WO2021147274A1 - 一种气敏-气相电子鼻仪器和发酵-恶臭多状态参数在线分析方法 - Google Patents
一种气敏-气相电子鼻仪器和发酵-恶臭多状态参数在线分析方法 Download PDFInfo
- Publication number
- WO2021147274A1 WO2021147274A1 PCT/CN2020/102885 CN2020102885W WO2021147274A1 WO 2021147274 A1 WO2021147274 A1 WO 2021147274A1 CN 2020102885 W CN2020102885 W CN 2020102885W WO 2021147274 A1 WO2021147274 A1 WO 2021147274A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- gas
- iii
- way solenoid
- valve iii
- solenoid valve
- Prior art date
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 133
- 238000000855 fermentation Methods 0.000 claims abstract description 98
- 230000004151 fermentation Effects 0.000 claims abstract description 97
- 230000004044 response Effects 0.000 claims abstract description 62
- 238000002347 injection Methods 0.000 claims abstract description 33
- 239000007924 injection Substances 0.000 claims abstract description 33
- 238000000034 method Methods 0.000 claims abstract description 23
- 230000014759 maintenance of location Effects 0.000 claims abstract description 22
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 15
- 239000007789 gas Substances 0.000 claims description 454
- 238000005070 sampling Methods 0.000 claims description 152
- 238000001514 detection method Methods 0.000 claims description 140
- 238000003965 capillary gas chromatography Methods 0.000 claims description 54
- 238000004817 gas chromatography Methods 0.000 claims description 32
- 230000008447 perception Effects 0.000 claims description 28
- 238000000926 separation method Methods 0.000 claims description 24
- 238000010801 machine learning Methods 0.000 claims description 23
- 230000009471 action Effects 0.000 claims description 22
- 238000010586 diagram Methods 0.000 claims description 22
- 238000011084 recovery Methods 0.000 claims description 20
- 239000012159 carrier gas Substances 0.000 claims description 19
- 150000001875 compounds Chemical class 0.000 claims description 19
- 238000012544 monitoring process Methods 0.000 claims description 19
- 230000007774 longterm Effects 0.000 claims description 18
- 230000035945 sensitivity Effects 0.000 claims description 18
- 239000003344 environmental pollutant Substances 0.000 claims description 16
- 231100000719 pollutant Toxicity 0.000 claims description 16
- 230000007613 environmental effect Effects 0.000 claims description 13
- 238000000746 purification Methods 0.000 claims description 12
- 239000000126 substance Substances 0.000 claims description 12
- 238000004140 cleaning Methods 0.000 claims description 11
- 230000006870 function Effects 0.000 claims description 10
- 239000011159 matrix material Substances 0.000 claims description 10
- 230000007704 transition Effects 0.000 claims description 10
- QGJOPFRUJISHPQ-UHFFFAOYSA-N Carbon disulfide Chemical compound S=C=S QGJOPFRUJISHPQ-UHFFFAOYSA-N 0.000 claims description 9
- 238000004451 qualitative analysis Methods 0.000 claims description 9
- 238000013375 chromatographic separation Methods 0.000 claims description 8
- 238000013022 venting Methods 0.000 claims description 8
- 238000009413 insulation Methods 0.000 claims description 7
- 238000004949 mass spectrometry Methods 0.000 claims description 7
- BDERNNFJNOPAEC-UHFFFAOYSA-N propan-1-ol Chemical compound CCCO BDERNNFJNOPAEC-UHFFFAOYSA-N 0.000 claims description 7
- 230000000087 stabilizing effect Effects 0.000 claims description 7
- QGZKDVFQNNGYKY-UHFFFAOYSA-N Ammonia Chemical compound N QGZKDVFQNNGYKY-UHFFFAOYSA-N 0.000 claims description 6
- QMMFVYPAHWMCMS-UHFFFAOYSA-N Dimethyl sulfide Chemical compound CSC QMMFVYPAHWMCMS-UHFFFAOYSA-N 0.000 claims description 6
- LSDPWZHWYPCBBB-UHFFFAOYSA-N Methanethiol Chemical compound SC LSDPWZHWYPCBBB-UHFFFAOYSA-N 0.000 claims description 6
- PPBRXRYQALVLMV-UHFFFAOYSA-N Styrene Chemical compound C=CC1=CC=CC=C1 PPBRXRYQALVLMV-UHFFFAOYSA-N 0.000 claims description 6
- 230000004913 activation Effects 0.000 claims description 6
- 238000011010 flushing procedure Methods 0.000 claims description 6
- 238000010438 heat treatment Methods 0.000 claims description 6
- 238000004445 quantitative analysis Methods 0.000 claims description 6
- 230000001953 sensory effect Effects 0.000 claims description 6
- 238000002798 spectrophotometry method Methods 0.000 claims description 6
- GETQZCLCWQTVFV-UHFFFAOYSA-N trimethylamine Chemical compound CN(C)C GETQZCLCWQTVFV-UHFFFAOYSA-N 0.000 claims description 6
- 230000000694 effects Effects 0.000 claims description 5
- 239000002737 fuel gas Substances 0.000 claims description 5
- 230000002045 lasting effect Effects 0.000 claims description 5
- 239000002243 precursor Substances 0.000 claims description 5
- 238000010926 purge Methods 0.000 claims description 5
- WRMNZCZEMHIOCP-UHFFFAOYSA-N 2-phenylethanol Chemical compound OCCC1=CC=CC=C1 WRMNZCZEMHIOCP-UHFFFAOYSA-N 0.000 claims description 4
- ULGZDMOVFRHVEP-RWJQBGPGSA-N Erythromycin Chemical compound O([C@@H]1[C@@H](C)C(=O)O[C@@H]([C@@]([C@H](O)[C@@H](C)C(=O)[C@H](C)C[C@@](C)(O)[C@H](O[C@H]2[C@@H]([C@H](C[C@@H](C)O2)N(C)C)O)[C@H]1C)(C)O)CC)[C@H]1C[C@@](C)(OC)[C@@H](O)[C@H](C)O1 ULGZDMOVFRHVEP-RWJQBGPGSA-N 0.000 claims description 4
- 206010037544 Purging Diseases 0.000 claims description 4
- WQOXQRCZOLPYPM-UHFFFAOYSA-N dimethyl disulfide Chemical compound CSSC WQOXQRCZOLPYPM-UHFFFAOYSA-N 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000012545 processing Methods 0.000 claims description 4
- RWSOTUBLDIXVET-UHFFFAOYSA-N Dihydrogen sulfide Chemical compound S RWSOTUBLDIXVET-UHFFFAOYSA-N 0.000 claims description 3
- 229930182555 Penicillin Natural products 0.000 claims description 3
- JGSARLDLIJGVTE-MBNYWOFBSA-N Penicillin G Chemical compound N([C@H]1[C@H]2SC([C@@H](N2C1=O)C(O)=O)(C)C)C(=O)CC1=CC=CC=C1 JGSARLDLIJGVTE-MBNYWOFBSA-N 0.000 claims description 3
- 229910021529 ammonia Inorganic materials 0.000 claims description 3
- 230000001186 cumulative effect Effects 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 229940049954 penicillin Drugs 0.000 claims description 3
- WLJVXDMOQOGPHL-PPJXEINESA-N 2-phenylacetic acid Chemical compound O[14C](=O)CC1=CC=CC=C1 WLJVXDMOQOGPHL-PPJXEINESA-N 0.000 claims description 2
- 238000006243 chemical reaction Methods 0.000 claims description 2
- 238000010411 cooking Methods 0.000 claims description 2
- 125000004122 cyclic group Chemical group 0.000 claims description 2
- 229960003276 erythromycin Drugs 0.000 claims description 2
- 230000010365 information processing Effects 0.000 claims description 2
- 244000144972 livestock Species 0.000 claims description 2
- LRHPLDYGYMQRHN-UHFFFAOYSA-N n-Butanol Substances CCCCO LRHPLDYGYMQRHN-UHFFFAOYSA-N 0.000 claims description 2
- 238000005192 partition Methods 0.000 claims description 2
- 244000144977 poultry Species 0.000 claims description 2
- 239000013558 reference substance Substances 0.000 claims description 2
- 230000008786 sensory perception of smell Effects 0.000 claims description 2
- 239000010865 sewage Substances 0.000 claims description 2
- 235000013555 soy sauce Nutrition 0.000 claims description 2
- 239000000758 substrate Substances 0.000 claims description 2
- 235000021419 vinegar Nutrition 0.000 claims description 2
- 239000000052 vinegar Substances 0.000 claims description 2
- 238000013528 artificial neural network Methods 0.000 claims 1
- 230000004927 fusion Effects 0.000 abstract description 10
- 238000004587 chromatography analysis Methods 0.000 abstract description 4
- 230000001360 synchronised effect Effects 0.000 abstract 1
- 235000019645 odor Nutrition 0.000 description 102
- 210000001331 nose Anatomy 0.000 description 84
- 238000005516 engineering process Methods 0.000 description 10
- 239000000203 mixture Substances 0.000 description 10
- 239000000463 material Substances 0.000 description 7
- LFQSCWFLJHTTHZ-UHFFFAOYSA-N Ethanol Chemical compound CCO LFQSCWFLJHTTHZ-UHFFFAOYSA-N 0.000 description 6
- 229910052799 carbon Inorganic materials 0.000 description 6
- 230000008859 change Effects 0.000 description 6
- 238000012360 testing method Methods 0.000 description 6
- UFHFLCQGNIYNRP-UHFFFAOYSA-N Hydrogen Chemical compound [H][H] UFHFLCQGNIYNRP-UHFFFAOYSA-N 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 5
- 229910052739 hydrogen Inorganic materials 0.000 description 5
- 239000001257 hydrogen Substances 0.000 description 5
- 238000001228 spectrum Methods 0.000 description 5
- 238000003491 array Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 4
- 238000010187 selection method Methods 0.000 description 4
- 230000035943 smell Effects 0.000 description 4
- 230000008094 contradictory effect Effects 0.000 description 3
- 230000007547 defect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 3
- 238000004904 shortening Methods 0.000 description 3
- 238000010897 surface acoustic wave method Methods 0.000 description 3
- OKTJSMMVPCPJKN-UHFFFAOYSA-N Carbon Chemical compound [C] OKTJSMMVPCPJKN-UHFFFAOYSA-N 0.000 description 2
- VGGSQFUCUMXWEO-UHFFFAOYSA-N Ethene Chemical compound C=C VGGSQFUCUMXWEO-UHFFFAOYSA-N 0.000 description 2
- 239000005977 Ethylene Substances 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 239000002322 conducting polymer Substances 0.000 description 2
- 229920001940 conductive polymer Polymers 0.000 description 2
- 238000007791 dehumidification Methods 0.000 description 2
- 239000000428 dust Substances 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 235000013305 food Nutrition 0.000 description 2
- 150000002484 inorganic compounds Chemical class 0.000 description 2
- 229910010272 inorganic material Inorganic materials 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 239000000243 solution Substances 0.000 description 2
- 239000012855 volatile organic compound Substances 0.000 description 2
- 101710178035 Chorismate synthase 2 Proteins 0.000 description 1
- 101710152694 Cysteine synthase 2 Proteins 0.000 description 1
- 230000005526 G1 to G0 transition Effects 0.000 description 1
- 229910006404 SnO 2 Inorganic materials 0.000 description 1
- UCKMPCXJQFINFW-UHFFFAOYSA-N Sulphide Chemical compound [S-2] UCKMPCXJQFINFW-UHFFFAOYSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 235000013405 beer Nutrition 0.000 description 1
- 238000009835 boiling Methods 0.000 description 1
- 230000002301 combined effect Effects 0.000 description 1
- 238000002485 combustion reaction Methods 0.000 description 1
- 230000001351 cycling effect Effects 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000007418 data mining Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 239000003205 fragrance Substances 0.000 description 1
- 125000000524 functional group Chemical group 0.000 description 1
- 230000036541 health Effects 0.000 description 1
- 238000010921 in-depth analysis Methods 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 239000007788 liquid Substances 0.000 description 1
- 229910044991 metal oxide Inorganic materials 0.000 description 1
- 150000004706 metal oxides Chemical class 0.000 description 1
- LPUQAYUQRXPFSQ-DFWYDOINSA-M monosodium L-glutamate Chemical compound [Na+].[O-]C(=O)[C@@H](N)CCC(O)=O LPUQAYUQRXPFSQ-DFWYDOINSA-M 0.000 description 1
- 235000013923 monosodium glutamate Nutrition 0.000 description 1
- 239000004223 monosodium glutamate Substances 0.000 description 1
- 238000011527 multiparameter analysis Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 150000002894 organic compounds Chemical class 0.000 description 1
- 238000007500 overflow downdraw method Methods 0.000 description 1
- 230000001590 oxidative effect Effects 0.000 description 1
- 239000012169 petroleum derived wax Substances 0.000 description 1
- 235000019381 petroleum wax Nutrition 0.000 description 1
- 229960003424 phenylacetic acid Drugs 0.000 description 1
- 239000003279 phenylacetic acid Substances 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 108090000623 proteins and genes Proteins 0.000 description 1
- 239000010453 quartz Substances 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- VYPSYNLAJGMNEJ-UHFFFAOYSA-N silicon dioxide Inorganic materials O=[Si]=O VYPSYNLAJGMNEJ-UHFFFAOYSA-N 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 238000009834 vaporization Methods 0.000 description 1
- 230000008016 vaporization Effects 0.000 description 1
- PHXATPHONSXBIL-UHFFFAOYSA-N xi-gamma-Undecalactone Chemical compound CCCCCCCC1CCC(=O)O1 PHXATPHONSXBIL-UHFFFAOYSA-N 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
- G01N30/88—Integrated analysis systems specially adapted therefor, not covered by a single one of the groups G01N30/04 - G01N30/86
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N1/00—Sampling; Preparing specimens for investigation
- G01N1/02—Devices for withdrawing samples
- G01N1/22—Devices for withdrawing samples in the gaseous state
- G01N1/24—Suction devices
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N33/00—Investigating or analysing materials by specific methods not covered by groups G01N1/00 - G01N31/00
- G01N33/0004—Gaseous mixtures, e.g. polluted air
- G01N33/0009—General constructional details of gas analysers, e.g. portable test equipment
- G01N33/0027—General constructional details of gas analysers, e.g. portable test equipment concerning the detector
- G01N33/0036—General constructional details of gas analysers, e.g. portable test equipment concerning the detector specially adapted to detect a particular component
- G01N33/0047—Organic compounds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
- G01N30/88—Integrated analysis systems specially adapted therefor, not covered by a single one of the groups G01N30/04 - G01N30/86
- G01N2030/8809—Integrated analysis systems specially adapted therefor, not covered by a single one of the groups G01N30/04 - G01N30/86 analysis specially adapted for the sample
- G01N2030/8813—Integrated analysis systems specially adapted therefor, not covered by a single one of the groups G01N30/04 - G01N30/86 analysis specially adapted for the sample biological materials
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N30/00—Investigating or analysing materials by separation into components using adsorption, absorption or similar phenomena or using ion-exchange, e.g. chromatography or field flow fractionation
- G01N30/02—Column chromatography
- G01N30/86—Signal analysis
Definitions
- the present invention-a gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online detection and analysis method is oriented to the needs of multi-source online detection and multi-parameter analysis of the process represented by biological fermentation and malodor pollution, and involves artificial intelligence, Computers, bioengineering, environmental protection, analytical chemistry and other technical fields mainly solve the problem of low sensitivity of gas sensors, single sensing information selection method causes poor selectivity of gas sensor arrays, and complete separation of peaks and peaks.
- a series of problems such as the optimized combination of array and gas chromatography column structure, the selection and fusion of multi-source sensing information, etc., to achieve the goal of long-term cycling multi-source online detection and multi-parameter online analysis of the biological fermentation and malodor pollution process by the electronic nose instrument.
- Olfactory simulation-The electronic nose method uses multiple gas sensors with overlapping performance to form an array to achieve rapid detection of odors, and uses machine learning methods for qualitative and quantitative analysis of odors. Odor online detection and analysis technology has become the core application technology of bioengineering, environmental protection, food and other industries. Electronic nose instruments have attracted attention because of their fast speed, non-contact, and easy operation. The current research status of electronic nose theory and technology is that the sensitivity of the gas sensor has reached the order of 10 -7 (V/V), which is 0.1ppm, but the selectivity is poor, which leads to the stability of the electronic nose instrument and the poor linearity and qualitative and quantitative ability; The nose and other products are even more of an international blank. In the context of major needs, electronic nose technology has been included in the "863", scientific and technological support and key research and development plans of the Ministry of Science and Technology for many times.
- the premise of multi-parameter real-time estimation, prediction and control of biological fermentation and malodor pollution processes is online detection of multiple process parameters.
- the environmental odor pollution time spans years and months; the biological fermentation process is short 1-2 days, long tens of days (such as beer fermentation), months or even years. It is a bit exaggerated to describe the changes in the state of the biological fermentation and foul-smelling process as "fast-changing", but the detection and analysis cycle in units of "hours" must be too long.
- the state of the tested object will generally not change much within 1 minute, that is, the detection cycle is less than 1 minute. It is not necessary; conversely, the biological fermentation or malodor pollution state may change greatly within 1 hour.
- the basic premise of the electronic nose instrument's "continuous online” detection and analysis is that its core, the gas sensor array, has a significant ability to perceive the measured object.
- the performance indicators that the gas sensor should achieve include: sufficiently high sensitivity (above ppm level), fast enough response speed (within 1min), stable working state, high degree of commercialization, and long life (3-5 years) , Its own size is small, and the selectivity is better.
- the literature [1] lists metal oxide semi-conductor (MOS) type, electrochemical (EC) type, and conducting polymer (CP) type. , Quartz microbalance (QMB) type, surface acoustic wave (SAW) type, photo ionization detector (PID) type, the sensing performance of the 6 commonly used gas sensors.
- MOS metal oxide semi-conductor
- EC electrochemical
- CP conducting polymer
- QMB Quartz microbalance
- SAW surface acoustic wave
- PID photo ionization detector
- the EC type gas sensor has better selectivity, but is much larger in size, has a life span of at least one year, and has a lower sensitivity by an order of magnitude or more.
- the PID type sensor is not only large in size, narrow in sensing range, high in price, but also has a life span of only about half a year.
- the EC type and PID type gas sensors are only suitable for the detection of malodorous pollutants.
- the sensitivity of QMB and SAW type gas sensors is more than one order of magnitude lower than that of MOS type.
- the sensitive film material needs to be further developed and the size needs to be further reduced.
- the MOS type gas sensor represented by SnO 2 is most suitable for the sensing element of the electronic nose instrument.
- the single-type gas sensor and its array made of the above 6 kinds of sensitive materials have very limited sensing capabilities and do not meet the requirements for online detection of process objects such as biological fermentation and malodor pollution.
- 2The existing electronic nose is not sensitive to the odor of a pig farm.
- the sensitivity of the gas sensor has reached the order of 10 -7 (V/V), but this is only for a specific MOS sensor to a specific odor component, and is not a common phenomenon.
- the most typical example is the detection of malodorous pollutants by electronic nose instruments and the prediction of the concentration of main malodorous compounds.
- the specific indicators specified by GB14554 include ammonia NH 3 , hydrogen sulfide H 2 S, carbon disulfide CS 2 , trimethylamine C 3 H 9 N, methyl mercaptan CH 4 S, methyl sulfide C 2 H 6 S, dimethyl disulfide C 2 H 6 S 2 , styrene C 8 H 8, a total of 8 specific compound concentration index values, plus the odor unit OU (odor unit) value, referred to as 8+1 odor pollutant concentration control index value.
- odor unit OU odor unit
- Gas chromatography has good selectivity, but MOS gas sensor has poor selectivity. However, this difference is only relative, and the "qualitative ability" of gas chromatography for unknown samples is still “weak". That is to say, in the absence of internal/external standard sample spectra, it is impossible to determine the composition and composition of unknown samples based on the spectra obtained from one measurement.
- the second defect of gas chromatography is that the "selectivity" of chromatographic columns is not universal. Only under certain conditions, a certain chromatographic column is sensitive to a certain sample, that is, a certain chromatographic column can only detect a certain range of a certain sample. When the sampling conditions, test conditions, or the parameters of the chromatographic column itself change, the chromatographic sensing parameters of a specific sample change accordingly.
- the odor is a mixture of tens, hundreds, or even thousands of compounds, and the molecular weight of all components is less than 300 Dalton. Retention time is an important qualitative analysis parameter of chromatography, and the chromatographic retention time of 8 malodorous compounds specified by GB14554 is mostly less than 8min.
- a capillary column with a larger inner diameter, such as ⁇ 0.53mm, and the column length can be 30m.
- the GC constant temperature studio is designed and manufactured; hydrogen is used as carrier gas and fuel gas, and the temperature is programmed and the gas to be measured.
- the sampling and carrier gas pushing processes are precisely controlled; the capillary column and the entire module should be easily replaced and installed.
- the sample flow rate of the tested gas can be 1.0-15ml/min, and the sample time can be 0.5 ⁇ 1.5sec.
- the sample time can be 0.5 ⁇ 1.5sec.
- the semi-separation/incomplete separation phenomenon of the chromatogram is the result of the combined effect of many factors such as the measured gas composition, the characteristics of the chromatographic column, the working parameter setting of the chromatograph, the performance of the detector, and the recording time of the recorder. Incomplete or half separation of chromatographic peaks is a common phenomenon, and complete separation is only an ideal or limit situation. The more components of the object to be measured, the more difficult it is to separate the peaks and peaks completely, at the cost of a long detection time. In the marathon that started for a period of time, although the winners and runners-up were not produced, the trend of winning and losing has been divided. The winners and runners-up are among the "teams running in front of the competition team".
- the semi-separated chromatogram is a part of the full-separated chromatogram, which is equivalent to the "team running in front of the competition team" in a marathon.
- the semi-separated chromatogram obtained from the same sample at different times will remain unchanged, and the positional relationship between the semi-separated chromatogram and the fully separated chromatogram will also remain unchanged.
- the semi-separation chromatogram can be used to infer some of the main characteristics of the full separation chromatogram, for example, to infer the presence and content of some long-retention components that do not appear on the semi-separation chromatogram.
- process analysis such as biological fermentation and foul-smelling pollution, we only need to get the information that reflects the main state parameters.
- the semi-separation chromatogram actually contains the main information of the full separation chromatogram. The key is how to get the required information from the figure. Information and explain.
- a single chromatographic column and a single-type sensitive material gas sensor array are limited.
- Gas chromatography is more difficult to analyze inorganic substances and easily decomposed high-boiling organic substances, and it is more difficult to qualitatively identify unknown substances. It is not suitable for the analysis of some single compounds with strong polarity or complex compounds with large differences in polarity, and some do not contain carbon. compound of.
- gas chromatography using a hydrogen flame ionization detector (FID) cannot effectively detect inorganic compounds. This is the reason why the present invention proposes an on-line detection and analysis method of an electronic nose instrument that integrates a gas sensor array and a capillary gas chromatography column.
- GB14554 stipulates: NH 3 and CS 2 these two kinds of odor pollutants concentration are measured by spectrophotometry, H 2 S, C 3 H 9 N, CH 4 S, C 2 H 6 S, C 2 H 6 The concentration of 6 kinds of odor pollutants including S 2 , C 8 H 8, etc. were detected by gas chromatography. It is worth noting that the three national standards GB/T14676-14678 respectively specify the gas chromatography detection methods for the last six odorous pollutants, and the detectors, chromatographic columns and working conditions are actually different from each other. We found that these national standards specify the use of 2 different sizes of packed columns to measure 6 kinds of malodorous compounds.
- a single chromatographic column cannot simultaneously detect the six malodorous compounds specified by GB14554.
- the choice of a chromatographic column must consider many factors such as the column's own material, stationary phase, inner diameter, film thickness, column length, polarity and non-polarity of the sample to be tested.
- the single-type gas sensor array has poor selectivity, limited overlapping sensing range, and insufficient sensitivity, which does not meet the online detection requirements for objects such as biological fermentation and odor pollution.
- Chromatography has the advantages of high sensitivity and good selectivity.
- the disadvantage is that the separation time is long, that is, the detection period is long, the instrument structure is complicated, and the working conditions are harsh.
- the existing usage is not suitable for online odor detection at all.
- the second defect of gas chromatography is that the "selectivity" of chromatographic columns is not universal.
- a certain chromatographic column is sensitive to a certain sample, that is, a certain chromatographic column can only detect a certain range of a certain sample.
- the chromatographic sensing parameters of a specific sample change accordingly.
- the third defect of gas chromatography is that it is difficult or even impossible to achieve "complete separation" of multi-component chromatographic peaks. The more the components, the closer the polarity between the components, the closer the retention time, the more difficult it is to completely separate the peaks and peaks.
- the gas sensor has the advantages of fast response speed and low working conditions, but the disadvantage is that the selectivity is poor and the sensitivity is not ideal.
- the GC method has the advantages of high sensitivity and good selectivity.
- the disadvantage is that the separation time is long, that is, the detection period is long, the instrument structure is complicated, and the working conditions are harsh. The existing usage is completely unsuitable for long-term online detection. It must be pointed out that the difference of "GC column selectivity is better, MOS gas sensor selectivity is poor" is only relative, and the "qualitative ability" of gas chromatography for unknown samples is still “weak".
- the gas sensor array and the capillary gas chromatography column are in sharp contrast, and the fusion of the two can achieve the effect of complementing each other and complement each other.
- the characteristics of odor are: (1) The composition is numerous and changes all the time. Taking malodorous pollutants as an example, except for a few inorganic substances such as H 2 S, NH 3 , and SO 2 , most of the smelling components are organic substances, namely "Volatile Organic Compounds (VOCs)". (2) Some ingredients have very low olfactory thresholds, but their contribution to odor intensity is great; vice versa. A dilemma encountered in practical applications of electronic noses is that some components contribute little to the odor intensity, but the gas sensor is very sensitive; vice versa. The gas sensor is used for odor online detection, and its performance indicators should include: high enough sensitivity, fast enough response speed, stable working state, high degree of commercialization, long life, small size, and good selectivity. We should deeply understand the characteristics of different gas sensors and design small gas sensor array modules to effectively solve the problems of poor stability, noise elimination, temperature and humidity compensation, and convenient replacement.
- the MOS gas sensor represented by S n O 2 material has a very fast response speed to some odors, for example, it only takes 2s for ethanol volatilization gas; while for other odors, the response speed is very slow, even up to 60s. Or longer, such as the perception of the volatile gas of ⁇ -undecanolide C 11 H 20 O 2 specified by GB/T14675.
- This phenomenon tells us that although the steady-state maximum value of the response curve of the same gas sensor to two odors may be the same, the peak time and the area under the curve may be different; or the area under the curve may be the same, but the steady-state maximum value may be the same as the maximum value. The peak time may be different, and so on.
- the shape of the response curve of the gas sensor is related to the odor composition, involving many factors such as molecular weight, carbon number, polarity, and functional groups.
- Triangular stability means that three sides (straight lines) are connected end to end to form a stable structure, which is not deformed under force. Parallelograms are easily deformed under force and are unstable; similarly, polygons with more than 3 sides are unstable.
- the inspiration given by the principle of triangle stability is that it is impossible to determine a triangle structure only by knowing two of the parameters (2 side lengths, 2 included angles, 1 side length and 1 included angle); In addition, knowing only one parameter (two cases of one side length and one included angle) is even worse.
- T 0 10min
- the perceptual information characteristics of the fermentation object or malodorous pollutants are linear in order to improve the response speed of gas chromatography.
- an electronic nose instrument can detect multiple fermentation tanks or multiple odor pollution observation points in a specific area at the same time, 24 hours a day, in units of years and months, which can be fixed point detection or mobile point detection;
- Big data such as health, finance, transportation, commerce, and genes are profoundly changing the way people live and work.
- big data on the ecological environment has been put on the agenda, and the government environmental protection department is vigorously promoting it.
- the present invention is based on the existing invention patents "A system and method for multi-point centralized online monitoring and analysis of malodorous gas” (see application number: 2018104716131), “Big data-driven online analysis method of multi-point centralized electronic nose instrument for malodorous gas” “(See application number: 2018104717083) and “A multi-channel integrated olfactory simulation instrument and on-line analysis method of biological fermentation process” (see application number: 201310405315.X), invented an electronic nose instrument and biological fermentation/malodor Pollution process online detection and analysis methods to solve the problem of long-term online detection of multiple fermentation processes or multiple malodorous pollution points, identification of odor types, and online quantitative prediction of qualitative indicators of odor intensity and control indicator values of multiple compound concentrations.
- the electronic nose instrument includes a gas sensor array module I, a capillary gas chromatography column module II, a gas automatic sampling module III, and a computer
- the control and analysis module IV, and the auxiliary gas source V realize the long-term cycle online detection and intelligent analysis of multiple biological fermentation processes or multiple malodor pollution processes.
- Gas sensor array module I includes gas sensor array I-1, gas sensor array annular working chamber I-2, resistance heating element I-3, fan I-4, heat insulation layer I-5 and partition I-6 , Located in the middle right part of the electronic nose instrument.
- Capillary gas chromatography column module II includes capillary gas chromatography column II-1, detector II-2, amplifier II-3, recorder II-4, injection port II-5, resistance heating wire II-6, fan II-7 And insulation layer II-8, located on the upper right of the electronic nose instrument.
- the gas automatic sampling module III includes: the first to the fifth two-position two-way solenoid valve III-1 to III-5, five first purifiers III-6, the first micro vacuum pump III-7, and the first flow meter III -8, the sixth two-position two-way solenoid valve III-9, the first throttle valve III-10, two-position three-way solenoid valve III-11, three-position four-way solenoid valve III-12, the second miniature vacuum pump III- 13.
- the fifth throttle valve III-25 is located at the lower right of the electronic nose instrument.
- Computer control and analysis module IV includes computer motherboard IV-1, A/D data acquisition card IV-2, drive and control circuit board IV-3, 4-channel precision DC power supply IV-4, display IV-5, WIFI module IV-6, located on the left side of the electronic nose instrument.
- a biological fermentation process/fermentation tank or a foul-smelling point is referred to as a detection point for short;
- the gas to be measured at one detection point is sucked into the gas sensor array module I and the capillary gas chromatography column module II by two miniature vacuum pumps III-7 and III-13, respectively.
- the sensor array I-1 and the capillary gas chromatography column II-1 produce sensitive responses, and the electronic nose instrument therefore obtains a group of gas sensor array response curves and a gas chromatogram, which are obtained by the electronic nose instrument sensing a sample of the gas to be measured
- the gas sensor/gas chromatograph analog signal The gas sensor/gas chromatograph analog signal.
- the electronic nose instrument does not pursue the complete separation of chromatogram peaks/peaks.
- the computer control and analysis module IV selects the first 10 maximum chromatographic peaks v gci ( ⁇ ) and the corresponding ones from the semi-separated chromatogram.
- the electronic nose instrument senses a biological fermentation process or a odorous pollution point of the measured gas, and the computer control and analysis module IV will extract the 16 response curves from the gas sensor array I-1
- a sample This is the basis for the qualitative and quantitative analysis of the biological fermentation process or odor pollution process by the electronic nose instrument.
- the electronic nose instrument forms the main body of the odor big data X through the long-term online detection of multiple biological fermentation processes and multiple odor pollution points over many years; the data set X also includes the offline detection data of conventional analytical instruments such as gas chromatography, mass spectrometry, and spectrophotometry.
- the OU value data of odor concentration obtained by professional laboratory sniffing, the data of biological fermentation types such as penicillin, erythromycin, soy sauce, vinegar, cooking wine, monosodium glutamate, etc.
- a subset of data set X establishes the corresponding relationship between gas sensitivity/chromatographic response and multiple biological fermentation processes/odor pollution types and main component concentrations.
- the perceptual components of the smell big data X are subjected to normalization preprocessing, and the machine learning model of the computer control and analysis module IV learns the smell big data X offline to determine its structure and parameters.
- the machine learning model learns the gas-sensitivity-chromatographic response online to fine-tune the model parameters.
- the gas sensor array I-1 and its annular working chamber I-2 are located in a 55 ⁇ 0.1°C thermostat.
- the gas sensor array module I has experienced the initial recovery of the gas sensor array T 0 -120s, the clean air accurate calibration 40s, the equilibrium 5s, the measured gas headspace sampling 60s, and the transition 5s. ⁇ Environmental purification air flushing for 10 seconds, a total of 6 stages.
- the gas types and flow rates of these six stages are as follows: 1Environmental clean air 6,500ml/min, 2Clean air 1,000ml/min, 3No gas flow, 4Test gas 1,000ml/min, 5Environmental clean air 1,000ml/ min, 6Environmental purification air 6,500ml/min; "transition” mainly refers to the conversion from the measured gas to environmental purification air.
- the first flow meter III-8 was finally discharged to the outdoors for 60s; the gas sensor array I-1 therefore had a sensitive response to the measured gas and was stored in a temporary file of the computer control and analysis module IV.
- the time interval [T 0 -120s, T 0 -80s] of the single gas sampling cycle T 0 is the clean air calibration phase of the gas sensor array module I, the three-position four-way solenoid valve III-12 is at position "1", The sixth, seventh and eighth two-way solenoid valves III-9, III-14 and III-15 are all disconnected, and the clean air in the clean air bottle V-2 flows through the first one at a flow rate of 1,000ml/min.
- Environmental clean air refers to the air where the electronic nose instrument is located after dust removal, dehumidification and aseptic treatment. It is only used for the initial recovery of the gas sensor array I-1, the annular working chamber I-2 and related The inner wall of the gas pipeline is flushed, and the accumulated heat of the gas sensor array is taken away.
- the three-position four-way solenoid valve III-12 is at position "2"
- the sixth The two-position two-way solenoid valve III-9 is turned on
- the eighth-two-position two-way solenoid valve III-15 is turned off.
- the environmentally purified air flows through the three-position four-way solenoid valve III-12 at a flow rate of 6,500ml/min.
- the pressure valve III-16, the annular working chamber I-2 and its internal gas sensor array I-1, the sixth two-position two-way solenoid valve III-9, and the first flow meter III-8 are finally discharged to the outdoors, Continue for T 0 -110s.
- the gas sensor array I-1 was initially restored to the baseline state under the action of environmentally purified air; because the eighth two-position two-way solenoid valve III-15 was disconnected, the first to fifth five two-position two-way solenoid valves Whether the solenoid valves III-1 to III-5 are turned on or not does not affect the initial recovery of the gas sensor array I-1.
- the capillary gas chromatographic column module II goes through three stages of headspace sampling of the tested gas for 1 second, chromatographic separation of the tested gas T 0 -16s, and venting and cleaning and purging for 15 seconds; H 2 It doubles as carrier gas and fuel gas, and clean air is used as fuel gas.
- the first 1s of the gas sampling cycle T 0 is the headspace sampling phase of the measured gas of the capillary gas chromatography column module II.
- the time interval [1s, T 0 -10s] of the single cycle T 0 of gas sampling is the measured gas separation phase of the capillary gas chromatography column module II, and the two-position three-way solenoid valve III-11 is in position "2", 72
- the position two-way solenoid valve III-14 is disconnected, and the measured gas from the detection point k is disconnected accordingly, which lasts T 0 -11s.
- the measured gas injected into the injection port II-5 of the chromatographic column module II is driven by the carrier gas H 2 of a certain pressure and flow rate to separate in the capillary gas chromatographic column II-1, and therefore the detector II-2 produces sensing
- the recorder II-4 records the sensory response of the column II-1 time period T 0 -10s in the time interval [0,T 0 -10s], and stores it in the computer for control and analysis In the temporary file of Module IV.
- the last 10s of the gas sampling cycle T 0 is [T 0 -10s, T 0 ] time interval is the venting of the capillary gas chromatographic column II-1, that is, the cleaning and purge stage, the first to the fifth five two-bit two Among the three-way solenoid valves III-1 to III-5, the one that was originally on, that is, III-k, is off, and one of the other four that was originally closed, that is, III-( ⁇ k), is on; the two-position, three-way solenoid valve III -11 is at position "2", the seventh two-position two-way solenoid valve III-14 is turned on, and the eighth second-position two-way solenoid valve III-15 is turned off.
- Two-position solenoid valve III-( ⁇ k), the seventh two-position two-way solenoid valve III-14, and the two-position three-way solenoid valve III-11 are directly discharged to the outdoors. The function of this stage is to remove the odor residues from the kth detection point in the current gas sampling cycle of the relevant pipeline, and gradually replace it with the measured gas at the ( ⁇ k)th detection point for the next gas injection. Prepare for another bio-fermentation process or malodor pollution monitoring point in a single cycle test, lasting 10s.
- the time interval [T 0 -10s, T 0 ] of the single cycle T 0 of gas sampling is the time interval for information selection and analysis at the same time.
- the computer control and analysis module IV is used for gas sampling from the time interval [T 0 -75s, T 0 -15s].
- Select 48 sensing information including steady-state peak value v gsi ( ⁇ ) from the voltage response curve of sensitive sensor array I-1.
- Select 21 perceptual components including the first 10 maximum chromatographic peak values v gci ( ⁇ ) from the chromatogram in the time period [0,T 0 -10s]. This is the basis for the electronic nose instrument to analyze the biological fermentation process or the odor-contaminated area.
- the computer control and analysis module IV performs odor type identification and quantitative prediction of intensity and main concentration index values based on the perception vector x( ⁇ ) and odor big data X.
- the time interval [T 0 -10s, T 0 ] is the information selection and analysis time period of 10s.
- the computer control and analysis module IV compares the gas sensor array module I and the capillary gas chromatograph
- the column module II both perform sensing information selection and analysis and processing operations at the same time.
- the computer control and analysis module IV selects the steady-state peak value v gs_i ( ⁇ ) and the corresponding voltage response curve from each voltage response curve of the gas sensor array I-1 in the time period of [T 0 -75s, T 0 -15s] that is 60s long.
- the peak time t gs_i ( ⁇ ) and the area under the curve Ags _i ( ⁇ ), the three sensing information components, are instantaneously long T 0 -10s from the capillary gas chromatography column II-1 in the time period of [0,T 0 -10s]
- the response component is stored in a temporary file on the computer hard disk.
- the computer control and analysis module IV selects the first q ⁇ 10 from the semi-separated chromatogram
- the maximum chromatographic peak value v gci ( ⁇ ), the corresponding retention time t gci ( ⁇ ) and the area under the chromatogram curve A gc ( ⁇ ), the insufficient chromatographic peak value and the corresponding retention time are filled with zero.
- the last 10s of the gas sampling cycle T 0 is the information processing and analysis time period in the [T 0 -10s, T 0 ] interval.
- the modular machine learning model of the computer control and analysis module IV is based on the gas sensor/chromatographic recent sensing time series matrix X( ⁇ -q) performs odor type identification and quantitative prediction of intensity and main components of biological fermentation process or malodor pollution monitoring point, including: biological fermentation process type and malodor pollution type identification, biological fermentation process cell concentration, substrate concentration, product Quantitative estimation of concentration, quantitative estimation of the concentration of precursor substances in the fermentation process such as n-propanol, phenethyl alcohol, and quantitative prediction of the concentration of 8+1 odor pollutants designated by GB14554; here, ⁇ is the current time, and q is the time that has recently passed , ⁇ -q is the recent time interval.
- Odor big data X also includes: electronic nose instrument gas sensing/chromatographic sensing data of the headspace volatile gas of a variety of single compounds with a concentration of 0.1-1,0000ppm, and offline detection data of conventional analytical instruments such as gas chromatography, mass spectrometry, and spectrophotometry; The professional laboratory sniffs the data.
- the single compound specifically includes n-propanol and phenylacetic acid, the precursors of the biological fermentation process, the eight malodorous compounds specified by GB14554, and the standard reference substance of odor concentration OU specified by European standard EN13725—n-butanol.
- the machine learning model is composed of multiple modular deep convolutional neural networks; the number of single-output deep convolutional neural network modules and the number of main components of fermentation broth predicted in the biological fermentation process, the number of main concentration indicators of odor pollutants, and the number of types of objects to be tested Equal, one-to-one correspondence.
- a single-output deep convolutional neural network consists of an input layer, 3 convolutional layers, 2 downsampling layers and 1 output unit.
- each hidden layer and output layer are all Sigmoid modified activation functions
- each single-output deep convolutional neural network adopts an error back propagation offline layer-by-layer learning algorithm, which mainly learns the labeled data in the odor big data and the odor big data with known components and has the necessary intelligence; convolutional layer
- the scan window size is 5 ⁇ 5, and the overlap scan step is 1; the convolution kernel is a combination of sine, cosine, polynomial, Gaussian, Sigmoid, wavelet, and Laplace kernels.
- the scan window size of the down-sampling layer is 2 ⁇ 2, and the step size is 2 for non-overlapping scans, and the maximum, mean and mean square error features are extracted.
- n single-output deep convolutional neural network models perform odor type recognition based on the current time ⁇ of the gas/gas chromatography and the recently occurred time series perception matrix X( ⁇ -q), and estimate and predict the current time ⁇ one by one. And future ⁇ +1, ⁇ +2, ⁇ +3 odor intensity and main component concentration values.
- the electronic nose instrument performs long-term cycle online detection and online analysis and prediction of multiple biological fermentation processes/malodor pollution points, including the following steps:
- the three-position four-way solenoid valve III-12 is in position "2", the sixth and second two-way solenoid valve III-9 is turned on, and the eighth-second two-way solenoid valve III-15 is turned off; at the first miniature vacuum pump III- Under the suction effect of 7, the environmentally purified air flows through the three-position four-way solenoid valve III-12, the pressure regulator valve III-16, the annular working chamber I-2 and its gas sensor array I in sequence at a flow rate of 6,500 ml/min. -1.
- the sixth two-position two-way solenoid valve III-9 and the first flow meter III-8 are finally discharged to the outdoors, and the internal temperature of the annular working chamber I-1 of the gas sensor array reaches a constant 55 ⁇ 0.1°C.
- the two-position three-way solenoid valve III-11 is at position "2"
- the seventh two-position two-way solenoid valve III-14 is disconnected, and the capillary gas chromatographic column II-1 gradually returns to the baseline under the push of the carrier gas H 2 State, the internal temperature of the chromatographic column thermostat reaches a constant 250 ⁇ 0.1°C,
- Gas sensor array module I It goes through six gas sampling stages in sequence: 1360s initial recovery, 240s precise calibration, 35s balance, 460s headspace sampling, 55s transition, and 610s cleaning and initial recovery. .
- the gas sensor array I-1 therefore generates a sensitive response which is stored in a temporary file corresponding to the computer control and analysis module IV.
- the computer control and analysis module IV is from the single gas sensor of the gas sensor array module I in the [405s,465s] time period of 60s
- Select three sensing information: steady-state peak value v gs_i ( ⁇ ), peak time t gs_i ( ⁇ ), and total curve area Ags _i ( ⁇ ) from a single voltage response curve; an array I-1 composed of 16 gas sensors A total of 16*3 48 perceptual components are obtained.
- the computer control and analysis module IV selects the top 10 maximum chromatographic peak values v gc_i ( ⁇ ) and the corresponding retention time t gc_i ( ⁇ ) and the total chromatogram curve from the 470s semi-separated chromatogram of the capillary column module II. Under the area A gc ( ⁇ ), a total of 21 perceptual components are obtained.
- the computer control and analysis module IV obtains a 69-dimensional sensing vector x( ⁇ ) ⁇ R 69 from the sensing information of the gas sensor array I module and the capillary column module II; then, machine learning Based on the perception vector x( ⁇ ) and odor big data X, the model recognizes odor types and quantitatively predicts the intensity and main components.
- the monitor displays the monitoring and prediction results, and transmits them to the central control room and multiple fixed/mobile terminals through the Internet.
- One of the fifth two-position two-way solenoid valve is turned on.
- the electronic nose instrument realizes the cyclic online detection and identification of the measured gas at 1 to 5 detection points, as well as the quantitative prediction of odor intensity and multiple concentration index values.
- Figure 1 shows the present invention—a gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online detection and analysis method—gas-sensing/chromatographic multi-sensing information fusion, electronic nose instrument development, and online detection and analysis of fermentation tail gas and malodorous odor Schematic diagram of the technical route of multi-process parameter analysis.
- Fig. 2 is a schematic diagram of the working principle of the present invention—a gas-sensing-gas phase electronic nose instrument and a fermentation-malodor multi-state parameter online detection and analysis method—the electronic nose instrument.
- Fig. 3 is a schematic diagram of the present invention—a gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online detection and analysis method—gas-sensing sensor array module and its gas circuit working principle.
- Fig. 4 is a schematic diagram of the present invention—a gas-sensing—gas phase electronic nose instrument and a fermentation—odor multi-state parameter online detection and analysis method—capillary gas chromatography column module and its gas path working principle.
- Fig. 5 is a schematic diagram of the present invention—a gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online detection and analysis method—gas-sensing sensor array module and capillary gas chromatography column module.
- Fig. 10 is a schematic diagram of the present invention—a gas-sensing-gas-phase electronic nose instrument and fermentation-malodor multi-state parameter online detection and analysis method—modular deep convolutional neural network model multi-parameter "divide-and-conquer” online quantitative prediction schematic diagram.
- Figure 1 is a schematic diagram of the technology route of the present invention's gas sensor array and capillary gas chromatography column sensing information fusion, electronic nose instrument development, fermentation tail gas and malodorous multi-source online detection and multi-process parameter analysis.
- the technical route shown in Figure 1 includes: (1) Performance evaluation and selection of gas-sensing and chromatographic sensing components. In-depth analysis of the difference in characteristics between the gas sensor and the capillary gas chromatography column, and is committed to complementing each other's strengths. (2) Modularization of components such as gas sensor array. The gas sensor array, capillary gas chromatography column, automatic gas sampling, computer control and analysis and other important components realize structural modularization. (3) The gas-sensitivity/chromatographic online perception model and information fusion.
- the machine learning model learns odor big data X offline to optimize and determine the model structure and parameters; in the decision-making stage, the machine learning model online learns the gas sensor/chromatogram recent response to fine-tune the parameters, and based on the gas sensor/gas current sensing vector x( ⁇ ) Determine the biological fermentation process or the type of odor pollution online, quantitatively predict the concentration of the main components of the fermentation broth during the biological fermentation process or the concentration of the 8 components of the odor pollutant specified by the national standard GB14554 and the OU value of the odor concentration, to achieve multiple biological fermentation processes and multiple odor pollution Long-term cycle online detection and online analysis of complex odors.
- Fig. 2 is a schematic diagram of the working principle of the gas sensor/gas chromatography integrated electronic nose instrument of the present invention.
- the electronic nose instrument mainly includes: gas sensor array module I, capillary gas chromatography column module II, gas automatic sampling module III, computer control and analysis module IV, and hydrogen cylinder V and clean air cylinder VI.
- Hydrogen is also used as the carrier gas and fuel gas of the capillary gas chromatography column module II hydrogen flame ionization detector FID; on the one hand, clean air is used as the supporting gas of the capillary column module II, and on the other hand, it is used as the calibration gas of the gas sensor array module I (not combustion).
- FIG. 3 and 4 are schematic diagrams of the gas sensor array module I, the capillary gas chromatography column module II and the gas path of the electronic nose instrument of the present invention.
- the main components of the gas sensor array module I include: gas sensor array I-1, gas sensor array annular working chamber I-2, resistance heating element I-3, fan I-4, heat insulation layer I-5 and insulation Board I-6, located in the middle right part of the electronic nose instrument.
- the main components of the capillary gas chromatography column module II include: capillary gas chromatography column II-1, detector II-2, amplifier II-3, recorder II-4, injection port II-5, resistance heating wire II-6, Fan II-7 and insulation layer II-8 are located at the upper right of the electronic nose instrument.
- the function of the gas sensor array module I and the capillary gas chromatography column module II is to convert the chemical and physical information of odors into electrical signals online.
- the components related to the gas automatic sampling module III and the gas sensor array module I include: the first to the fifth two-position two-way solenoid valve III-1 to III-5, the first purifier III-6, and the first miniature vacuum pump III-7, the first flow meter III-8, the sixth two-position two-way solenoid valve III-9, the first throttle valve III-10, the three-position four-way solenoid valve III-12, the seventh two-position two-way solenoid valve Valve III-14, the eighth two-position two-way solenoid valve III-15, the stabilizing valve III-16, the first pressure reducing valve III-17, the second throttle valve III-18, and the second purifier III-19.
- the components related to the gas automatic sampling module III and the capillary gas chromatography column module II include: two-position three-way solenoid valve III-11, second miniature vacuum pump III-13, second pressure reducing valve III-20, and third purifier III-21, the third throttle valve III-22, the second flow meter III-23, the fourth throttle valve III-24, and the fifth throttle valve III-25.
- the gas automatic injection module III is located at the bottom right of the electronic nose instrument.
- the main components of the computer control and analysis module IV include: computer motherboard IV-1, A/D data acquisition card IV-2, drive and control circuit board IV-3, 4-channel precision DC power supply IV-4, display IV- 5.
- WIFI module IV-6 located on the left side of the electronic nose instrument. The function of the WIFI module IV-6 is to transmit the sensing information of the gas sensor array module I and the capillary gas chromatography column module II to the designated fixed/mobile terminal in real time.
- Figure 5 is a schematic diagram of the gas sensor array module I and the capillary gas chromatography column module II of the electronic nose instrument.
- the gas sensor array and the capillary gas chromatography column are both located in two constant temperature boxes with different constant temperatures, forming two modules, which can be easily replaced as needed.
- the adjustable time period is mainly the environmental purification air flushing of the gas sensor array module I/the initial recovery phase of the gas sensor and the separation phase of the capillary gas chromatography column module II.
- Figure 6(a) shows the single cycle of gas sampling of the capillary column module II, including three stages: 1 headspace sampling of the tested gas, 2 chromatographic separation of the tested gas, and 3 column venting.
- 1 The headspace sampling phase of the tested gas is at the beginning of a single injection cycle T 0 , the sampling time range is 0.5 ⁇ 1.5s, and the default is 1s; the sampling flow range is 1.5 ⁇ 15ml/min, and the default is 6ml/min; cumulative The injection volume ranges from 0.0125 to 0.375ml, and the default is 0.1ml.
- the two-position three-way solenoid valve III-11 is at position "1"
- the seventh two-position two-way solenoid valve III-14 is turned on
- the first to fifth two-position two-way solenoid valves are turned on.
- One of solenoid valves III-1 to III-5 is turned on
- the eighth two-position two-way solenoid valve III-15 is turned off.
- the first two-position two-way solenoid valve III-1 Assuming that the first two-position two-way solenoid valve III-1 is turned on, at this time, one of the biological fermentation process (fermentation tank) or the measured gas at the odor pollution point, such as the first detection point, is at the second micro vacuum pump III-13 Under the suction effect, it flows through the first two-position two-way solenoid valve III-1, the seventh two-position two-way solenoid valve III-14, the two-position three-way solenoid valve III-11, and the fourth throttle valve III- 24. It is mixed with carrier gas H 2 at the injection port II-5, so it flows into the capillary gas chromatography column II-1.
- the tested gas injection volume is 0.1ml, which meets the optimal injection volume requirements for capillary gas chromatography columns.
- the two-position three-way solenoid valve III-11 is at position "2"
- the seventh two-position two-way solenoid valve III-14 is disconnected, that is, the measured gas is disconnected.
- the gas to be measured is separated in the capillary gas chromatography column II-1.
- the two-position three-way solenoid valve III-11 is at position "2"
- the seventh two-position two-way solenoid valve III-14 Conducted, one of the first to fifth five two-position two-way solenoid valves III-1 to III-5 is turned on (but the original conduction is closed), and the eighth two-position two-way solenoid valve III-15 is disconnected .
- the first two-position two-way solenoid valve III-1 Assuming that the first two-position two-way solenoid valve III-1 is turned on, under the suction action of the second miniature vacuum pump III-13, it flows through the first two-position two-way solenoid valve III-1 and the seventh two-way solenoid valve in turn The solenoid valve III-14 and the two-position three-way solenoid valve III-11 are then discharged to the outdoors.
- the function of this stage is to remove the residues of related pipes in this single gas sampling cycle, and prepare for the next single gas sampling cycle. It must be pointed out that the position of the three-position four-way solenoid valve III-12 is determined by the following table 2.
- Table 2 shows the operating parameters of the gas sensor array module I and the on/off status of the relevant solenoid valve in a single gas sampling cycle T 0.
- one of the first to fifth five two-position two-way solenoid valves III-1 to III-5 leads The three-position four-way solenoid valve III-12 is in position "0", the sixth and seventh two-position two-way solenoid valves III-9 are disconnected from III-14, and the eighth two-position two-way solenoid valve III-15 is on. Pass. Under the suction action of the first micro vacuum pump III-7, the measured gas from one of the 5 biological fermentation processes (fermentation tanks) or malodorous pollution points (such as the first detection point) flows through at a flow rate of 1,000ml/min.
- the three-position four-way solenoid valve III-12 is at position "1"
- the sixth, seventh and eighth two-way solenoid valve Valves III-9, III-14 and III-15 are disconnected, and the clean air in the clean air bottle VI is 1,000ml/min
- the flow rate flows through the first pressure reducing valve III-17, the second throttle valve III-18, the second purifier III-19, the three-position four-way solenoid valve III-12, the pressure regulator valve III-16, and the ring operation
- the cavity I-2 and its gas sensor array I-1, the first throttle valve III-10, and the first flow meter III-8 are finally discharged to the outdoors for 40s. During this period, the gas sensor array I-1 is accurately restored to the reference state under the action of clean air. Since the eighth two-position two-way solenoid valve III-15 is disconnected, the conduction/disconnection of the first to fifth two-position two-way solenoid valves III-1 to III-5 does not affect the gas sensor array. Calibration of I-1.
- the three-position four-way solenoid valve III-12 is in position "2""
- the sixth two-position two-way solenoid valve III-9 is on
- the eighth two-position two-way solenoid valve III-15 is off
- the environmentally purified air flows through the three-position four-way solenoid valve III at a flow rate of 6,500ml/min.
- “environmentally purified air” refers to the air where the electronic nose instrument is located after dust removal, dehumidification and aseptic treatment. It is only used for the initial recovery of the gas sensor array I-1.
- the gas sensor array is ring-shaped. Flushing of the inner wall of the working chamber I-2 and related gas pipelines, and taking away the accumulated heat of the gas sensor array.
- the gas sensor array module I and the capillary gas chromatography column module II enter the information selection and analysis area at the same time.
- Computer control and analysis module IV from [T 0 -75s, T 0 -15s ]
- Each gas sensor voltage response curve steady-state period selected peak v gsi ( ⁇ ), corresponding to the peak time t gsi ( ⁇ ), the area under the curve Ags i ( ⁇ ) these three perception information; select the first 10 maximum chromatographic peaks v gci ( ⁇ ) and the corresponding 10 retention from the chromatogram of the time period [0,T 0 -10s] Time t gci ( ⁇ ), area under the chromatogram curve A gc ( ⁇ ), a total of 21 chromatographic sensing components.
- the figure shows the corresponding curve examples of the three gas sensors TGS822, TGS826 and TGS832 respectively for petroleum wax samples, 2,000ppm ethylene gas and 5,000ppm ethanol volatilization gas.
- Our approach is to add zero for the insufficient number of chromatographic peaks and corresponding retention times.
- the semi-separated chromatogram in Figure 9(b) has more than 10 chromatographic peaks, and we can select the first 10 largest chromatographic peaks.
- the present invention regards the semi-separated chromatogram as a part of the electronic nose instrument's perception information, which is the pattern, combines the perception information of the gas sensor array to establish odor big data, and realizes unknown odor identification, qualitative analysis and quantitative prediction of main components with the help of artificial intelligence machine learning methods .
- Figure 10 is a schematic diagram of multi-parameter "divide and conquer" quantitative prediction of a deep convolutional neural network machine learning model oriented to "continuous online” analysis.
- the specific steps include: according to the time series matrix X( ⁇ -q) obtained by the gas sensor array module I and the capillary gas chromatography column module II recently, a plurality of single-output deep convolutional neural networks are used to predict the types of fermentation and odor pollution one by one , Odor intensity and main component concentration value.
- ⁇ is the current time
- q is the time that has recently passed
- ⁇ -q is the recent time interval. Therefore, the dimension scale of the time series matrix X( ⁇ -q) is R 69 ⁇ ( ⁇ - q+1) .
- the value of q is generally about 6 hours in the recent period of fermentation or malodor pollution.
- the first task is to establish big odor data, including: gas sensor array module I and capillary gas chromatography column module II online sensing data for a large number of biological fermentation processes and odor-contaminated areas over the years ; Off-line monitoring data of conventional instruments such as chromatographs, mass spectrometers, and spectrophotometers; odor label data of known types and components; and sensory evaluation data.
- the next thing to do is the fusion of the sensing data of the gas sensor array and the sensing data of the capillary gas chromatography column, including normalization and dimensionality reduction preprocessing.
- one-by-one quantitative prediction problem that is, one-by-one fitting problem of n curves/surfaces is decomposed into n curves/surfaces one by one.
- the invention adopts multiple modular single-output deep convolutional neural networks to realize multi-parameter online quantitative prediction.
- a single-output deep convolutional neural network consists of an input layer, 3 convolutional layers, 2 down-sampling layers and 1 output unit. It mainly learns the labeled data and the data with known components in the odor big data.
- the activation functions of each hidden layer and output layer are modified Sigmoid activation functions Using error back propagation offline learning algorithm layer by layer.
- the scan window size of the convolutional layer can be 5 ⁇ 5, and the overlap scan step length can be 1; the convolution kernel is a combination of sine, cosine, polynomial, Gaussian, Sigmoid, wavelet, and Laplace kernels; down-sampling layer
- the size of the scanning window can be 2 ⁇ 2, and the non-overlapping scanning means that the step size is 2.
- the maximum, mean and mean square error features are extracted from each scanning window.
- n single-output deep convolutional neural network models predict the imminent moments of ⁇ +1, ⁇ +2, ⁇ +3, etc. one by one based on the gas sensor/gas chromatography recent sensing time series matrix X( ⁇ -q) Multiple quantitative index values, including odor type, intensity and concentration of main components.
- the gas chromatography time series perception response matrix is equivalent to predicting possible changes in the odor intensity and main components of the next 8 minutes, 16 minutes and 24 minutes based on the fermentation process or the changes in the malodorous environment in the past 1.2 hours.
Landscapes
- Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Chemical & Material Sciences (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Pathology (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- Immunology (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Artificial Intelligence (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Evolutionary Computation (AREA)
- Mathematical Physics (AREA)
- General Engineering & Computer Science (AREA)
- Medicinal Chemistry (AREA)
- Food Science & Technology (AREA)
- Combustion & Propulsion (AREA)
- Sampling And Sample Adjustment (AREA)
- Investigating Or Analyzing Materials By The Use Of Fluid Adsorption Or Reactions (AREA)
Abstract
一种气敏-气相电子鼻仪器和发酵-恶臭多状态参数在线分析方法，仪器组成单元包括气敏传感器阵列、毛细管气相色谱柱、气体自动进样、计算机控制与分析诸模块。气体进样单周期T 0＝300-600s，气敏和色谱二模块气体进样流量与进样体积不相等，进样时间不同步；计算机控制与分析模块从时长60s单条气敏响应曲线中选择稳态峰值、出峰时间、总曲线下面积这3个感知信息，从时长T 0-10s半分离色谱图中选择10个最大峰值与保留时间、总曲线下面积共21个感知信息；模块化深度卷积神经网络依据归一融合的69维实时感知信息和气味大数据实现最大T＝5T 0的5个发酵或恶臭污染过程循环在线识别、气味强度与多指标量化估计。
Description
本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，面向以生物发酵、恶臭污染为代表的过程多源在线检测与多参数分析需求，涉及人工智能、计算机、生物工程、环境保护、分析化学等技术领域，主要解决气敏传感器灵敏度较低、单一感知信息选择法致使气敏传感器阵列选择性差、峰峰完全分离做法致使色谱法在线性差、气敏传感器阵列与气相色谱柱结构优化组合、多源感知信息选择与融合等一系列问题，以实现电子鼻仪器对生物发酵、恶臭污染过程长期循环多源在线检测与多参数在线分析的目标。
嗅觉模拟—电子鼻方法用多个性能重叠的气敏元件组成阵列实现气味快速检测，用机器学习方法进行气味定性定量分析。气味在线检测与分析技术已成为生物工程、环境保护、食品等行业的核心应用技术，电子鼻仪器因具有速度快、非接触、操作简便等特点而备受关注。电子鼻理论与技术研究现状是，气敏传感器灵敏度最高已达10
-7(V/V)即0.1ppm数量级，但选择性差，导致电子鼻仪器稳定性、在线性和定性定量能力差；发酵电子鼻等产品更属于国际空白。在重大需求背景下，电子鼻技术多次列入科技部“863”、科技支撑和重点研发计划。
生物发酵和恶臭污染过程多参数实时(real-time)估计、预测与控制的前提是多个过程参数在线(online)检测。环境恶臭污染时间跨度经年累月；生物发酵过程短的1-2天，长的数十天(例如啤酒发酵)、数月乃至数年。生物发酵和恶臭污染过程状态变化用“瞬息万变”形容有点夸张，但是，以“小时”为单位的检测与分析周期肯定太长。我们认为，被测对象状态1min内一般不会发生很大变化，即检测周期小于1min并不必要；反过来，生物发酵或恶臭污染状态1小时内完全可能发生很大变化，将周期1小时及以上的定时人工取样“断续”检测看成“在线”检测是不合适的。据此，电子鼻仪器对单一生物发酵过程(发酵罐)或单一恶臭污染监测点的在线检测与分析周期不宜超过T
0＝10min，对多个发酵罐或多个恶臭监测点的循环在线检测与分析周期T＝n*T
0不宜超过1小时，以此判断一种检测与分析方法是否“在线”是比较合理的。
靠人工嗅闻尾气以分析生物发酵过程是不现实且不可想象的。不仅如此，靠嗅闻量化确定臭气浓度、食品与香料香精气味强度等指标的做法因为过程十分繁琐、成本高、效率低、客观性差、可操作性差而倍受诟病。长期嗅闻恶臭气味会对身体产生严重伤害，与人们追求美好生活的愿望和所处的人工智能时代格格不入。因此，复杂气味多源在线检测、识别和多种组成成分同时在线量化预测既是复杂的理论问题，更是迫切需要解决的技术与应用问题。
电子鼻技术主要发展趋势之一是，以多个具有必要灵敏度的气敏器件组成阵列，着重利用大数据和人工智能技术来提高对复杂气味的定性定量能力，进而实现气味类型识别和强度与多项主要成分量化估计。“长期连续在线检测与分析”是电子鼻的一种主要工作方式，主要面向生物发酵、恶臭污染等对象的过程连续在线检测、在线定性分析与多项主要成分浓度在线预测，其特点是被测气体来源充沛，气体进样单周期固定(例如5min)，气体抽取流量与抽取持续时间固定，周而复始，电子鼻仪器靠对发酵尾气/恶臭气体的一次次感知进行定性定量分析。
电子鼻仪器“连续在线”检测与分析工作的基本前提是其核心—气敏传感器阵列对被测对象具有显著感知能力。从应用角度出发，气敏传感器应达到的性能指标包括：灵敏度足够高(ppm级以上)，响应速度足够快(1min以内)，工作状态稳定，商品化程度高，寿命长(3-5年)，自身尺寸小，选择性较好。
依据敏感材料和工作原理的不同，文献[1]列出了金属氧化物半导体(metal oxide semi-conductor,MOS)型，电化学(electrochemical,EC)型，导电聚合物(conducting polymer,CP)型，石英微平衡(quartz microbalance,QMB)型，声表面波(surface acoustic wave,SAW)型、光离子(photo ionization detector,PID)型这6种常用气敏元件的感知性能。与MOS型相比，EC型气敏传感器选择性要好一些，但尺寸大许多，寿命至少短1年以上，灵敏度低一个数量级或以上。同样与MOS型相比，PID型传感器不仅尺寸大、感知范围窄、价格高，而且寿命仅半年左右。不仅如此，EC型与PID型气敏传感器仅适用于恶臭污染物检测。QMB和SAW型气敏元件的灵敏度比MOS型低1个数量级以上，敏感膜材料有待进一步开发，尺寸有待进一步缩小。综合考虑各种因素，以SnO
2为代表的MOS型气敏传感器最适宜用做电子鼻仪器的感知元件。
必须指出，上述6种敏感材料制成的单一型气敏传感器及其阵列的感知能力十分有限，不满足生物发酵、恶臭污染等过程对象的在线检测要求。我们的大量实验指出：①即使灵敏度最高的MOS气敏元件对青霉素发酵前体—苯乙酸仍不够敏感；②现有电子鼻对某一养猪场的臭味都不敏感。前已述及，气敏传感器灵敏度最高已达10
-7(V/V)数量级，但这只是特定MOS型传感器对特定气味成分而言的，不是普遍现象。最典型的例子是电子鼻仪器对恶臭污染物检测与主要恶臭化合物浓度指标预测。GB14554指定的具体指标包括氨NH
3、硫化氢H
2S、二硫化碳CS
2、三甲胺C
3H
9N、甲硫醇CH
4S、甲硫醚C
2H
6S、二甲二硫醚C
2H
6S
2、苯乙烯C
8H
8共8种具体化合物浓度指标值，加上臭气浓度OU(odor unit)值，简称8+1种恶臭污染物浓度控制指标值。现在，对CS
2、C
3H
9N、CH
4S、C
2H
6S、C
2H
6S
2、C
8H
8这6种恶臭有机化合物均敏感且选择性好的气敏传感器阵列并不存在，短期内尚难以研制出来。这就是说，仅靠上述6种气敏材料制成传感器阵列实现8+1种恶臭污染指标值的在线检测与预测是难以办到的。
大量冗余气敏元件组成阵列来检测众多气味之路走不通；一方面是仪器结构十分复杂，另一方面是气敏传感器灵敏度不够、重叠感知范围有限[1]，气相色谱法因此引起了人们的高度关注。色谱电子鼻商品已经出现，例如法国αMOS公司的HeraclesII气相电子鼻。本质上，HeraclesII电子鼻是基于色谱峰峰完全分离、单进样周期T
0＝5-8min的气体一次性现场随时检测与分析，不适用于长期连续在线检测。
气相色谱法选择性好，MOS气敏传感器选择性差。但是，这种差别只是相对的，气相色谱法对未知样品的“定性能力”仍然是“弱”的。也就是说，在无内/外标样品谱图的情况下，仅凭一次测量得到的谱图根本无法确定未知样品的成分及其组成。气相色谱法缺陷之二是，色谱柱“选择能力”没有普遍性。只有在特定条件下，特定色谱柱才对特定样品敏感，即特定色谱柱只能检测特定范围的特定样品。当进样条件、测试条件或色谱柱自身参数发生变化时，特定样品的色谱感知参数随之变化。
必须指出，气相色谱法的核心是分离，而不是检测。提高色谱分离度的有效方法包括：(1),适当增加柱长度；(2),适当减少进样量和进样时间；(3),适当降低载气流速；(4),适当降低色谱柱温度；(5),适当提高汽化室温度。我们必须清楚，适当升高色谱柱温和/或适当增大载气流速有利于缩短保留时间，提高色谱峰峰分离度和缩短保留时间二者有时是互相矛盾的。
气味是数十、数百、乃至数千种化合物混合体，且所有组成成分分子量均小于300Dalton。保留时间是色谱法的重要定性分析参数，而GB14554指定的8种恶臭化合物色谱保留时间大多小于8min。为提高气相色谱法的检测速度，我们可选择较大内径的毛细管柱，例如φ0.53mm，柱长可为30m，设计制作GC恒温工作室；氢气兼作载气和燃气，程序升温、被测气体进样和载气推送过程均精密控制；毛细管柱及整个模块应便捷更换和安装。在T
0≤10min周期内，被测气体进样流量可为1.0-15ml/min，进样时长可为0.5～1.5sec。这时，我们得到的可能是一幅T
0≤10min有限时长的半分离多峰图。
色谱图的半分离/未完全分离现象是由被测气体成分、色谱柱自身特性、色谱仪工作参数设置、检测器性能、记录仪记录时间诸多因素共同作用的结果。色谱峰峰未完全或半分离是普遍现象，完全分离只是理想或极限情况。被测对象组分越多，峰峰完全分离就越困难，且以检测时间长为代价。开赛一段时间的马拉松比赛，尽管冠亚军未产生，但胜负趋势已分，冠亚军就在“跑在比赛队伍前面的团队”中。这是气相色谱法利用半分离色谱图进行在线检测与分析的生物学依据。半分离色谱图是全分离色谱图的一部分，相当于是马拉松比赛“跑在比赛队伍前面的团队”。只要被测样品成分和色谱柱测试条件保持不变，同一 样品不同时间测试得到的半分离色谱图就保持不变，半分离色谱图与全分离色谱图的位置关系也保持不变。这就是说，我们可以用半分离色谱图推测全分离色谱图的一些主要特性，例如，推测未在半分离色谱图上出现的一些保留时间长的成分存在与否与含量。对生物发酵、恶臭污染等过程分析而言，我们只要得到体现主要状态参数的信息就够了，半分离色谱图实际已蕴含了全分离色谱图的主要信息，关键是怎样从图中得到需要的信息，并进行解释。
从感知范围来说，单一色谱柱和单一型敏感材料的气敏传感器阵列都是有限的。气相色谱法对无机物和易分解的高沸点有机物分析起来比较困难，对未知物定性比较困难，不适用于分析一些极性强的单一化合物或极性差别大的复杂化合物，以及一些不含碳的化合物。例如，采用氢火焰离子化检测器(Hydrogen flame ionization detector,FID)的气相色谱法就无法有效检测无机化合物。这就是本发明提出融合气敏传感器阵列与毛细管气相色谱柱的电子鼻仪器在线检测与分析方法的动因。为什么要将气敏传感器阵列与毛细管气相色谱柱二者融合起来？原因之一是，气敏传感器不仅选择性较差，而且对有些化合物的灵敏度差强人意。例如，对一些非还原/氧化性无机化合物、对青霉素发酵前体—苯乙酸等，仅靠现有气敏传感器阵列实现GB14554指定的8+1种恶臭污染指标在线量化预测尚不可能。原因之二是，气相色谱法在线性差，单一色谱柱选择能力有限。例如，气相色谱法只能检测热稳定性好的样品；据不完全统计，Agilent公司提供上千种即用色谱柱。“色谱柱选择与更换操作”这一事实说明，单一色谱柱的检测范围是有限的。
一个典型例子是，GB14554规定：NH
3和CS
2这2种恶臭污染物浓度用分光光度法检测，H
2S、C
3H
9N、CH
4S、C
2H
6S、C
2H
6S
2、C
8H
8等6种恶臭污染物浓度用气相色谱法检测。值得关注的是，GB/T14676-14678这3项国家标准分别指定了后6种恶臭污染物的气相色谱检测方法，其中的检测器、色谱柱和工作条件竟互不相同。我们发现，这几项国标指定用2种不同尺寸的填充柱分别测量其中的6种恶臭化合物。这就是说，单一色谱柱不能同时检测GB14554指定的6种恶臭化合物。总之，选择色谱柱须考虑色谱柱自身材料、固定相、内径、膜厚、柱长度、被测样品极性与非极性等诸多因素。
单一型气敏传感器阵列选择性较差，重叠感知范围有限，灵敏度亦不够，不满足生物发酵、恶臭污染等对象的在线检测要求。色谱法的优点是灵敏度高、选择性较好，缺点是分离时间即检测周期长，仪器结构复杂，工作条件苛刻，现有用法根本不适用于气味在线检测。在无内/外标样品谱图的情况下，仅凭一次测量的谱图根本无法确定未知样品的成分与组成。气相色谱法缺陷之二是，色谱柱“选择能力”没有普遍性。只有在特定条件下，特定色谱柱才对特定样品敏感，即特定色谱柱只能检测特定范围的特定样品。当进样条件、测试条件、色谱柱自身参数其中之一变化时，特定样品的色谱感知参数随之变化。气相色谱法缺陷之三是，实现多组分色谱峰“完全分离”是困难的，乃至于不可能的。组分越多，组分间极性越相近，保留时间越相近，峰峰完全分离就越困难。我们认为，色谱图多组分谱峰完全分离是相对的、很少的；反之，多组分色谱峰不完全分离是绝对的、普遍的。从操作参数角度上讲，提高色谱分离度和缩短保留时间二者有时是互相矛盾的。
气敏传感器优点是响应速度快，工作条件要求低，缺点是选择性较差，灵敏度不够理想。GC法的优点是灵敏度高、选择性较好，缺点是分离时间即检测周期长，仪器结构复杂，工作条件苛刻，现有用法完全不适用于长期在线检测。必须指出，“GC柱选择性较好，MOS气敏传感器选择性较差”这种差别只是相对的，气相色谱法对未知样品的“定性能力”仍然是“弱”的。气敏传感器阵列与毛细管气相色谱柱二者形成鲜明对照，二者的融合可达到取长补短的功效，相得益彰。为实现对发酵过程和恶臭污染物较宽范围的在线感知，需要解决的问题是，如何将气敏传感器阵列与色谱柱组合起来，优势互补，实现单周期为5-10min左右的长期循环在线检测。为了实现气敏传感器阵列—毛细管气相色谱柱相融合的电子鼻仪器在线检测与分析方法，我们须解决以下气味感知理论与分析技术问题：
(A)气敏—色谱多感知信息选择与融合和电子鼻仪器在线感知能力问题
气味的特点是，(1)组成成分众多且时刻变化。以恶臭污染物为例，呈味成分除H
2S、NH
3、SO
2等少数无机物外，大多数为有机物质，即“挥发性有机化合物(Volatile Organic Compounds,VOCs)”。(2)有些成 分嗅觉阈值很低，但对气味强度贡献度却很大；反之亦然。电子鼻在实际应用中遇到的一个困境是，有些成分对气味强度贡献度很小，气敏传感器却很敏感；反之亦然。气敏传感器用作气味在线检测是，应具有的性能指标包括：灵敏度足够高，响应速度足够快，工作状态稳定，商品化程度高，寿命长，自身尺寸小，选择性较好。我们应深入理解不同气敏元件的特点，设计小型气敏传感器阵列模块，有效解决稳定性差、噪声消除、温湿度补偿、便捷更换等问题。
大量实验指出，以S
nO
2材料为代表的MOS气敏传感器对有些气味响应速度很快，例如对乙醇挥发气只需2s即可；而对另一些气味则响应速度很慢，甚至达60s或更长，例如对GB/T14675指定的一种标准臭液—γ-十一碳(烷)酸内酯C
11H
20O
2挥发气感知就是如此。这一现象告诉我们，尽管同一气敏传感器对两种气味的响应曲线稳态最大值可能相同，但出峰时间与曲线下面积可能不同；或者曲线下面积可能相同，但稳态最大值与出峰时间可能不同，等等。总之，气敏传感器响应曲线形状与气味组成有关，涉及分子量、碳数、极性、官能团诸多因素。
三角形稳定性是指三条边(直线)首尾相接形成稳定结构，具有受力不变形的特点。平行四边形受力易变形，是不稳定的；类似地，边数大于3的多边形都是不稳定的。三角形稳定性原理给我们的启发是，仅知道其中两个参数(2条边长、2个夹角、1条边长与1个夹角共三种情况)是无法确定一个三角形结构的；无须赘言，只知道一个参数(1条边长、1个夹角共两种情况)更是不行。
受三角稳定性原理启发，我们应从单条气敏传感器响应曲线中同时提取多个特征信息，例如，同时选择“稳态”响应最大值、出峰时间值和曲线下面积等特征信息，相当于从预处理角度提高电子鼻仪器的选择性。色谱柱响应速度比气敏传感器至少低一个数量级，企求色谱峰/峰完全分离的做法致使色谱法不满足气味在线检测要求。受马拉松比赛生活原型启发，可以从规定时间区间(例如T
0＝10min左右)的半分离色谱图上提取前若干个最高峰值与对应保留时间，再加上谱图曲线下面积，作为毛细色谱柱对发酵对象或恶臭污染物的感知信息特征，以提高气相色谱法的响应速度即在线性。
如何从气敏传感器阵列响应曲线和半分离色谱图中同时选择多个特征信息并融合起来，以提高电子鼻仪器的在线定性定量分析能力，是本发明要解决的一个主要问题。
(B)气敏传感器阵列等功能部件优化组合和电子鼻仪器集成化与自动化问题
气味组成成分众多，环境变化多端，企图用冗余气敏元件组成阵列来检测所有气味是不经济的，甚至是不现实的。我们前已指出，单一色谱柱和单一型气敏传感器阵列的感知范围都是有限的。因此，我们应发明气敏传感器阵列和气相色谱柱优化与融合方法，将气味感知、气体自动进样***、驱动与控制电路、计算机等模块化并集成在一个测试箱内，研制尺寸小、重量轻、操作简便的多点集中式电子鼻仪器；精密控制仪器内部各部件的工作状态，优化仪器内部的工作条件，以内部“不变”应对外部“万变”。理想情况是，一台电子鼻仪器能对多个发酵罐或在特定区域多个恶臭污染观测点以年月为单位的每天24小时同时在线检测，即可固定点检测，也可移动点检测；利用简单有效的机器学习模型与算法实现对气味强度、主要成分浓度的实时在线分析和预测，并利用WIFI技术，实时把检测数据和分析结果通过云端传输到监控中心及各种终端，实现基于Internet网的特定区域远程监控。
(C)基于大数据和机器学习的电子鼻仪器在线分析能力与智能化问题
人类社会处于大数据和人工智能时代。健康、金融、交通、商业、基因等大数据正在深刻地改变人们的生活和工作方式。在我国，生态环境大数据已提上议事日程，政府环保部门正在大力推动中。
没有对大量气味在线测试产生的多源感知数据，没有嗅辨数据和色/质谱等常规仪器的成分检测数据，企图单纯靠单一型气敏传感器阵列、单一气相色谱柱和单一机器学习模型来在线估计复杂气味的强度与多种组成成分浓度是不现实的。尽管很多电子鼻正是这样做的，但由此产生的检测数据的作用是十分有限的，得到的结果因而是不可信的。
由于气味复杂性和环境多变性，小数据不足以用来训练有效的机器学习模型以识别多种气味类型和量化预测复杂气味组成成分。我们应以气敏/色谱多源感知数据、嗅辨数据、色/质谱等常规仪器检测数据为 基础，建立气味大数据。有了气味大数据，机器学习方法就能依据当前感知信息，通过数据挖掘来识别气味类型和量化预测众多组成成分浓度。大数据和复杂气味多成分在线预测是矛盾的两个方面，有效解决途径是，深入研究并采用尽可能简单有效的机器学习模型与算法来实现气味的类型识别和气味强度、多种主要组成成分浓度的实时量化预测。
参考文献：
[1]P.Boeker,On'Electronic Nose'methodology,Sensors&Actuators B-Chemical,2014,204:2–17.
发明内容
本发明是在现有发明专利《一种恶臭气体多点集中式在线监测与分析***及方法》(参见申请号：2018104716131)、《大数据驱动的恶臭气体多点集中式电子鼻仪器在线分析方法》(参见申请号：2018104717083)和《一种多通道集成嗅觉模拟仪器和生物发酵过程在线分析方法》(参见申请号：201310405315.X)的基础上，发明一种电子鼻仪器和生物发酵/恶臭污染过程在线检测与分析方法，以解决多个发酵过程或多个恶臭污染点的长期在线检测、气味类型识别和气味强度定性指标与多种化合物浓度控制指标值的在线量化预测问题。
为了实现上述目的，本发明提供如下技术方案：
在一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法中，电子鼻仪器包括气敏传感器阵列模块I、毛细管气相色谱柱模块II、气体自动进样模块III、计算机控制与分析模块IV，以及辅助气源V，实现多个生物发酵过程或多个恶臭污染过程的长期循环在线检测与智能分析。
气敏传感器阵列模块I包括气敏传感器阵列I-1、气敏传感器阵列环形工作腔I-2、电阻加热元件I-3、风扇I-4、隔热层I-5和隔板I-6，位于电子鼻仪器右中部。
毛细管气相色谱柱模块II包括毛细管气相色谱柱II-1、检测器II-2、放大器II-3、记录仪II-4、进样口II-5、电阻加热丝II-6、风扇II-7和隔热层II-8，位于电子鼻仪器右上部。
气体自动进样模块III包括：第一～第五二位二通电磁阀III-1～III-5、5个第一净化器III-6、第一微型真空泵III-7、第一流量计III-8、第六二位二通电磁阀III-9、第一节流阀III-10、二位三通电磁阀III-11、三位四通电磁阀III-12、第二微型真空泵III-13、第七二位二通电磁阀III-14、第八二位二通电磁阀III-15、稳压阀III-16、第一减压阀III-17、第二节流阀III-18、第二净化器III-19、第二减压阀III-20、第三净化器III-21、第三节流阀III-22、第二流量计III-23、第四节流阀III-24、第五节流阀III-25，位于电子鼻仪器右下部。
计算机控制与分析模块IV包括计算机主板IV-1、A/D数据采集卡IV-2、驱动与控制电路板IV-3、4路精密直流稳压电源IV-4、显示器IV-5、WIFI模块IV-6，位于电子鼻仪器左侧。
一个生物发酵过程/发酵罐或一个恶臭污染点被简称为检测点；电子鼻仪器对一个检测点的被测气体进样单周期为T
0＝300-600s，默认T
0＝480s。在气体进样单周期T
0内，一个检测点的被测气体被2个微型真空泵III-7和III-13分别抽吸到气敏传感器阵列模块I和毛细管气相色谱柱模块II内，气敏传感器阵列I-1和毛细管气相色谱柱II-1产生敏感响应，电子鼻仪器因此得到1组气敏传感器阵列响应曲线和1幅气相色谱图，这是电子鼻仪器感知一个被测气体样品而得到的气敏/气相色谱模拟信号。
在气体进样单周期T
0内，计算机控制与分析模块IV从气敏传感器阵列I-1时长60s的每条电压响应曲线中选择稳态峰值v
gs_i(τ)、对应的出峰时间t
gs_i(τ)、曲线下面积Ags
_i(τ)这3个感知信息，以满足三角稳定性原理，提高气敏传感器阵列的定性定量能力。若气敏传感器阵列I-1由16个气敏传感器组成，则i＝1,2,…,16，计算机控制与分析模块IV在气体进样单周期T
0内从气敏传感器阵列16条响应曲线中共得到16*3＝48个感知分量。
在气体进样单周期T
0内，电子鼻仪器不追求色谱图峰/峰完全分离，计算机控制与分析模块IV从半分 离色谱图上选择前10个最大色谱峰值v
gci(τ)和相对应的保留时间t
gci(τ)、色谱图曲线下面积A
gc(τ)，i＝1,2,…,10，共得到21个感知分量，以提高气相色谱柱的在线检测能力。
在气体进样单周期T
0内，电子鼻仪器感知一个生物发酵过程或一个恶臭污染点的被测气体，计算机控制与分析模块IV将从气敏传感器阵列I-1的16条响应曲线提取的48个感知分量和从毛细管色谱柱II-1半分离色谱图提取的21个感知分量融合起来，得到一个m＝48+21＝69维的感知向量x(τ)∈R
69，称之为样本，这是电子鼻仪器对生物发酵过程或恶臭污染过程进行定性定量分析的依据。
电子鼻仪器对n(≤5)个生物发酵过程或n(≤5)个恶臭监测点的气体循环进样周期为T＝n*T
0，依次得到n个样本，依次存储在计算机硬盘的n个对应数据文件里，并通过WIFI路由模块将样本数据发送到云端和指定的固定/移动终端。若T
0＝480s，则被测气体循环进样周期为T＝n*T
0＝n*480s，相当于一个发酵罐或一个恶臭污染点每隔n*480s被检测一次。
电子鼻仪器通过对多个生物发酵过程、多个恶臭污染点经年累月的长期在线检测，形成气味大数据X的主体；数据集X还包括气相色谱、质谱、分光光度等常规分析仪器离线检测数据，专业人员实验室嗅辨得到的臭气浓度OU值数据，操作人员记录的青霉素、红霉素、酱油、食醋、料酒、味精等生物发酵类型数据和化工园区、垃圾填埋场、污水处理厂、畜禽养殖场等恶臭污染监测区域类型数据；数据集X的一部分子集建立了气敏/色谱响应与多个生物发酵过程/恶臭污染类型以及主要成分浓度的对应关系。
在学习阶段，气味大数据X的各感知分量被施以归一化预处理，计算机控制与分析模块IV的机器学习模型离线学习气味大数据X以确定其结构和参数。在决策阶段，机器学习模型在线学习气敏-色谱近期响应以微调模型参数，依据气敏/气相近期已发生感知时间序列阵在线确定多个生物发酵过程和恶臭污染类型，量化预测生物发酵过程发酵液主要成分浓度或国标GB14554指定的氨NH
3、硫化氢H
2S、二硫化碳CS
2、三甲胺C
3H
9N、甲硫醇CH
4S、甲硫醚C
2H
6S、二甲二硫醚C
2H
6S
2、苯乙烯C
8H
8这8种恶臭化合物和臭气浓度OU值共8+1种恶臭污染物浓度指标值。
气敏传感器阵列I-1及其环形工作腔I-2位于55±0.1℃恒温箱内。在气体进样单周期T
0内，气敏传感器阵列模块I依次经历了气敏传感器阵列初步恢复T
0-120s、洁净空气精确标定40s、平衡5s、被测气体顶空进样60s、过渡5s、环境净化空气冲洗10s共6个阶段。这6个阶段的气体类型与流量依次是：①环境净化空气6,500ml/min、②洁净空气1,000ml/min、③气体不流动、④被测气体1,000ml/min、⑤环境净化空气1,000ml/min、⑥环境净化空气6,500ml/min；“过渡”主要指从被测气体到环境净化空气的转换。
气体进样单周期T
0的[T
0-75s,T
0-15s]时间区间为气敏传感器阵列模块I的被测气体顶空进样阶段，第一～第五这5个二位二通电磁阀其中之一III-k(k＝1,2,…,5)导通，三位四通电磁阀III-12处于位置“0”，第六和第七二位二通电磁阀III-9与III-14断开、第八二位二通电磁阀III-15导通。在第一微型真空泵III-7的抽吸作用下，一个检测点的被测气体以1,000ml/min的流量依次流过第k二位二通电磁阀III-k(k＝1,2,…,5)、第八二位二通电磁阀III-15、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续60s；气敏传感器阵列I-1因此对被测气体产生敏感响应，并被存储在计算机控制与分析模块IV的临时文件里。
气体进样单周期T
0的[T
0-120s,T
0-80s]时间区间为气敏传感器阵列模块I的洁净空气标定阶段，三位四通电磁阀III-12处于位置“1”，第六、第七和第八二位二通电磁阀III-9、III-14和III-15均断开，洁净空气瓶V-2中的洁净空气以1,000ml/min的流量依次流经第一减压阀III-17、第二节流阀III-18、第二净化器III-19、三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续40s。在此期间，气敏传感器阵列I-1在洁净空气的作用下精确恢复到基准状态。由于第八二位二通电磁阀III-15断开，第一～第五这5个二位二通电磁阀III-1～III-5的导通与否不影响气敏传感器阵列I-1的标定。
“环境净化空气”是指电子鼻仪器所处的室外空气经除尘、去湿及无菌处理后的空气，仅用于气敏传感器阵列I-1的初步恢复、环形工作腔I-2和相关气路管道内壁的冲洗、以及气敏传感器阵列累积热量的带走。在气体进样单周期T
0的[0,T
0-120s]和[T
0-10s,T
0]这两个时间段，三位四通电磁阀III-12处于位置“2”， 第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开，环境净化空气以6,500ml/min的流量依次流经三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外，持续T
0-110s。在此期间，气敏传感器阵列I-1在环境净化空气作用下初步恢复到基准状态；由于第八二位二通电磁阀III-15断开，第一～第五这5个二位二通电磁阀III-1～III-5导通与否不影响气敏传感器阵列I-1的初步恢复。
商用毛细管色谱柱II-1尺寸默认为长度L×内径φd×膜厚δ＝30m×φ0.53mm×0.25μm，位于250-300±0.1℃的恒温箱内。在气体进样单周期T
0内，毛细管气相色谱柱模块II依次经历被测气体顶空进样1s、被测气体色谱分离T
0-16s、放空与清洗吹扫15s共三个阶段；H
2兼作载气和燃气，洁净空气为助燃气。
气体进样单周期T
0的最初1s是毛细管气相色谱柱模块II的被测气体顶空进样阶段，第一～第五二位二通电磁阀之一III-k(k＝1,2,…,5)导通，二位三通电磁阀III-11处于位置“1”，第七二位二通电磁阀III-14导通，第八二位二通电磁阀III-15断开。这时，检测点k的被测气体在第二微型真空泵III-13的抽吸作用下，依次流经第一～第五二位二通电磁阀之一III-k(k＝1,2,…,5)、第七二位二通电磁阀III-14、二位三通电磁阀III-11、第四节流阀III-24，在进样口II-5处与载气H
2混合，因此流入毛细管气相色谱柱II-1，持续1s；被测气体默认进样流量6ml/min，默认进样持续时间1s，默认累积进样量0.1ml。
气体进样单周期T
0的[1s,T
0-10s]时间区间为毛细管气相色谱柱模块II的被测气体分离阶段，二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开，来自检测点k的被测气体因此断开，历时T
0-11s。注入色谱柱模块II进样口II-5的被测气体在一定压力和流量的载气H
2的推动作用下，在毛细管气相色谱柱II-1内产生分离，检测器II-2因此产生感知，经放大器II-3放大后，记录仪II-4将[0,T
0-10s]时间区间内即色谱柱II-1时长T
0-10s的感知响应记录下来，并存储在计算机控制与分析模块IV的临时文件里。
气体进样单周期T
0的最后10s即[T
0-10s,T
0]时间区间是毛细管气相色谱柱II-1的放空亦即清洗吹扫阶段，第一～第五这5个二位二通电磁阀III-1～III-5中，原来导通的那一个即III-k断开，原来关闭的其它4个之一即III-(～k)导通；二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14导通，第八二位二通电磁阀III-15断开。假设二位二通电磁阀III-(～k)导通(k＝1,2,…,5)，在第二微型真空泵III-13的抽吸作用下，以330ml/min流量依次流经二位二通电磁阀III-(～k)、第七二位二通电磁阀III-14、二位三通电磁阀III-11，然后直接被排出到室外。这一阶段的作用是，清除相关管道在现行气体进样单周期来自第k个检测点的气味残留，并逐渐由第(～k)个检测点的被测气体所取代，为下一气体进样单周期检测另一个生物发酵过程或恶臭污染监测点做准备，持续10s。
气体进样单周期T
0的[T
0-10s,T
0]时间区间同时为信息选择与分析时间段，计算机控制与分析模块IV从[T
0-75s,T
0-15s]时间段的气敏传感器阵列I-1电压响应曲线中选择包括稳态峰值v
gsi(τ)在内的48个感知信息。从[0,T
0-10s]时间段的色谱图上选择包括前10个最大色谱峰值v
gci(τ)在内的21个感知分量。这是电子鼻仪器对生物发酵过程或恶臭污染区域进行分析的依据。计算机控制与分析模块IV依据感知向量x(τ)和气味大数据X进行气味类型识别和强度与主要浓度指标值量化预测。
在气体进样单周期T
0情况下，当只检测一个生物发酵过程或一个恶臭污染点时，则气体检测与分析周期循环是T＝T
0。若同时检测k个生物发酵过程/恶臭污染点，则其中一个生物发酵过程/恶臭污染点的循环检测与分析周期是T＝k*T
0。在长期循环监测过程中，其中一个生物发酵过程/恶臭污染点退出，则气体循环检测与分析周期变成T＝(k-1)*T
0。类似地，在长期循环监测过程中，一个新的生物发酵过程/恶臭污染点中途加入，则循环检测与分析周期变成T＝(k+1)*T
0。自一个生物发酵过程/恶臭污染点退出/加入时刻起，对应的数据文件记录周期相应变化。
在气体进样单周期T
0内，[T
0-10s,T
0]时间区间即为时长10s的信息选择与分析时间段，计算机控制与分析模块IV对气敏传感器阵列模块I和毛细管气相色谱柱模块II二者同时进行感知信息选择与分析处理操作。计算机控制与分析模块IV从气敏传感器阵列I-1在[T
0-75s,T
0-15s]时间段即时长60s的每条电压响应曲线中选择选择稳态峰值v
gs_i(τ)、对应的出峰时间t
gs_i(τ)、曲线下面积Ags
_i(τ)这3个感知信息分量，从毛细管气相色谱柱II-1在[0,T
0-10s]时间段即时长T
0-10s的一幅半分离色谱图上选择前10个最大色谱峰值 v
gc_i(τ)和10个相对应的保留时间t
gc_i(τ)、1个色谱图曲线下面积A
gc(τ)共21个感知响应分量，存储在计算机硬盘的临时文件里。
在气体进样单周期T
0内，若时长T
0-10s的半分离色谱图的色谱峰个数q小于10，则计算机控制与分析模块IV从该半分离色谱图上选择前q<10个最大色谱峰值v
gci(τ)、相对应的保留时间t
gci(τ)以及色谱图曲线下面积A
gc(τ)，不足的色谱峰值和对应保留时间补零，这时得到的色谱感知信息是x
gc(τ)＝{(h
gc1(τ),h
gc2(τ),…,h
gc,q(τ),0,…,0)；(t
gc1(τ),t
gc2(τ),…,t
gc,q(τ),0,…,0)；A
gc(τ)}。
气体进样单周期T
0最后10s即[T
0-10s,T
0]区间的信息处理与分析时间段，计算机控制与分析模块IV的模块化机器学习模型依据气敏/色谱近期感知时间序列矩阵X(τ-q)对生物发酵过程或恶臭污染监测点进行气味类型识别和强度与主要成分量化预测，包括：生物发酵过程类型与恶臭污染类型识别，生物发酵过程细胞浓度、底物浓度、产物浓度量化估计，正丙醇、苯乙醇等发酵过程前体物质浓度量化估计，GB14554指定的8+1种恶臭污染物浓度指标值量化预测；这里，τ为当前时间，q为近期已过去的时间，τ-q是近期时间间隔。
气味大数据X还包括：浓度为0.1-1,0000ppm的多种单一化合物顶空挥发气的电子鼻仪器气敏/色谱感知数据，气相色谱、质谱和分光光度等常规分析仪器的离线检测数据；专业人员实验室嗅辨数据。单一化合物特别包括生物发酵过程前体物质正丙醇与苯乙酸、GB14554指定的8种恶臭化合物，以及欧洲标准EN13725指定的臭气浓度OU标准参照物—正丁醇。
机器学习模型由多个模块化深度卷积神经网络组成；单输出深度卷积神经网络模块数与被预测的生物发酵过程发酵液主要成分数、恶臭污染物主要浓度指标数、被测对象类型数相等，一一对应。一个单输出深度卷积神经网络由一个输入层、3个卷积层、2个下采样层和1个输出单元组成，各隐层与输出层活化函数均为Sigmoid修正活化函数
在学习阶段，各个单输出深度卷积神经网络均采用误差反传离线逐层学习算法，主要学习气味大数据中有标签的数据和组成成分已知的气味大数据而具有必要智能；卷积层扫描窗尺寸为5×5，重叠扫描步长为1；卷积核为正弦核、余弦核、多项式核、Gaussian核、Sigmoid核、小波核和Laplace核的组合。下采样层扫描窗尺寸为2×2，不重叠扫描即步长为2，提取最大值、均值和均方差特征。在决策阶段，n个单输出深度卷积神经网络模型依据气敏/气相色谱当前时刻τ和近期已发生的时间序列感知矩阵X(τ-q)进行气味类型识别、一一估计预测当前时刻τ和未来τ+1、τ+2、τ+3时刻的气味强度与主要组成成分浓度值。
应用本发明的气敏-气相电子鼻仪器和发酵-恶臭多状态参数在线分析方法，电子鼻仪器对多个生物发酵过程/恶臭污染点长期循环在线检测和在线分析预测，包括以下步骤：
(1)开机：仪器预热30min；
修改屏幕菜单“气体进样单周期T
0”设置，默认值T
0＝8min；5个检测点气体循环进样周期为T＝5T
0。
三位四通电磁阀III-12处于位置“2”，第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开；在第一微型真空泵III-7的抽吸作用下，环境净化空气以6,500毫升/分钟的流量依次流经三位四通电磁阀III-12，稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外，气敏传感器阵列环形工作腔I-1内部温度达到恒定的55±0.1℃。
二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开，在载气H
2的推动作用下，毛细管气相色谱柱II-1逐步恢复到基准状态，色谱柱恒温箱内部温度达到恒定的250±0.1℃，
(2)气体循环进样周期开始：单击显示器IV-5屏幕菜单的“检测点k开通”选项，k＝1,2,…,5，电子鼻仪器长期持续检测直至操作人员单击“检测点k断开”选项为止。电子鼻仪器依次对5个检测点进行循环检测，计算机控制与分析模块IV自动生成5个文本文件，以存储气敏传感器阵列I-1和毛细管气相色谱柱模块II对5个检测点气体的感知响应数据。
(3)检测点k气体进样单周期开始；以T
0＝8min为例：
(3.1)气敏传感器阵列模块I：依次历经①360s的初步恢复、②40s的精确标定、③5s的平衡、④60s的顶空进样、⑤5s的过渡和⑥10s的清洗与初步恢复共六个气体进样阶段。
(3.1a)初步恢复：在气体进样单周期T
0第0-360s，三位四通电磁阀III-12处于位置“2”，第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开。在第一微型真空泵III-7抽吸作用下，环境净化空气以6,500ml/min的流量依次流经三位四通电磁阀III-12，稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外；气敏传感器阵列I-1初步恢复到基准状态。
(3.1b)精确标定：在气体进样单周期T
0第360-400s，三位四通电磁阀III-12处于位置“1”，第六、第七和第八二位二通电磁阀III-9、III-14和III-15均断开，洁净空气以1,000ml/min的流量依次流经第一减压阀III-17、第二节流阀III-18、第二净化器III-19、三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续40s；气敏传感器阵列I-1因此精确恢复到基准状态。
(3.1c)平衡：在气体进样单周期T
0第400-405s，三位四通电磁阀III-12处于位置“0”，第六和第八二位二通电磁阀III-9与III-15均断开，气敏传感器阵列环形工作腔I-2内部无气体流动，持续5s。
(3.1d)顶空进样：在气体进样单周期T
0第405-465s，第一～第五这5个二位二通电磁阀其中之一III-k(k＝1,2,…,5)导通，三位四通电磁阀III-12处于位置“0”，第六和第七二位二通电磁阀III-9与III-14断开、第八二位二通电磁阀III-15导通。在第一微型真空泵III-7的抽吸作用下，一个检测点的被测气体以1,000ml/min的流量依次流过二位二通电磁阀III-k(k＝1,2,…,5)、第八二位二通电磁阀III-15、稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续60s。气敏传感器阵列I-1因此产生敏感响应被存储在计算机控制与分析模块IV对应的临时文件里。
(3.1e)过渡：在气体进样单周期T
0第465-470s，三位四通电磁阀III-12处于位置“2”，第八二位二通电磁阀III-15断开，第六和第七二位二通电磁阀III-9与III-14保持断开。在第一微型真空泵III-7抽吸作用下，环境净化空气以1,000ml/min的流量依次流经三位四通电磁阀III-12，稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外。
(3.1f)清洗与初步恢复：在气体进样单周期T
0第470-480s，与“过渡”阶段相比，除第六二位二通电磁阀III-9由“断开”转为“导通”，其余阀位置相同。环境净化空气流量因此从“1,000ml/min”转变为“6,500ml/min”；这一阶段与即将开始的下一单周期“初步恢复”阶段的阀位置与工作状态完全相同和衔接。
(3.2)毛细管气相色谱柱II模块：依次历经①1s的顶空进样、②469s的色谱分离和③10s的放空与清洗共3个气体进样阶段。
(3.2a)顶空进样：在气体进样单周期T
0第0-1s，第一～第五这5个二位二通电磁阀之一III-k(k＝1,2,…,5)导通，二位三通电磁阀III-11处于位置“1”，第七二位二通电磁阀III-14导通，第八二位二通电磁阀III-15断开。在第二微型真空泵III-13的抽吸作用下，检测点k的被测气体依次流经第一～第五二位二通电磁阀之一III-k(k＝1,2,…,5)、第七二位二通电磁阀III-14、二位三通电磁阀III-11、第四节流阀III-24，在进样口II-5处与载气H
2混合，流入毛细管气相色谱柱II-1，持续1s。
(3.2b)色谱分离：在气体进样单周期T
0第1-470s，二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开。被测气体在一定压力和流量的载气H
2的推动作用下，在毛细管气相色谱柱II-1内分离，检测器II-2因此产生感知响应，经放大器II-3放大后，记录仪II-4将[0,470s]区间时长470s的感知响应记录下来，形成半分离色谱峰图，并存储在计算机控制与分析模块IV对应的临时文件里。
(3.3)信息选择与分析：在气体进样单周期T
0第470-480s，计算机控制与分析模块IV从气敏传感器阵列模块I的单个气敏传感器在[405s,465s]时间段时长60s的单条电压响应曲线中选择稳态峰值v
gs_i(τ)、出峰时间t
gs_i(τ)、总曲线下面积Ags
_i(τ)这3个感知信息；16个气敏传感器组成的阵列I-1共得到 16*3＝48个感知分量。与此同时，计算机控制与分析模块IV从毛细管色谱柱模块II时长470s的半分离色谱图上选择前10个最大色谱峰值v
gc_i(τ)和对应保留时间t
gc_i(τ)、总色谱图曲线下面积A
gc(τ)，共得到21个感知分量。在单周期T
0内，计算机控制与分析模块IV从气敏传感器阵列I模块和毛细管色谱柱模块II的感知信息中共得到1个69维的感知向量x(τ)∈R
69；然后，机器学习模型依据感知向量x(τ)和气味大数据X进行气味类型识别和强度与主要成分量化预测，显示器显示监测和预测结果，并通过Internet网络传送到中央控制室和多个固定/移动终端。
(3.4)检测点k结束与下一个检测点开始。
第一～第五这5个二位二通电磁阀其中之一III-k(k＝1,2,…,5)由原来的导通转为断开，与下一个检测点对应的第一～第五二位二通电磁阀其中之一导通。
(4)重复步骤(3.1)～(3.4)，电子鼻仪器实现1～5个检测点被测气体的循环在线检测、识别和气味强度与多项浓度指标值量化预测。
图1是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气敏/色谱多感知信息融合、电子鼻仪器研制和发酵尾气与恶臭气味在线检测与多过程参数分析技术路线示意图。
图2是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—电子鼻仪器工作原理示意图。
图3是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气敏传感器阵列模块及其气路工作原理示意图。
图4是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—毛细管气相色谱柱模块及其气路工作原理示意图。
图5是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气敏传感器阵列模块和毛细管气相色谱柱模块示意图。
图6是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气体进样单周期T
0＝480s内，毛细色谱柱和气敏阵列模块气体进样时间、流量和气敏传感器响应变化情况示意图。
图7是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气体进样单周期T
0＝480s内，气敏传感器响应曲线多信息选择示意图。
图8是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气体进样单周期T
0＝480s内，半分离色谱图多信息选择示意图。
图9是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—气体进样单周期T
0＝480s内，当色谱峰数小于或大于10时，半分离色谱图多信息选择示意图。
图10是本发明—一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法—模块化深度卷积神经网络模型多参数“分而治之”在线量化预测示意图。
下面结合附图对本发明作进一步详细描述。
图1是本发明的气敏传感器阵列与毛细管气相色谱柱感知信息融合、电子鼻仪器研制和发酵尾气与恶臭气味多源在线检测与多过程参数分析技术路线示意图。图1显示的技术路线包括：(1),气敏、色谱感知元件性能评价与选择。深入分析气敏传感器和毛细管气相色谱柱之间的特性差异，致力于二者的取长补短、 优势互补。(2),气敏传感器阵列等部件模块化。气敏传感器阵列、毛细管气相色谱柱、气体自动进样、计算机控制与分析等重要部件实现结构模块化。(3),气敏/色谱在线感知模型与信息融合。发明满足三角稳定性原理的气敏传感器响应曲线多信息选择方法和模拟马拉松比赛场景的半分离色谱图多信息选择方法，实现气敏传感器阵列和毛细管色谱柱的在线感知与信息融合。(4),建立气味大数据。以对大量生物发酵过程或恶臭污染点的电子鼻仪器气敏/色谱多源在线感知数据、专业人员实验室嗅辨数据、色/质谱与分光光度等常规仪器离线检测数据为基础，形成气味大数据X。(5),机器学习模型离线学习与在线微调。机器学习模型离线学习气味大数据X，以优化确定模型结构与参数；在决策阶段，机器学习模型在线学习气敏/色谱近期响应以微调参数，并依据气敏/气相当前感知向量x(τ)在线确定生物发酵过程或恶臭污染类型，量化预测生物发酵过程发酵液主要成分浓度或国标GB14554指定的恶臭污染物8种成分浓度与臭气浓度OU值，实现多个生物发酵过程和多个恶臭污染点的复杂气味长期循环在线检测与在线分析。
图2是本发明的气敏/气相色谱相融合的电子鼻仪器工作原理示意图。电子鼻仪器主要包括：气敏传感器阵列模块I，毛细管气相色谱柱模块II，气体自动进样模块III，计算机控制与分析模块IV，以及氢气瓶V与洁净空气瓶VI。氢气兼作毛细管气相色谱柱模块II氢火焰离子化检测器FID的载气和燃气；洁净空气一方面作为毛细管色谱柱模块II的助燃气，另一方面作为气敏传感器阵列模块I的标定气体(不燃烧)。
图3和图4分别是本发明的电子鼻仪器气敏传感器阵列模块I、毛细管气相色谱柱模块II及其气路工作原理示意图。
气敏传感器阵列模块I主要组成单元包括：气敏传感器阵列I-1、气敏传感器阵列环形工作腔I-2、电阻加热元件I-3、风扇I-4、隔热层I-5和隔板I-6，位于电子鼻仪器右中部。毛细管气相色谱柱模块II主要组成单元包括：毛细管气相色谱柱II-1、检测器II-2、放大器II-3、记录仪II-4、进样口II-5、电阻加热丝II-6、风扇II-7和隔热层II-8，位于电子鼻仪器右上部。气敏传感器阵列模块I和毛细管气相色谱柱模块II的作用是将气味的化学与物理信息在线转换为电信号。
气体自动进样模块III与气敏传感器阵列模块I相关的组成单元包括：第一～第五二位二通电磁阀III-1～III-5、第一净化器III-6、第一微型真空泵III-7、第一流量计III-8、第六二位二通电磁阀III-9、第一节流阀III-10、三位四通电磁阀III-12、第七二位二通电磁阀III-14、第八二位二通电磁阀III-15、稳压阀III-16、第一减压阀III-17、第二节流阀III-18、第二净化器III-19。
气体自动进样模块III与毛细管气相色谱柱模块II相关的组成单元包括：二位三通电磁阀III-11、第二微型真空泵III-13、第二减压阀III-20、第三净化器III-21、第三节流阀III-22、第二流量计III-23、第四节流阀III-24、第五节流阀III-25。气体自动进样模块III位于电子鼻仪器右下方。
计算机控制与分析模块IV主要组成单元包括：计算机主板IV-1、A/D数据采集卡IV-2、驱动与控制电路板IV-3、4路精密直流稳压电源IV-4、显示器IV-5、WIFI模块IV-6，位于电子鼻仪器左侧。WIFI模块IV-6的作用是实时将气敏传感器阵列模块I和毛细管气相色谱柱模块II的感知信息传送给指定的固定/移动终端。
图5是电子鼻仪器的气敏传感器阵列模块I和毛细管气相色谱柱模块II示意图。气敏传感器阵列和毛细管气相色谱柱均位于恒温温度不同的两个恒温箱内，形成两个模块，可根据需要便捷更换。
图6是电子鼻仪器在气体进样单周期T
0＝480s内，气敏传感器阵列模块I和毛细管气相色谱柱模块II气体进样时间、流量和气敏传感器响应变化情况示意图。气体进样单周期可在T
0＝5～10min之间调整，图6仅给出了默认的气体进样单周期T
0＝480s例子。可调整的时间段主要为气敏传感器阵列模块I的环境净化空气冲洗/气敏传感器初步恢复阶段和毛细管气相色谱柱模块II的分离阶段。图6表明，在气体进样单周期为T
0＝480s内，气敏传感器阵列模块I和毛细管气相色谱柱模块II对被测气体的进样流量与累积进样量不相等，进样时间不同步，但在最后10s阶段对信息选择与分析区是同时进行的。
图6(a)给出了毛细管色谱柱模块II的气体进样单周期情况，包括①被测气体顶空进样、②被测气体色谱分离和③色谱柱放空3个阶段。①被测气体顶空进样阶段处于进样单周期T
0的开始阶段，进样时长范围为0.5～1.5s，默认1s；进样流量范围为1.5～15ml/min，默认6ml/min；累计进样量范围为0.0125～0.375ml，默认0.1ml。
表1,在气体进样单周期T
0＝300-600s(默认480s)，毛细管气相色谱柱模块II工作参数与相关电磁阀导通/断开状况
请参见图6，并结合图4，表1给出了在气体进样单周期T
0＝480s内，毛细管气相色谱柱模块II工作参数和相关电磁阀导通/断开状况。在被测气体进样阶段①，二位三通电磁阀III-11处于位置“1”，第七二位二通电磁阀III-14导通，第一～第五这五个二位二通电磁阀III-1～III-5之一导通，第八二位二通电磁阀III-15断开。假设第一二位二通电磁阀III-1导通，这时，其中一个生物发酵过程(发酵罐)或恶臭污染点的被测气体，例如第一检测点，在第二微型真空泵III-13的抽吸作用下，依次流经第一二位二通电磁阀III-1、第七二位二通电磁阀III-14、二位三通电磁阀III-11、第四节流阀III-24，在进样口II-5处与载气H
2混合，因此流入毛细管气相色谱柱II-1。
在被测气体进样阶段①，若默认进样流量6ml/min，默认进样持续时间1s，则被测气体进样量为0.1ml，符合毛细管气相色谱柱的最佳进样量要求。在T
0＝480s历时369s的色谱分离阶段②，二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开，亦即被测气体断开。在此期间，在载气H
2的推送作用下，被测气体在毛细管气相色谱柱II-1内实现分离。
在气体进样单周期T
0最后10s的色谱柱放空阶段③，即清洗、吹扫阶段，二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14导通，第一～第五这5个二位二通电磁阀III-1～III-5之一导通(但原来导通的关闭)，第八二位二通电磁阀III-15断开。假设第一二位二通电磁阀III-1导通，在第二微型真空泵III-13的抽吸作用下，依次流经第一二位二通电磁阀III-1、第七二位二通电磁阀III-14、二位三通电磁阀III-11，然后被排出到室外。这一阶段的作用是，清除相关管道在本次气体进样单周期的残留，为下一气体进样单周期做准备。必须指出，三位四通电磁阀III-12的位置由随后给出的表2所确定。
请参见图6，并结合图3，表2给出了在气体进样单周期T
0内，气敏传感器阵列模块I工作参数与相关电磁阀导通/断开状况。
以下以气体进样单周期T
0＝480s为例，详细阐述气敏传感器阵列模块I的几个主要工作状态。
在被测气体顶空进样阶段④，即气体进样单周期T
0的405-465s时间段，第一～第五这5个二位二通电磁阀III-1～III-5之一导通，三位四通电磁阀III-12处于位置“0”，第六和第七二位二通电磁阀III-9与III-14断开、第八二位二通电磁阀III-15导通。在第一微型真空泵III-7的抽吸作用下，5个生物发酵过程(发酵罐)或恶臭污染点之一(例如第一检测点)的被测气体以1,000ml/min的流量依次流过第一～第五这5个二位二通电磁阀III-1～III-5之一、第八二位二通电磁阀III-15、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续60s。在此期间，气敏传感器阵列I-1对被测气体产生敏感响应。
在洁净空气标定阶段②，即气体进样单周期T
0的360-400s时间段，三位四通电磁阀III-12处于位置“1”，第六、第七和第八二位二通电磁阀III-9、III-14和III-15断开，洁净空气瓶VI中的洁净空气以1,000ml/min
表2,在气体进样单周期T
0＝300-600s(默认480s)，气敏传感器阵列模块I工作参数与相关电磁阀导通/断开状况
的流量依次流经第一减压阀III-17、第二节流阀III-18、第二净化器III-19、三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续40s。在此期间，气敏传感器阵列I-1在洁净空气的作用下精确恢复到基准状态。由于第八二位二通电磁阀III-15断开，第一～第五这五个二位二通电磁阀III-1～III-5的导通/断开与否不影响气敏传感器阵列I-1的标定。
在气敏传感器初步恢复阶段①和环境净化空气冲洗阶段⑥，即气体进样单周期T
0的0-360s和470-480两个时间段，三位四通电磁阀III-12处于位置“2”，第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开，环境净化空气以6,500ml/min的流量依次流经三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外，持续370s。在环境净化空气作用下，气敏传感器阵列I-1初步恢复到基准状态。由于第八二位二通电磁阀III-15断开，第一～第五位二通电磁阀III-1～III-5、和第六与第七二位二通电磁阀III-9与III-14导通/断开与否不影响气敏传感器阵列I-1的初步恢复。
必须指出，“环境净化空气”是指电子鼻仪器所处的室外空气经除尘、去湿及无菌处理后的空气，仅用于气敏传感器阵列I-1的初步恢复，气敏传感器阵列环形工作腔I-2及相关气路管道内壁的冲洗，以及气敏传感器阵列累积热量的带走。
请参见图6，在气体进样单周期T
0的最后10s时间段，气敏传感器阵列模块I和毛细管气相色谱柱模块II同时进入信息选择与分析区。计算机控制与分析模块IV从[T
0-75s,T
0-15s]时间段的每条气敏传感器电压响应曲线中选择稳态峰值v
gsi(τ)、相对应的出峰时间t
gsi(τ)、曲线下面积Ags
i(τ)这3个感知信息；从[0,T
0-10s]时间段的色谱图上选择前10个最大色谱峰值v
gci(τ)和相对应的10个保留时间t
gci(τ)、色谱图曲线下面积A
gc(τ)，共21个色谱感知分量。这是电子鼻仪器对生物发酵过程或恶臭污染区域进行分析预测的依据，是建立气味大数据X的基础；计算机控制与分析模块IV的机器学习模型依据感知向量x(τ)进行气味类型的识别和强度与主要浓度指标值的量化预测。
图7是气体进样单周期T
0＝480s内，气敏传感器响应曲线多特征选择示意图。图中给出了TGS822、TGS826和TGS832这3个气敏传感器分别对石油蜡样品、2,000ppm乙烯气及5,000ppm乙醇挥发气的相应曲线示例。其中，图7(b)和图7(c)两幅图的电压响应曲线稳态最大值相等，即v
a＝v
b。若仅依据常规的单一的电压响应曲线稳态最大值特征选择法，电子鼻仪器此时不能区别2,000ppm乙烯气和5,000ppm乙醇挥发气。经仔细观察，我们发现，图7(b)和图7(c)两幅图显示了情况1：尽管电压响应稳态最大值相等，但峰值对应出峰时间不相等，曲线下面积也互不相等。类似地，还有情况2：出峰时间相等，但峰值与曲线下面积互不相等。情况3：曲线下面积相等，但出峰时间与峰值互不相等。
依据图7，本发明提出，从一个气敏传感器i(＝1,2,…,16)的响应曲线上，同时选择电压响应稳态最大值v
gsi(τ)，与之对应的从被测气体顶空采样开始之刻起的出峰时间t
gsi(τ)，再加上被测气体顶空采样时间段60s的曲线下面积A
gsi(τ)。若气敏传感器阵列由16个敏感元件组成，则在气体进样单周期T
0历时10s的信息选择与处理区，计算机控制与分析模块IV从16条响应曲线中依次选择3*16＝48个特征数值作为气敏传感器阵列模块I对被测气体的一次感知信息，记为x
gs(τ)＝{(v
gs1(τ),v
gs2(τ),…,v
gs16(τ))；(t
gs1(τ),t
gs2(τ),…,t
gs16(τ))；(A
gc1(τ),A
gc2(τ),…,A
gc16(τ))}。
图8为气体进样单周期T
0＝480s内，一幅半分离色谱图的信息选择示意图。在气体进样单周期T
0历时5s的信息选择区，计算机控制与分析模块IV从这幅半分离色谱图中依次选择10组{峰高h
gci(τ)，保留时间t
gci(τ)}(i＝1,2,…,10)和指定时长470s的谱图曲线下面积A
gc(τ)共21个特征数值作为毛细管色谱柱模块II对被测气体的一次感知信息，记为x
gc(τ)＝{(h
gc1(τ),h
gc2(τ),…,h
gc10(τ))；(t
gc1(τ),t
gc2(τ),…,t
gc10(τ))；A
gc(τ)}。
图9为气体进样单周期T
0＝480s内，两个半分离色谱图特征选择示意图。图9(a)的半分离色谱图只有8个色谱峰，只能得到8个峰值h
gci(τ)(i＝1,2,…,8)以及对应保留时间t
gci(τ)(i＝1,2,…,8)，再加上半分离色谱图曲线下面积A
gc(τ)。我们的做法是，数量不足的色谱峰值和对应保留时间补零，因此，依据图9(a)，最终得到的色谱感知信息是x
gc(τ)＝{(h
gc1(τ),h
gc2(τ),…,h
gc8(τ),0,0)；(t
gc1(τ),t
gc2(τ),…,t
gc8(τ),0,0)；A
gc(τ)}。图9(b)的半分离色谱图有10个以上色谱峰，我们从中选择前10个最大色谱峰即可。
本发明将半分离色谱图看成电子鼻仪器感知信息即模式的一部分，结合气敏传感器阵列感知信息，建立气味大数据，藉助人工智能机器学习方法实现未知气味识别、定性分析和主要成分量化预测。在气体单采样周期T
0历时10s的信息选择与处理区，计算机控制与分析模块IV将气敏传感器阵列模块I和毛细管色谱柱模块II不同时间段对被测气体感知信息融合起来，并进行归一化预处理，于是得到电子鼻仪器对一个被测气体样品的感知信息向量x(τ)＝x
gs(τ)+x
gc(τ)＝{(v
gs1(τ),v
gs2(τ),…,v
gs16(τ))；(t
gs1(τ),t
gs2(τ),…,t
gs16(τ))；(A
gc1(τ),A
gc2(τ),…,A
gc16(τ))；(h
gc1(τ),h
gc2(τ),…,h
gc10(τ))；(t
gc1(τ),t
gc2(τ),…,t
gc10(τ))；A
gc}∈R
69。感知向量x(τ)∈R
69是电子鼻仪器对生物发酵过程和恶臭污染点的气味进行在线类型识别和主要成分量化预测的依据。
图10是面向“连续在线”分析方式的深度卷积神经网络机器学习模型多参数“分而治之”量化预测示意图。具体步骤包括：依据气敏传感器阵列模块I与毛细管气相色谱柱模块II近期感知得到的时间序列矩阵X(τ-q)，由多个单输出深度卷积神经网络一一预测发酵与恶臭污染类型、气味强度与主要成分浓度值。这里，τ为当前时间，q为近期已过去的时间，τ-q是近期时间间隔。因此，时间序列矩阵X(τ-q)维数尺度为R
69×(τ-
q+1)。q的取值一般以发酵或恶臭污染过程近期6小时左右时间长度为宜。
为了确定模块化卷积神经网络模型结构和参数，首要任务是建立气味大数据，包括：气敏传感器阵列模块I与毛细管气相色谱柱模块II对大量生物发酵过程和恶臭污染区域经年累月的在线感知数据；色谱仪、质谱仪、分光光度仪等常规仪器的离线监测数据；已知类型与组成成分的气味标签数据；以及感官评定数据。
接下来要做的事情是气敏传感器阵列感知数据与毛细管气相色谱柱感知数据的融合，包括归一化和降维预处理。为了降低气味大数据分析难度，我们采用“分而治之”策略，将一个复杂的多气味定性定量分析问题，即复杂的多气味类型识别问题和复杂的多气味强度与组成成分量化估计问题，分解为多个气味类型一一识别和多个较简单的单一气味强度与重要组成成分一一量化预测问题，也就是将一个n条曲线/曲面整体拟合问题分解n条曲线/曲面一一拟合问题，并由n个单输出深度卷积神经网络模型解决，一一对应。
本发明采用多个模块化单输出深度卷积神经网络实现多参数在线量化预测。一个单输出深度卷积神经网络由一个输入层、3个卷积层、2个下采样层和1个输出单元组成，主要学习气味大数据中有标签的数据和组成成分已知的数据。各隐层与输出层活化函数均为修正的Sigmoid活化函数
采用误差反传离线逐层学习算法。卷积层扫描窗尺寸可为5×5，重叠扫描步长可为1；卷积核为正弦核、余弦核、多项式核、Gaussian核、Sigmoid核、小波核和Laplace核的组合；下采样层扫描窗尺寸可为2×2，不重叠扫描即步长为2，从每个扫描窗提取最大值、均值和均方差特征。在决策阶段，n个单输出深度卷积神经网络模型依据气敏/气相色谱近期感知时间序列矩阵X(τ-q)一一预测τ+1、τ+2、τ+3等即将发生时刻的多项量化指标值，包括气味类型、强度与主要组成成分浓度值。
气敏/气相色谱近期感知时间序列矩阵X(τ-q)的具体组成元素是：
本发明取m＝69，q＝9。
当只检测1个点时，即气体循环进样周期T与气体单采样周期T
0＝8min相等，设置“q＝9”相当于依据从当前时刻τ到过去1.2小时这个时间段的气敏/气相色谱时间序列感知响应矩阵，等价于依据过去1.2小时时间段的发酵过程或恶臭环境的变化，来预测未来8min、16min和24min的气味强度和主要组成成分可能的变化。当检测5个点时，即气体循环进样周期T＝5T
0＝40min相等，设置“q＝9”相当于依据从当前时刻τ到过去6小时这个时间段的气敏/气相色谱时间序列感知响应矩阵，来预测未来40min、80min和120min的气味强度和主要组成成分可能的变化。
Claims (14)
- 一种气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，电子鼻仪器包括气敏传感器阵列模块I、毛细管气相色谱柱模块II、气体自动进样模块III、计算机控制与分析模块IV，以及辅助气源V，实现多个生物发酵过程或多个恶臭污染过程的长期循环在线检测与智能分析；所述的气敏传感器阵列模块I包括气敏传感器阵列I-1、气敏传感器阵列环形工作腔I-2、电阻加热元件I-3、风扇I-4、隔热层I-5和隔板I-6，位于电子鼻仪器右中部；所述的毛细管气相色谱柱模块II包括毛细管气相色谱柱II-1、检测器II-2、放大器II-3、记录仪II-4、进样口II-5、电阻加热丝II-6、风扇II-7和隔热层II-8，位于电子鼻仪器右上部；所述的气体自动进样模块III包括：第一～第五二位二通电磁阀III-1～III-5、5个第一净化器III-6、第一微型真空泵III-7、第一流量计III-8、第六二位二通电磁阀III-9、第一节流阀III-10、二位三通电磁阀III-11、三位四通电磁阀III-12、第二微型真空泵III-13、第七二位二通电磁阀III-14、第八二位二通电磁阀III-15、稳压阀III-16、第一减压阀III-17、第二节流阀III-18、第二净化器III-19、第二减压阀III-20、第三净化器III-21、第三节流阀III-22、第二流量计III-23、第四节流阀III-24、第五节流阀III-25，位于电子鼻仪器右下部；所述的计算机控制与分析模块IV包括计算机主板IV-1、A/D数据采集卡IV-2、驱动与控制电路板IV-3、4路精密直流稳压电源IV-4、显示器IV-5、WIFI模块IV-6，位于电子鼻仪器左侧；一个生物发酵过程或一个恶臭污染点被简称为检测点；电子鼻仪器对一个检测点的被测气体进样单周期为T 0＝300-600s，默认T 0＝480s；在气体进样单周期T 0内，一个检测点的被测气体被2个微型真空泵III-7和III-13分别抽吸到气敏传感器阵列模块I和毛细管气相色谱柱模块II内，气敏传感器阵列I-1和毛细管气相色谱柱II-1产生敏感响应，电子鼻仪器因此得到1组气敏传感器阵列响应曲线和1幅气相色谱图，这是电子鼻仪器感知一个被测气体样品而得到的气敏/气相色谱模拟信号；在气体进样单周期T 0内，计算机控制与分析模块IV从气敏传感器阵列I-1时长60s的每条电压响应曲线中选择稳态峰值v gs_i(τ)、对应的出峰时间t gs_i(τ)、曲线下面积Ags _i(τ)这3个感知信息，以满足三角稳定性原理，提高气敏传感器阵列的定性定量能力；若气敏传感器阵列I-1由16个气敏传感器组成，则i＝1,2,…,16，计算机控制与分析模块IV在气体进样单周期T 0内从气敏传感器阵列16条响应曲线中共得到16*3＝48个感知分量；在气体进样单周期T 0内，电子鼻仪器不追求色谱图峰/峰完全分离，计算机控制与分析模块IV从半分离色谱图上选择前10个最大色谱峰值v gci(τ)和相对应的保留时间t gci(τ)、色谱图曲线下面积A gc(τ)，i＝1,2,…,10，共得到21个感知分量，以提高气相色谱柱的在线检测能力；在气体进样单周期T 0内，电子鼻仪器感知一个生物发酵过程或一个恶臭污染点的被测气体，计算机控制与分析模块IV将从气敏传感器阵列I-1的16条响应曲线提取的48个感知分量和从毛细管色谱柱II-1半分离色谱图提取的21个感知分量融合起来，得到一个m＝48+21＝69维的感知向量x(τ)∈R 69，称之为样本，这是电子鼻仪器对生物发酵过程或恶臭污染过程进行定性定量分析的依据；电子鼻仪器对n(≤5)个生物发酵过程或n(≤5)个恶臭监测点的气体循环进样周期为T＝nT 0，依次得到n个样本，依次存储在计算机硬盘的n个对应数据文件里，并通过WIFI路由模块将样本数据发送到云端和指定的固定/移动终端；若T 0＝480s，则被测气体循环进样周期为T＝nT 0＝n*480s，相当于一个发酵罐或一个恶臭污染点每隔n*480s被检测一次；电子鼻仪器通过对多个生物发酵过程、多个恶臭污染点经年累月的长期在线检测，形成气味大数据X的主体；数据集X还包括气相色谱、质谱、分光光度等常规分析仪器离线检测数据，专业人员实验室嗅辨得到的臭气浓度OU(odor unit)值数据，操作人员记录的青霉素、红霉素、食醋、酱油、料酒、味精等生物发酵类型数据和化工园区、垃圾填埋场、污水处理厂、畜禽养殖场等恶臭污染监测区域类型数据；数据集X的一部分子集建立了气敏/色谱响应与多个生物发酵过程/恶臭污染类型以及主要成分浓度的对应关系；在学习阶段，气味大数据X的各感知分量被施以归一化预处理，计算机控制与分析模块IV的机器学习模型离线学习气味大数据X以确定其结构和参数；在决策阶段，机器学习模型在线学习气敏-色谱近期响应以微调模型参数，依据气敏/气相近期感知时间序列阵在线确定多个生物发酵过程和恶臭污染类型，量化预测生物发酵过程发酵液主要成分浓度或国标GB14554指定的氨NH 3、硫化氢H 2S、二硫化碳CS 2、三甲胺C 3H 9N、甲硫醇CH 4S、甲硫醚C 2H 6S、二甲二硫醚C 2H 6S 2、苯乙烯C 8H 8这8种恶臭化合物和臭气浓度OU值共8+1种恶臭污染物浓度指标值。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气敏传感器阵列I-1及其环形工作腔I-2位于55±0.1℃恒温箱内；在气体进样单周期T 0内，气敏传感器阵列模块I依次经历了气敏传感器阵列初步恢复T 0-120s、洁净空气精确标定40s、平衡5s、被测气体顶空进样60s、过渡5s、环境净化空气冲洗10s共6个阶段；这6个阶段的气体类型与流量依次是：①环境净化空气6,500ml/min、②洁净空气1,000ml/min、③气体不流动、④被测气体1,000ml/min、⑤环境净化空气1,000ml/min、⑥环境净化空气6,500ml/min；“过渡”主要指从被测气体到环境净化空气的转换。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气体进样单周期T 0的[T 0-75s,T 0-15s]时间区间为气敏传感器阵列模块I的被测气体顶空进样阶段，第一～第五这5个二位二通电磁阀其中之一III-k(k＝1,2,…,5)导通，三位四通电磁阀III-12处于位置“0”，第六和第七二位二通电磁阀III-9与III-14断开、第八二位二通电磁阀III-15导通；在第一微型真空泵III-7的抽吸作用下，一个检测点的被测气体以1,000ml/min的流量依次流过第k二位二通电磁阀III-k(k＝1,2,…,5)、第八二位二通电磁阀III-15、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续60s；气敏传感器阵列I-1因此对被测气体产生敏感响应，并被存储在计算机控制与分析模块IV的临时文件里。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气体进样单周期T 0的[T 0-120s,T 0-80s]时间区间为气敏传感器阵列模块I的洁净空气标定阶段，三位四通电磁阀III-12处于位置“1”，第六、第七和第八二位二通电磁阀III-9、III-14和III-15均断开，洁净空气瓶V-2中的洁净空气以1,000ml/min的流量依次流经第一减压阀III-17、第二节流阀III-18、第二净化器III-19、三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续40s；在此期间，气敏传感器阵列I-1在洁净空气的作用下精确恢复到基准状态；由于第八二位二通电磁阀III-15断开，第一～第五这5个二位二通电磁阀III-1～III-5的导通与否不影响气敏传感器阵列I-1的标定。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，“环境净化空气”是指电子鼻仪器所处的室外空气经除尘、去湿及无菌处理后的空气，仅用于气敏传感器阵列I-1的初步恢复、环形工作腔I-2和相关气路管道内壁的冲洗、以及气敏传感器阵列累积热量的带走；在气体进样单周期T 0的[0,T 0-120s]和[T 0-10s,T 0]这两个时间段，三位四通电磁阀III-12处于位置“2”，第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开，环境净化空气以6,500ml/min的流量依次流经三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外，持续T 0-110s；在此期间，气敏传感器阵列I-1在环境净化空气作用下初步恢复到基准状态；由于第八二位二通电磁阀III-15断开，第一～第五这5个二位二通电磁阀III-1～III-5导通与否不影响气敏传感器阵列I-1的初步恢复。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，商用毛细管色谱柱II-1尺寸默认为长度L×内径φd×膜厚δ＝30m×φ0.53mm×0.25μm，位于250-300±0.1℃的恒温箱内；在气体进样单周期T 0内，毛细管气相色谱柱模块II依次经历被测气体顶空进样1s、被测气体色谱分离T 0-16s、放空与清洗吹扫15s共三个阶段；H 2兼作载气和燃气，洁净空气为助燃气；气体进样单周期T 0的最初1s是毛细管气相色谱柱模块II的被测气体顶空进样阶段，第一～第五二位二通电磁阀之一III-k(k＝1,2,…,5)导通，二位三通电磁阀III-11处于位置“1”，第七二位二通电磁阀III-14导通，第八二位二通电磁阀III-15断开；这时，检测点k的被测气体在第二微型真空泵III-13的抽吸作用下，依次流经第一～第五二位二通电磁阀之一III-k(k＝1,2,…,5)、第七二位二通电磁阀III-14、二位三通电磁阀III-11、第四节流阀III-24，在进样口II-5处与载气H 2混合，因此流入毛细管气相色谱柱II-1，持续1s；被测气体默认进样流量6ml/min，默认进样持续时间1s，默认累积进样量0.1ml。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气体进样单周期T 0的[1s,T 0-10s]时间区间为毛细管气相色谱柱模块II的被测气体分离阶段，二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开，来自检测点k的被测气体因此断开，历时T 0-11s；注入气相色谱柱模块II进样口II-5的被测气体在一定压力和流量的载气H 2的推动作用下，在毛细管气相色谱柱II-1内产生分离，检测器II-2因此产生感知，经放大器II-3放大后，记录仪II-4将[0,T 0-10s]时间区间内即色谱柱II-1时长T 0-10s的感知响应记录下来，并存储在计算机控制与分析模块IV的临时文件里。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气体进样单周期T 0的最后10s即[T 0-10s,T 0]时间区间是毛细管气相色谱柱II-1的放空亦即清洗吹扫阶段，第一～第五这5个二位二通电磁阀III-1～III-5中，原来导通的那一个即III-k断开，原来关闭的其它4个之一即III-(～k)导通；二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14导通，第八二位二通电磁阀III-15断开；假设二位二通电磁阀III-(～k)导通(k＝1,2,…,5)，在第二微型真空泵III-13的抽吸作用下，以330ml/min流量依次流经二位二通电磁阀III-(～k)、第七二位二通电磁阀III-14、二位三通电磁阀III-11，然后直接被排出到室外；这一阶段的作用是，清除相关管道在现行气体进样单周期来自第k个检测点的气味残留，并逐渐由第～k个检测点的被测气体所取代，为下一气体进样单周期检测另一个生物发酵过程或恶臭污染监测点做准备，持续10s；气体进样单周期T 0的[T 0-10s,T 0]时间区间同时为信息选择与分析时间段，计算机控制与分析模块IV从[T 0-75s,T 0-15s]时间段的气敏传感器阵列I-1电压响应曲线中选择包括稳态峰值v gsi(τ)在内的48个感知信息；从[0,T 0-10s]时间段的色谱图上选择包括前10个最大色谱峰值v gci(τ)在内的21个感知分量；这是电子鼻仪器对生物发酵过程或恶臭污染区域进行分析的依据；计算机控制与分析模块IV依据感知向量x(τ)和气味大数据X进行气味类型识别和强度与主要浓度指标值量化预测。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，在气体进样单周期T 0情况下，当只检测一个生物发酵过程或一个恶臭污染点时，则气体检测与分析周期循环是T＝T 0；若同时检测k个生物发酵过程/恶臭污染点，则其中一个生物发酵过程/恶臭污染点的循环检测与分析周期是T＝k*T 0；在长期循环监测过程中，其中一个生物发酵过程/恶臭污染点退出，则气体循环检测与分析周期变成T＝(k-1)*T 0；类似地，在长期循环监测过程中，一个新的生物发酵过程/恶臭污染点中途加入，则循环检测与分析周期变成T＝(k+1)*T 0；自一个生物发酵过程/恶臭污染点退出/加入时刻起，对应的数据文件记录周期相应变化。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，在气体进样单周期T 0内，[T 0-10s,T 0]时间区间即为时长10s的信息选择与分析时间段，计算机控制与分析模块IV对气敏传感器阵列模块I和毛细管气相色谱柱模块II二者同时进行感知信息选择与分析处理操作；计算机控制与分析模块IV从气敏传感器阵列I-1在[T 0-75s,T 0-15s]时间段即时长60s的每条电压响应曲线中选择选择稳态峰值v gs_i(τ)、对应的出峰时间t gs_i(τ)、曲线下面积Ags _i(τ)这3个感知信息分量，从毛细管气相色谱柱II-1在[0,T 0-10s]时间段即时长T 0-10s的一幅半分离色谱图上选择前10个最大色谱峰值v gc_i(τ)和10个相对应的保留时间t gc_i(τ)、1个色谱图曲线下面积A gc(τ)共21个感知响应分量，存储在计算机硬盘的临时文件里；在气体进样单周期T 0内，若时长T 0-10s的半分离色谱图的色谱峰个数q小于10，则计算机控制与分析模块IV从该半分离色谱图上选择前q<10个最大色谱峰值v gci(τ)、相对应的保留时间t gci(τ)以及色谱图曲线下面积A gc(τ)，不足的色谱峰值和对应保留时间补零，这时得到的色谱感知信息是x gc(τ)＝{(h gc1(τ),h gc2(τ),…,h gc,q(τ) ,0,…,0)；(t gc1(τ),t gc2(τ),…,t gc,q(τ) ,0,…,0)；A gc(τ)}。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气体进样单周期T 0最后10s即[T 0-10s,T 0]区间的信息处理与分析时间段，计算机控制与分析模块IV的模块化机器学习模型依据气敏/色谱近期感知时间序列矩阵X(τ-q)对生物发酵过程或恶臭污染监测点进行气味类型识别和强度与主要成分量化预测，包括：生物发酵过程类型与恶臭污染类型识别，生物发酵过程细胞浓度、底物浓度、产物浓度量化估计，正丙醇、苯乙醇等发酵过程前体物质浓度量化估计，GB14554指定的8+1种恶臭污染物浓度指标值量化预测；这里，τ为当前时间，q为近期已过去的时间，τ-q是近期时间间隔。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，气味大数据X还包括：浓度为0.1-1,0000ppm的多种单一化合物顶空挥发气的电子鼻仪器气敏/色谱感知数据，气相色谱、质谱和分光光度等常规分析仪器的离线检测数据；专业人员实验室嗅辨数据；单一化合物特别包括生物发酵过程前体物质正丙醇与苯乙酸、GB14554指定的8种恶臭化合物，以及欧洲标准EN13725指定的臭气浓度OU标准参照物—正丁醇。
- 根据权利要求1所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，机器学习模型由多个模块化深度卷积神经网络组成；单输出深度卷积神经网络模块数与被预测的生物发酵过程发酵液主要成分数、恶臭污染物主要浓度指标数、被测对象类型数相等，一一对应；一个单输出深度卷积神经网络由一个输入层、3个卷积层、2个下采样层和1个输出单元组成，各隐层与输出层活化函数均为Sigmoid修正活化函数在学习阶段，各个单输出深度卷积神经网络均采用误差反传离线逐层学习算法，主要学习气味大数据中有标签的数据和组成成分已知的气味大数据而具有必要智能；卷积层扫描窗尺寸为5×5，重叠扫描步长为1；卷积核为正弦核、余弦核、多项式核、Gaussian核、Sigmoid核、小波核和Laplace核的组合；下采样层扫描窗尺寸为2×2，不重叠扫描即步长为2，提取最大值、均值和均方差特征；在决策阶段，n个单输出深度卷积神经网络模型依据气敏/气相色谱当前时刻τ和近期已发生的时间序列感知矩阵X(τ-q)进行气味类型识别、一一估计预测当前时刻τ和未来τ+1、τ+2、τ+3时刻的气味强度与主要组成成分浓度值。
- 应用权利要求1-13所述的气敏—气相电子鼻仪器和发酵—恶臭多状态参数在线检测与分析方法，其特征是，电子鼻仪器对多个生物发酵过程/恶臭污染点长期循环在线检测和在线分析预测，包括以下步骤：(1)开机：仪器预热30min；修改屏幕菜单“气体进样单周期T 0”设置，默认值T 0＝8min；5个检测点气体循环进样周期为T＝5T 0；三位四通电磁阀III-12处于位置“2”，第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开；在第一微型真空泵III-7的抽吸作用下，环境净化空气以6,500毫升/分钟的流量依次流经三位四通电磁阀III-12，稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外；气敏传感器阵列环形工作腔I-1内部温度达到恒定的55±0.1℃；二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开，在载气H 2的推动作用下，毛细管气相色谱柱II-1逐步恢复到基准状态，色谱柱恒温箱内部温度达到恒定的250±0.1℃；(2)气体循环进样周期开始：单击显示器IV-5屏幕菜单的“检测点k开通”选项，k＝1,2,…,5，电子鼻仪器长期持续检测直至操作人员单击“检测点k断开”选项为止；电子鼻仪器依次对5个检测点进行循环检测，计算机控制与分析模块IV自动生成5个文本文件，以存储气敏传感器阵列I-1和毛细管气相色谱柱模块II对5个检测点气体的感知响应数据；(3)检测点k气体进样单周期开始；以T 0＝8min为例：(3.1)气敏传感器阵列模块I：依次历经①360s的初步恢复、②40s的精确标定、③5s的平衡、④60s的顶空进样、⑤5s的过渡和⑥10s的清洗与初步恢复共六个气体进样阶段；(3.1a)初步恢复：在气体进样单周期T 0第0-360s，三位四通电磁阀III-12处于位置“2”，第六二位二通电磁阀III-9导通，第八二位二通电磁阀III-15断开；在第一微型真空泵III-7抽吸作用下，环境净化空气以6,500ml/min的流量依次流经三位四通电磁阀III-12，稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外；气敏传感器阵列I-1初步恢复到基准状态；(3.1b)精确标定：在气体进样单周期T 0第360-400s，三位四通电磁阀III-12处于位置“1”，第六、第七和第八二位二通电磁阀III-9、III-14和III-15均断开，洁净空气以1,000ml/min的流量依次流经第一减压阀III-17、第二节流阀III-18、第二净化器III-19、三位四通电磁阀III-12、稳压阀III-16、环形工作腔I-2及其内部的气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续40s；气敏传感器阵列I-1因此精确恢复到基准状态；(3.1c)平衡：在气体进样单周期T 0第400-405s，三位四通电磁阀III-12处于位置“0”，第六和第八二位二通电磁阀III-9与III-15均断开，气敏传感器阵列环形工作腔I-2内部无气体流动，持续5s；(3.1d)顶空进样：在气体进样单周期T 0第405-465s，第一～第五这5个二位二通电磁阀其中之一III-k(k＝1,2,…,5)导通，三位四通电磁阀III-12处于位置“0”，第六和第七二位二通电磁阀III-9与III-14断开、第八二位二通电磁阀III-15导通；在第一微型真空泵III-7的抽吸作用下，一个检测点的被测气体以1,000ml/min的流量依次流过二位二通电磁阀III-k(k＝1,2,…,5)、第八二位二通电磁阀III-15、稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第一节流阀III-10、第一流量计III-8，最后被排出到室外，持续60s；气敏传感器阵列I-1因此产生敏感响应被存储在计算机控制与分析模块IV对应的临时文件里；(3.1e)过渡：在气体进样单周期T 0第465-470s，三位四通电磁阀III-12处于位置“2”，第八二位二通电磁阀III-15断开，第六和第七二位二通电磁阀III-9与III-14保持断开；在第一微型真空泵III-7抽吸作用下，环境净化空气以1,000ml/min的流量依次流经三位四通电磁阀III-12，稳压阀III-16、环形工作腔I-2及其气敏传感器阵列I-1、第六二位二通电磁阀III-9、第一流量计III-8，最后被排出到室外；(3.1f)清洗与初步恢复：在气体进样单周期T 0第470-480s，与“过渡”阶段相比，除第六二位二通电磁阀III-9由“断开”转为“导通”，其余阀位置相同；环境净化空气流量因此从“1,000ml/min”转变为“6,500ml/min”；这一阶段与即将开始的下一单周期“初步恢复”阶段的阀位置与工作状态完全相同和衔接；(3.2)毛细管气相色谱柱II模块：依次历经①1s的顶空进样、②469s的色谱分离和③10s的放空与清洗共3个气体进样阶段；(3.2a)顶空进样：在气体进样单周期T 0第0-1s，第一～第五这5个二位二通电磁阀之一III-k(k＝1,2,…,5)导通，二位三通电磁阀III-11处于位置“1”，第七二位二通电磁阀III-14导通，第八二位二通电磁阀III-15断开；在第二微型真空泵III-13的抽吸作用下，检测点k的被测气体依次流经第一～第五二位二通电磁阀之一III-k(k＝1,2,…,5)、第七二位二通电磁阀III-14、二位三通电磁阀III-11、第四节流阀III-24，在进样口II-5处与载气H 2混合，流入毛细管气相色谱柱II-1，持续1s；(3.2b)色谱分离：在气体进样单周期T 0第1-470s，二位三通电磁阀III-11处于位置“2”，第七二位二通电磁阀III-14断开；被测气体在一定压力和流量的载气H 2的推动作用下，在毛细管气相色谱柱II-1内分离，检测器II-2因此产生感知响应，经放大器II-3放大后，记录仪II-4将[0,470s]区间时长470s的感知响应记录下来，形成半分离色谱峰图，并存储在计算机控制与分析模块IV对应 的临时文件里；(3.3)信息选择与分析：在气体进样单周期T 0第470-480s，计算机控制与分析模块IV从气敏传感器阵列模块I的单个气敏传感器在[405s,465s]时间段时长60s的单条电压响应曲线中选择稳态峰值v gs_i(τ)、出峰时间t gs_i(τ)、总曲线下面积Ags _i(τ)这3个感知信息；16个气敏传感器组成的阵列I-1共得到16*3＝48个感知分量；与此同时，计算机控制与分析模块IV从毛细管色谱柱模块II时长470s的半分离色谱图上选择前10个最大色谱峰值v gc_i(τ)和对应保留时间t gc_i(τ)、总色谱图曲线下面积A gc(τ)，共得到21个感知分量；在单周期T 0内，计算机控制与分析模块IV从气敏传感器阵列I模块和毛细管色谱柱模块II的感知信息中共得到1个69维的感知向量x(τ)∈R 69；然后，机器学习模型依据感知向量x(τ)和气味大数据X进行气味类型识别和强度与主要成分量化预测，显示器显示监测和预测结果，并通过Internet网络传送到中央控制室和多个固定/移动终端；(3.4)检测点k结束与下一个检测点开始；第一～第五这5个二位二通电磁阀其中之一III-k(k＝1,2,…,5)由原来的导通转为断开，与下一个检测点对应的第一～第五二位二通电磁阀其中之一导通；(4)重复步骤(3.1)～(3.4)，电子鼻仪器实现1～5个检测点被测气体的循环在线检测、识别和气味强度与多项浓度指标值量化预测。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| US17/794,767 US20230152287A1 (en) | 2020-01-23 | 2020-07-18 | Gas-sensitive-gas-chromatographic electronic nose instrument and online analysis method of multiple state parameters of fermentation and malodorous pollutant processes |
Applications Claiming Priority (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN200010077146.1 | 2020-01-23 | ||
| CN202010077147.6 | 2020-01-23 | ||
| CN202010077146.1A CN111443159B (zh) | 2020-01-23 | 2020-01-23 | 一种气敏-气相色谱多源感知和电子鼻仪器在线检测方法 |
| CN202010077147.6A CN111443160B (zh) | 2020-01-23 | 2020-01-23 | 一种气敏-气相色谱信息融合和电子鼻仪器在线分析方法 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2021147274A1 true WO2021147274A1 (zh) | 2021-07-29 |
Family
ID=76993394
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2020/102885 WO2021147274A1 (zh) | 2020-01-23 | 2020-07-18 | 一种气敏-气相电子鼻仪器和发酵-恶臭多状态参数在线分析方法 |
Country Status (2)
| Country | Link |
|---|---|
| US (1) | US20230152287A1 (zh) |
| WO (1) | WO2021147274A1 (zh) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114839307A (zh) * | 2022-04-26 | 2022-08-02 | 华中农业大学 | 一种基于热导池阵列的气敏检测芯片及其在食品新鲜度检测中的应用 |
| CN115372515A (zh) * | 2022-08-22 | 2022-11-22 | 辽宁中医药大学 | 一种基于电子鼻快速鉴别天麻粉真伪的方法 |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SE543370C2 (sv) * | 2019-06-10 | 2020-12-22 | Reddo Floor Solutions Ab | Torkanordning för en fuktskadad golvkonstruktion |
| CN114544868B (zh) * | 2022-01-20 | 2024-03-26 | 上海工程技术大学 | 一种消除干扰气体影响的气体检测方法与*** |
| CN117434186A (zh) * | 2023-12-22 | 2024-01-23 | 中国市政工程华北设计研究总院有限公司 | 一种携污雨水复杂臭味气体组成的同步识别与量化方法 |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1367381A (zh) * | 2002-03-15 | 2002-09-04 | 上海慧比智能科技有限公司 | 一种嗅觉模拟装置及其嗅觉模拟测试方法 |
| US20020182739A1 (en) * | 1999-04-07 | 2002-12-05 | Sadik Omowunmi A. | Rapid detection of aromas using integrated gas chromatography with multiarray sensors |
| EP1376122A1 (en) * | 2002-06-28 | 2004-01-02 | Alpha M.O.S. | Sampler of volatile components |
| CN101093217A (zh) * | 2007-07-03 | 2007-12-26 | 浙江大学 | 声表面波传感器与气相色谱分离柱联用的电子鼻及检测方法 |
| CN202092981U (zh) * | 2011-04-01 | 2011-12-28 | 深圳市安鑫宝科技发展有限公司 | 基于电子鼻技术的恶臭检测仪 |
| CN103454335A (zh) * | 2013-09-09 | 2013-12-18 | 华东理工大学 | 一种多通道集成嗅觉模拟仪器和生物发酵过程在线分析方法 |
| CN103675127A (zh) * | 2013-12-02 | 2014-03-26 | 上海应用技术学院 | 一种利用顶空气相色谱-质谱联用仪和电子鼻联用对食用菌中风味物质进行辨别的方法 |
| CN105738503A (zh) * | 2016-02-01 | 2016-07-06 | 浙江大学 | 一种基于电子鼻快速预测山核桃脂肪酸含量的方法 |
| CN107085018A (zh) * | 2017-04-27 | 2017-08-22 | 浙江清华长三角研究院 | 一种用于voc及恶臭类气体检测的电子鼻*** |
| CN107748165A (zh) * | 2017-11-15 | 2018-03-02 | 江苏大学 | 基于机器视觉与电子鼻融合技术的马铃薯龙葵素检测方法 |
| CN108709955A (zh) * | 2018-05-17 | 2018-10-26 | 华东理工大学 | 一种恶臭电子鼻仪器和恶臭气体多点集中式在线监测方法 |
| CN108896706A (zh) * | 2018-05-17 | 2018-11-27 | 华东理工大学 | 大数据驱动的恶臭气体多点集中式电子鼻仪器在线分析方法 |
| CN109724645A (zh) * | 2019-01-07 | 2019-05-07 | 江苏大学 | 一种实时监测金耳发酵过程状态的便携式电子鼻*** |
| CN110632189A (zh) * | 2019-08-22 | 2019-12-31 | 广州质量监督检测研究院 | 定制家具产品气味检测装置及其方法 |
| CN110702815A (zh) * | 2019-10-29 | 2020-01-17 | 江苏大学 | 一种面粉储存品质的嗅觉传感检测方法 |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US8305582B2 (en) * | 2009-09-01 | 2012-11-06 | Alltech Associates, Inc. | Methods and apparatus for analyzing samples and collecting sample fractions |
| CN104655739A (zh) * | 2013-11-21 | 2015-05-27 | 宁夏启元药业有限公司 | 顶空气相色谱法测定红霉素发酵尾气中丙醇含量的方法 |
| US10948467B2 (en) * | 2018-05-17 | 2021-03-16 | East China University Of Science And Technology | Online centralized monitoring and analysis method for multi-point malodorous gases using electronic nose instrument |
| CN111443161B (zh) * | 2020-01-23 | 2021-01-05 | 华东理工大学 | 电子鼻气敏-色谱信息融合和风味物质现场检测分析方法 |
| CN220381063U (zh) * | 2023-06-12 | 2024-01-23 | 国投生物科技投资有限公司 | 用于酒精和二氧化碳的气相色谱分析*** |
| CN220819933U (zh) * | 2023-08-10 | 2024-04-19 | 兰州中科安泰分析科技有限责任公司 | 一种类型多样化气相色谱毛细管柱多通道可视化生产*** |
| CN117929604A (zh) * | 2024-01-18 | 2024-04-26 | 马鞍山钢铁股份有限公司 | 一种有机硫化物检测的气相色谱自动进样装置及方法 |
-
2020
- 2020-07-18 WO PCT/CN2020/102885 patent/WO2021147274A1/zh active Application Filing
- 2020-07-18 US US17/794,767 patent/US20230152287A1/en active Pending
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20020182739A1 (en) * | 1999-04-07 | 2002-12-05 | Sadik Omowunmi A. | Rapid detection of aromas using integrated gas chromatography with multiarray sensors |
| CN1367381A (zh) * | 2002-03-15 | 2002-09-04 | 上海慧比智能科技有限公司 | 一种嗅觉模拟装置及其嗅觉模拟测试方法 |
| EP1376122A1 (en) * | 2002-06-28 | 2004-01-02 | Alpha M.O.S. | Sampler of volatile components |
| CN101093217A (zh) * | 2007-07-03 | 2007-12-26 | 浙江大学 | 声表面波传感器与气相色谱分离柱联用的电子鼻及检测方法 |
| CN202092981U (zh) * | 2011-04-01 | 2011-12-28 | 深圳市安鑫宝科技发展有限公司 | 基于电子鼻技术的恶臭检测仪 |
| CN103454335A (zh) * | 2013-09-09 | 2013-12-18 | 华东理工大学 | 一种多通道集成嗅觉模拟仪器和生物发酵过程在线分析方法 |
| CN103675127A (zh) * | 2013-12-02 | 2014-03-26 | 上海应用技术学院 | 一种利用顶空气相色谱-质谱联用仪和电子鼻联用对食用菌中风味物质进行辨别的方法 |
| CN105738503A (zh) * | 2016-02-01 | 2016-07-06 | 浙江大学 | 一种基于电子鼻快速预测山核桃脂肪酸含量的方法 |
| CN107085018A (zh) * | 2017-04-27 | 2017-08-22 | 浙江清华长三角研究院 | 一种用于voc及恶臭类气体检测的电子鼻*** |
| CN107748165A (zh) * | 2017-11-15 | 2018-03-02 | 江苏大学 | 基于机器视觉与电子鼻融合技术的马铃薯龙葵素检测方法 |
| CN108709955A (zh) * | 2018-05-17 | 2018-10-26 | 华东理工大学 | 一种恶臭电子鼻仪器和恶臭气体多点集中式在线监测方法 |
| CN108896706A (zh) * | 2018-05-17 | 2018-11-27 | 华东理工大学 | 大数据驱动的恶臭气体多点集中式电子鼻仪器在线分析方法 |
| CN109724645A (zh) * | 2019-01-07 | 2019-05-07 | 江苏大学 | 一种实时监测金耳发酵过程状态的便携式电子鼻*** |
| CN110632189A (zh) * | 2019-08-22 | 2019-12-31 | 广州质量监督检测研究院 | 定制家具产品气味检测装置及其方法 |
| CN110702815A (zh) * | 2019-10-29 | 2020-01-17 | 江苏大学 | 一种面粉储存品质的嗅觉传感检测方法 |
Non-Patent Citations (2)
| Title |
|---|
| LI JUN-CHENG; ZHANG SI-XIANG; ZHOU WEI; WANG XIAO-CHEN; LI ZHI-DONG; ZHOU QI; GUO XING; LI SHUAI-FU: "Detection of Sulfur Containing Odorous Gas Based on Micro-Gas Chromatography Technology", MODERN CHEMICAL INDUSTRY, vol. 36, no. 12, 31 December 2016 (2016-12-31), CN, pages 183 - 185,187, XP009529426, ISSN: 0253-4320, DOI: 10.16606/j.cnki.issn0253-4320.2016.12.047 * |
| ZHAO, XULAN ET AL.: "Analysis of Sulfur-Containing Odorous Gas Mixture by Capillary Gas Chromatography", CHINESE JOURNAL OF ANALYSIS LABORATORY, vol. 34, no. 11, 30 November 2015 (2015-11-30), pages 1263 - 1267, XP055832195 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114839307A (zh) * | 2022-04-26 | 2022-08-02 | 华中农业大学 | 一种基于热导池阵列的气敏检测芯片及其在食品新鲜度检测中的应用 |
| CN114839307B (zh) * | 2022-04-26 | 2024-05-03 | 华中农业大学 | 一种基于热导池阵列的气敏检测芯片及其在食品新鲜度检测中的应用 |
| CN115372515A (zh) * | 2022-08-22 | 2022-11-22 | 辽宁中医药大学 | 一种基于电子鼻快速鉴别天麻粉真伪的方法 |
Also Published As
| Publication number | Publication date |
|---|---|
| US20230152287A1 (en) | 2023-05-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021147274A1 (zh) | 一种气敏-气相电子鼻仪器和发酵-恶臭多状态参数在线分析方法 | |
| WO2021147275A1 (zh) | 电子鼻仪器气敏-色谱多信息融合和风味物质现场检测与分析方法 | |
| CN111443160B (zh) | 一种气敏-气相色谱信息融合和电子鼻仪器在线分析方法 | |
| CN110057983B (zh) | 一种基于观测数据和化学机制的臭氧来源解析方法 | |
| CN108896706B (zh) | 大数据驱动的恶臭气体多点集中式电子鼻仪器在线分析方法 | |
| WO2019218395A1 (zh) | 一种恶臭气体多点集中式电子鼻仪器在线监测与分析方法 | |
| CN111443159B (zh) | 一种气敏-气相色谱多源感知和电子鼻仪器在线检测方法 | |
| Polidori et al. | Local and regional secondary organic aerosol: Insights from a year of semi-continuous carbon measurements at Pittsburgh | |
| CN105388187B (zh) | 可控湿度的半导体气敏元件的测试方法 | |
| CN102353798A (zh) | 一种面向生物发酵过程的嗅觉模拟仪器与在线分析方法 | |
| CN103454335B (zh) | 一种多通道集成嗅觉模拟仪器和生物发酵过程在线分析方法 | |
| CN102466636B (zh) | 紫外荧光法检测海洋挥发性硫化物的装置及其制作方法 | |
| CN101694461A (zh) | 一种污染源气体排放通量红外多组份监测方法与*** | |
| CN102621272A (zh) | 一种大气碳氢化合物在线分析仪 | |
| CN109781809B (zh) | 一种甲醛浓度人工智能测算方法 | |
| CN105784802A (zh) | 一种复合恶臭气体传感器检测装置 | |
| CN112730667A (zh) | 一种在线挥发性有机物走航监测***及分析方法 | |
| CN205067439U (zh) | 一种VOCs浓度在线监测装置 | |
| CN210322885U (zh) | 一种包气带土壤气快速采样分析装置 | |
| CN102778445A (zh) | 一种标态干基智能分析仪和检测方法 | |
| CN102539795B (zh) | 一种自动测定反硝化产生的气体的装置 | |
| CN204514858U (zh) | 一种非甲烷总烃气体检测仪 | |
| CN111220689B (zh) | 一种移动式多点采集快速检测粮食霉变程度的方法 | |
| CN206848240U (zh) | 一种固定污染源voc在线监测*** | |
| Kulakova et al. | Automated control and control system of emissions into the atmospheric air of thermoelectric-power industries |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 20914871 Country of ref document: EP Kind code of ref document: A1 |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 20914871 Country of ref document: EP Kind code of ref document: A1 |