Two algorithms for processing the digitized readings of electronic noses, and computer programs to implement the algorithms, have been devised in a continuing effort to increase the utility of electronic noses as means of identifying airborne compounds and measuring their concentrations. One algorithm identifies the two vapors in a two-vapor mixture and estimates the concentration of each vapor (in principle, this algorithm could be extended to more than two vapors). The other algorithm identifies a single vapor and estimates its concentration.
An electronic nose consists of an array of sensors, all of which respond to a variety of chemicals. By design, each sensor is unique in its responses to these chemicals: some or all of the sensitivities of a given sensor to the various vapors differ from the corresponding sensitivities of another sensor. The two algorithms exploit these sensitivities and the differences among them.
The validity of the two-vapor algorithm depends on the validity of the assumption that, of all the vapors of interest, no more than two of them are present at the time of measurement. This algorithm utilizes the following mathematical model of the response of a given sensor to a given pair of vapors:
z = A + (BxC + DyE)F,
where z is the sensor response, x and y are the concentrations of the two vapors, and parameters A through F are obtained by least-squares best fit of sensor responses to known concentrations of the individual vapors and to known concentrations of mixtures of the two vapors. The reason for choosing this model is that this research has shown it to be the best for mixtures of vapors. The model equation defines a response surface of the given sensor for the given pair of vapors.
Given the responses of an electronic nose to an unknown single vapor or two-vapor mixture, the first step of this algorithm is to calculate the difference between (1) the actual response of each sensor and (2) the model response of the sensor for an assumed pair of vapors. This calculation yields an error surface for the given sensor for the given two vapors. Next, the error surfaces thus calculated for all the sensors in the array are combined to obtain an error surface for the electronic nose with respect for the assumed two vapors. Next, the process as described thus far is performed for a different pair of vapors. The process is repeated until error surfaces for all possible pairs of vapors have been calculated.
It is necessary to find the minimum point on the electronic-nose error surface for each pair of vapors. In the present version of the algorithm, this is done by sampling values on a grid and selecting the sample that has the minimum value. In a subsequent enhanced version of the algorithm, a more sophisticated technique (e.g., gradient descent) might be used to find the minimum. The pair of vapors for which the electronic-nose error surface has the lowest minimum value is deemed to be identified as the vapor pair sensed by the electronic nose. Provided that this identification is correct, the concentrations of the two vapors are the coordinates of the location of the minimum on the error surface for that pair.
The validity of the single-vapor algorithm depends on the validity of the assumption that, of all the vapors of interest, only one is present at the time of measurement. This algorithm utilizes the following mathematical model of the response of a given sensor to a single vapor:
z = A(1 - eBx),
where z is the sensor response, x is the concentration of the vapor, and parameters A and B are obtained by leasts squares best fit of sensor responses at known values of x. This model is appropriate because it gives both the expected zero response at zero concentration and saturation response at high concentration.
The first step of the single-vapor algorithm is to identify the vapor by applying standard statistical pattern recognition techniques to the responses of the electronic nose. Assuming that the vapor has been correctly identified, one could, in principle, estimate the concentration by applying the inverse of the model to the responses of all sensors in the nose. The question is how best to utilize the readings of all the sensors in the nose to obtain the best estimate. Research has answered the question: the best estimate is obtained by inverting the reading of a single sensor known to be best for the vapor that has been identified. Accordingly, the algorithm chooses the sensor found to be best for the identified vapor and calculates the concentration from the reading of that sensor.
This work was done by Rebecca Young of Kennedy Space Center and Bruce Linnell and Barbara Peterson of ASRC Aerospace. For more information, download the Technical Support Package (free white paper) at www.techbriefs.com/tsp under the Information Sciences category. KSC-12725/20
