Smart sensor system and method using a surface acoustic wave vapor sensor array and pattern recognition for selective trace organic vapor detection

• US 5,469,369 A
• Filed: 11/02/1992
• Issued: 11/21/1995
• Est. Priority Date: 11/02/1992
• Status: Expired due to Term

• Current Assignee: the united states of america as represented by the secretary of the navy
• Original Assignee: the united states of america as represented by the secretary of the navy
• Inventors: Di Lella, Daniel, Grate, Jay W., Rose-Pehrsson, Susan L.
• Primary Examiner(s): Ramirez, Ellis B.
• Assistant Examiner(s): ASSOUAD, PATRICK J
• Application Number:
  US07/970,750
• Time in Patent Office: 1,114 Days
• Field of Search: 364/496, 364/497, 364/498, 364/499, 340/632, 73/23.2
• US Class Current: 702/27
• CPC Class Codes: G01N 2291/0215 Mixtures of three or more gases, e.g. air
• CPC Class Codes: G01N 2291/0256 Adsorption, desorption, surface mass change, e.g. on biosensors
• CPC Class Codes: G01N 2291/0423 Surface waves, e.g. Rayleigh waves, Love waves
• CPC Class Codes: G01N 2291/0427 Flexural waves, plate waves, e.g. Lamb waves, tuning fork, cantilever
• CPC Class Codes: G01N 29/022 Fluid sensors based on microsensors, e.g. quartz crystal-microbalance [QCM], surface acoustic wave [SAW] devices, tuning forks, cantilevers, flexural plate wave [FPW] devices microdevices per se B81B
• CPC Class Codes: G01N 33/0031 comprising two or more sensors, e.g. a sensor array electrochemical electrode arrays G01N27/27
• CPC Class Codes: G01N 33/0034 comprising neural networks or related mathematical techniques
• CPC Class Codes: G01N 33/0057 for warfare agents or explosives properties of explosives G01N33/227

Abstract

A method and a system using that method are provided which employ a patternecognition algorithm to improve sensitivity in detecting hazardous vapors. The algorithm enables the discrimination of vapors of interest from non-hazardous substances at higher concentrations in varying relative humidity. A weight vector is generated corresponding to a N-space representation of a class comprising known vapors of interest, and a N-space representation of the unknown vapor is used to generate an unknown pattern vector. By calculating the dot product of the unknown pattern vector and the weight vector a determination can be made as to whether the unknown vapor is within the class. The weight vector is generated by selecting a training set comprising a subset of the known vapors of interest and background vapors and generating an N-space representation of the training set so as to create an associated weight vector.

Claims

1. A method of identifying an unknown vapor as either belonging to a class or not, wherein the class comprises known vapors of interest, said method comprising the steps of:

introducing a sample of an unknown vapor into at least one sensor coated with a vapor-sensitive coating,

generating a weight vector corresponding to a N-space representation of the class;

generating a N-space representation of the unknown vapor and generating an unknown pattern vector based thereon; and

calculating the dot product of said unknown pattern vector and the weight vector to determine whether the unknown vapor is within said class, wherein the dot product producing a positive number is a member of the class in question while a negative number is not.

2. The method recited in claim 1, wherein said step of generating a weight vector comprises the steps of:

selecting a training set comprising a subset of the known vapors of interest and background vapors;

generating an N-space representation of said training set so as to create an associated weight vector; and

storing said weight vector in memory for use in calculating said dot product.

3. The method recited in claim 2, wherein said step of generating an N-space representation of said training set further comprises the steps of:

generating vapor response data from a sensor array comprising at least two sensors;

interpreting a pattern of responses as a point in an N-dimensional feature space having N axes and defining a pattern vector extending from the origin of said axes to said point, where N is defined by the number of sensors in said sensor array; and

using a supervised learning technique to generate a discriminant function which defines a weight vector, a first cluster of points corresponding to the known vapors, and a second cluster of points corresponding to the background vapors.

4. The method recited in claim 3, wherein said step of using supervised learning techniques to generate a discriminant function further comprises the step of:

using a Bayes linear equation to iteratively generate said weight vector.

5. The method recited in claim 4 further comprising the step of

comparing results of said dot product with a vector associated with each known pattern result to determine if a correction factor is required;

and

iteratively repeating the last step until all results agree with known results.

6. The method recited in claim 3, wherein said step of using a supervised learning technique to generate a discriminant further comprises the step of:

using a nonparametric equation selected from the group consisting of linear learning machines, adaptive least squares classifiers and combinations thereof, to iteratively generate said weight vector.

7. The method recited in claim 3, wherein said step of using a supervised learning technique to generate a discriminant further comprises the step of:

using a parametric equation to iteratively generate said weight vector; and

passing said weight vector to at least one further supervised learning technique to increase the accuracy of said weight vector.

8. The method recited in claim 1, wherein said step of generating a N-space representation of the unknown vapor comprises the steps of:

generating unknown vapor response data from a sensor array comprising at least two sensors; and

interpreting a pattern of said unknown response as a point in an N-dimensional feature space having N axes and defining an unknown pattern vector extending from the origin of said axes to said point, where N is defined by the number of sensors in said sensor array.

9. A smart sensor system for determining whether an unknown vapor is within a known class of vapors, said system comprising:

a sampling means for generating vapor samples;

an array of sensors for generating vapor response data from said vapor samples;

means for interpreting said response data as points in an N-dimensional feature space having N axes and defining unknown pattern vectors, each vector being associated with one of said points and extending from the origin of said axes to said point, where N is defined by the number of sensors in said sensor array; and

means for calculating the dot product of said unknown pattern vector and a stored weighing vector associated with a N-space representation of the known class, and to determine if said vapor is within said class, said weight vector being defined by a discriminant function generated by a supervised learning technique.

10. The system recited in claim 9 wherein each of said sensors comprises a transducer having vapor a sensitive coating disposed thereon for providing sorbing of said vapors to thereby increase the mass and decrease the modulus of said coating.

11. The system recited in claim 10 wherein said vapor sensitive coating is selected from a group consisting of poly(ethylenimine), fluoropolyol, ethyl cellulose, poly(epichlorohydrin), poly(isobutylene), poly(ethylene maleate), poly(ethylene phthalate), and combinations thereof.

12. The system recited in claim 9 further comprising a temperature control means for maintaining said sensors at a predetermined temperature.