alcohol sensor1
Content
ensors an ens ng ys ems or Machin Mac hine e Olfact Olfaction ion – Ele Electr ctroni onicc Nose and Electronic Tongue Dr. Nabarun Bhattachar a C-DAC, Kolkata . October 29, 2009
Presentation Structure Human Olfaction Machine Olfaction Electronic Nose Design Details and Results
Electronic Tongue Desi De si n Det Detai ails ls an and d Res Resul ults ts
Roadmap for Research in Machine Olfaction Conclusion
.
Human Olfaction The olfactory region is located in the roof of the two nasal cavities when compounds (called odorants)) that are carried by odorants inhaled air stimulate receptors located in the olfactory epithelium. The mucous lipid , which is pro uce n e o ac ory epithelium,, assists in epithelium transporting the odorant molecules. Only volatile materials that are soluble in the mucous can interact with the olfactory recep ors an pro uces e signals that our brain interprets as odor.
Machine Olfaction Attempts to mimic mimic human senses senses of smell and taste by electronic means are called machine Olfaction. Sensors are the most crucial components in a machine olfaction system. Signal conditioning, data acquisition, data, data processing and pattern recognition are the crucial modules of an olfactory sensing .
Machine Olfaction System
Hum H uman an Pe Perc rcep epti tion on Eye:
V IS IO N
Machine Ma chine Sensing V I SIO N :C a m e r a
ar: S k in :
: T A C T IL E SENSES
c ro p o n e
TOUCH: Tactile
Devices -
Tongue:TASTE
TASTE: TAS TE: E-T E-Tongu ongue e
About Electronic Nose Electronic Nose senses complex odours using an Array of Sensors (called “sensor array”): each tuned for o our o a am y o vo at e compoun s. Odour stimulus imprints a characteristic electronic . This smell print is statistically classified and resolved with suitable attern reco nition en ine as a measurement of odour of the sample. In short, Electronic Nose is “A scientific, reliable, repeatable, physical, non-invasive, affordable real-time techniques for various applications like food quality assessment, environmental polution detection, medical applications, explosive detection etc.
Basic Block Diagram Odour Delivery System
Sensor Array
Classification
ent cat on
Signal Conditioning
Data Acquisition
our
an
ng
e very
Headspace Sampling Autosampling Stage Air
S1 Mass Flow Controller
S2
Solenoid Valves
S3
Sensor Cell
Bubbler System
Measurement Circuit
Syringe Needles Liquid Sample
Temperature Controlled Bath
ensors
Desirable Properties of Selectivity : Must respond to a range of chemical
species. Sensitivity : Should be sensitive to detect vapour
. Speed of Response : Response time should be in . Reproducibility : Sensors response characteristics should be reproducible. Reversibility : Should be able to recover immediately after exposure to gas. Portability : Should be small so that less sample volume may be used.
Interface Circuit Diagram
Each sesnsor is a MOS sensor made from a metal oxide film, e.g., Tin Oxide Volatiles under o redox reactions at the sensor surface, resulting in a chan e of conductivit across the sensor Each sensor is reversible.
Vc
V RL
RL
VH
GND
Measurement Circuit with MOS Sensor
within TTL range.
Signal Conditioning The output of the sensors is analogue voltage. stages,namely,buffering,amplification,filtering,co the USB card used in the system for data acquisition. No additional electronics has been used for this purpose
Data Acquisition Circuit USB 6009 card from the National Instruments has been used. DAQ system consists of sample and , to digital conversion module. Sample rate :250 Ksamples/second
Signal Pre-processing Steps of Signal Pre-processing: Baseline identification and mani ulation Compression
Baseline Handling Baseline refers to sensor response in no exposure condition Fractional techni ue of baseline manipulation is used for compensation a ainst drift and contrast enhancement.
y s t =
x s t − x s 0 x 0
Compression Technique Compression is a preprocessing stage where the response of sensor array is uti ize as a eature vector or a fingerprint by reducing the number of escriptors. The maximum value vector from the sensor output data has only been considered for data analysis. =
[S i1 max ..............S i8 max ]
Nose…
Olfaction Software
Software features: -Programmable Sequence Control -Dynamic Fermentation Profile Display -Data Logging -Alarm Annunciation - ex y o perm ea p an ers emse ves o ra n an the system as per their requirements.
cus om ze
a a na ys s
About Sensor Array Output .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
⎢ . ⎢ ⎢ b h1 A = ⎢ S 11
.
.
.
.
.⎥
bh2
.
.
.
S 12
.
.
.
b h8 ⎥ S 18 ⎥
S 22
.
.
.
S 28 ⎥
.
.
.
.
.⎥
.
.
.
.
.⎥
.
.
.
.
.⎥
⎢ b 21 ⎢ ⎢ . ⎢
⎢ S 21 ⎢ . ⎢ ⎢ . ⎢ . ⎢ ⎢⎣ S m 1
b 22
S m 2
.
.
.
⎥ b 28 ⎥ .⎥ ⎥
Sensor responses during headspace generation
.
⎥
⎥ ⎥
S m 8 ⎥⎦
Sensor responses when exposed to tea odour during sampling
Data is 8-dimensional Headspace Duration : 30 Seconds and Sampling Duration : 50 seconds 10 rea ings are scanne per secon Approximately 800 rows are there in any sniffing data matrix
Data Analysis Strategy MULTIVARIATE DATA
DATA EXPLORATION
PRINCIPAL COMPONENT ANALYSIS (PCA)
DATA QUANTIFICATION
AROMA SCORE CALCULATION BY 2NORM METHOD
DATA CO-RELATION
AROMA SCORE CALCULATION BY MAHALANOBIS DISTANCE METHOD
BACK PROPAGATION
ARTIFICIAL NEURAL NETWORK
RADIAL BASIS FUNCTION
PROBALISTIC NEURAL NETWORK
Results – Different Clones Well-defined clusters are found in PCA. 100% classification accurac observed in BP-MLP
Electronic Tongue
Tea Liquor
Array of
Wate r
Current Status vis-à-vis Work Plan
11th Plan Project proposed
Pilot Level Deployment
NTRF Funding
Future Scope of Research Hybrid sensor array consisting of MOS, CP Development of new sensor array sensitive Development of more efficient algorithms or e er c us er ng an c ass ca on Techniques for drift compensation Integration of E-Nose with E-Tongue and E-Vision s stems --- ENTV S stem
Presentation Structure Human Olfaction Machine Olfaction Electronic Nose Design Details and Results
Electronic Tongue Desi De si n Det Detai ails ls an and d Res Resul ults ts
Roadmap for Research in Machine Olfaction Conclusion
.
Human Olfaction The olfactory region is located in the roof of the two nasal cavities when compounds (called odorants)) that are carried by odorants inhaled air stimulate receptors located in the olfactory epithelium. The mucous lipid , which is pro uce n e o ac ory epithelium,, assists in epithelium transporting the odorant molecules. Only volatile materials that are soluble in the mucous can interact with the olfactory recep ors an pro uces e signals that our brain interprets as odor.
Machine Olfaction Attempts to mimic mimic human senses senses of smell and taste by electronic means are called machine Olfaction. Sensors are the most crucial components in a machine olfaction system. Signal conditioning, data acquisition, data, data processing and pattern recognition are the crucial modules of an olfactory sensing .
Machine Olfaction System
Hum H uman an Pe Perc rcep epti tion on Eye:
V IS IO N
Machine Ma chine Sensing V I SIO N :C a m e r a
ar: S k in :
: T A C T IL E SENSES
c ro p o n e
TOUCH: Tactile
Devices -
Tongue:TASTE
TASTE: TAS TE: E-T E-Tongu ongue e
About Electronic Nose Electronic Nose senses complex odours using an Array of Sensors (called “sensor array”): each tuned for o our o a am y o vo at e compoun s. Odour stimulus imprints a characteristic electronic . This smell print is statistically classified and resolved with suitable attern reco nition en ine as a measurement of odour of the sample. In short, Electronic Nose is “A scientific, reliable, repeatable, physical, non-invasive, affordable real-time techniques for various applications like food quality assessment, environmental polution detection, medical applications, explosive detection etc.
Basic Block Diagram Odour Delivery System
Sensor Array
Classification
ent cat on
Signal Conditioning
Data Acquisition
our
an
ng
e very
Headspace Sampling Autosampling Stage Air
S1 Mass Flow Controller
S2
Solenoid Valves
S3
Sensor Cell
Bubbler System
Measurement Circuit
Syringe Needles Liquid Sample
Temperature Controlled Bath
ensors
Desirable Properties of Selectivity : Must respond to a range of chemical
species. Sensitivity : Should be sensitive to detect vapour
. Speed of Response : Response time should be in . Reproducibility : Sensors response characteristics should be reproducible. Reversibility : Should be able to recover immediately after exposure to gas. Portability : Should be small so that less sample volume may be used.
Interface Circuit Diagram
Each sesnsor is a MOS sensor made from a metal oxide film, e.g., Tin Oxide Volatiles under o redox reactions at the sensor surface, resulting in a chan e of conductivit across the sensor Each sensor is reversible.
Vc
V RL
RL
VH
GND
Measurement Circuit with MOS Sensor
within TTL range.
Signal Conditioning The output of the sensors is analogue voltage. stages,namely,buffering,amplification,filtering,co the USB card used in the system for data acquisition. No additional electronics has been used for this purpose
Data Acquisition Circuit USB 6009 card from the National Instruments has been used. DAQ system consists of sample and , to digital conversion module. Sample rate :250 Ksamples/second
Signal Pre-processing Steps of Signal Pre-processing: Baseline identification and mani ulation Compression
Baseline Handling Baseline refers to sensor response in no exposure condition Fractional techni ue of baseline manipulation is used for compensation a ainst drift and contrast enhancement.
y s t =
x s t − x s 0 x 0
Compression Technique Compression is a preprocessing stage where the response of sensor array is uti ize as a eature vector or a fingerprint by reducing the number of escriptors. The maximum value vector from the sensor output data has only been considered for data analysis. =
[S i1 max ..............S i8 max ]
Nose…
Olfaction Software
Software features: -Programmable Sequence Control -Dynamic Fermentation Profile Display -Data Logging -Alarm Annunciation - ex y o perm ea p an ers emse ves o ra n an the system as per their requirements.
cus om ze
a a na ys s
About Sensor Array Output .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
⎢ . ⎢ ⎢ b h1 A = ⎢ S 11
.
.
.
.
.⎥
bh2
.
.
.
S 12
.
.
.
b h8 ⎥ S 18 ⎥
S 22
.
.
.
S 28 ⎥
.
.
.
.
.⎥
.
.
.
.
.⎥
.
.
.
.
.⎥
⎢ b 21 ⎢ ⎢ . ⎢
⎢ S 21 ⎢ . ⎢ ⎢ . ⎢ . ⎢ ⎢⎣ S m 1
b 22
S m 2
.
.
.
⎥ b 28 ⎥ .⎥ ⎥
Sensor responses during headspace generation
.
⎥
⎥ ⎥
S m 8 ⎥⎦
Sensor responses when exposed to tea odour during sampling
Data is 8-dimensional Headspace Duration : 30 Seconds and Sampling Duration : 50 seconds 10 rea ings are scanne per secon Approximately 800 rows are there in any sniffing data matrix
Data Analysis Strategy MULTIVARIATE DATA
DATA EXPLORATION
PRINCIPAL COMPONENT ANALYSIS (PCA)
DATA QUANTIFICATION
AROMA SCORE CALCULATION BY 2NORM METHOD
DATA CO-RELATION
AROMA SCORE CALCULATION BY MAHALANOBIS DISTANCE METHOD
BACK PROPAGATION
ARTIFICIAL NEURAL NETWORK
RADIAL BASIS FUNCTION
PROBALISTIC NEURAL NETWORK
Results – Different Clones Well-defined clusters are found in PCA. 100% classification accurac observed in BP-MLP
Electronic Tongue
Tea Liquor
Array of
Wate r
Current Status vis-à-vis Work Plan
11th Plan Project proposed
Pilot Level Deployment
NTRF Funding
Future Scope of Research Hybrid sensor array consisting of MOS, CP Development of new sensor array sensitive Development of more efficient algorithms or e er c us er ng an c ass ca on Techniques for drift compensation Integration of E-Nose with E-Tongue and E-Vision s stems --- ENTV S stem
