/q-surface fitting problem is decomposed into q curve/surface fitting problems, and q single-output machine learning models are used for solving them, one by one.

[0112] The n(n−1)/2 single-output machine learning models ω.sub.jk(j, k=1, 2, . . . , n and j≠k) in the first level and the q single-output machine learning models in the second level form the modular cascade machine learning model. Among them, each single-output machine learning model both in the first level and in the second level may be one single-output single-hidden-layer neural network, one decision tree, one support vector machine, etc. Here, the present disclosure employs single-output single-hidden-layer neural networks. The offline learning algorithm of the modular cascade machine learning model is mainly the error back-propagation algorithm, which mainly learns the labeled data and the data with the known constituents in the big odor data.

[0113] In the decision-making stage, the n(n−1)/2 single-output machine learning models in the first level form n vote recognition groups {Ω.sub.j}.sub.n, j=1, 2, . . . , n. A single-output machine learning model takes and only takes part in two vote recognition groups. For example, the single-output single-hidden-layer neural network ω.sub.jk votes not only in the vote recognition group Ω.sub.j but also in the group Ω.sub.k. The undetermined pattern x belongs to the class represented by the vote recognition group with the most votes, i.e., the winning vote recognition group. Then, the quantitative prediction group corresponding to the winning vote recognition group in the second level predicts multiple quantitative indicator values of the tested odor x, i.e., the overall intensity, the quality grade and the concentrations of multiple main components.