Active gas/odor sensing system using automatically controlled gas blender and numerical optimization technique

Abstract

As measurement of a vapor mixture composition is a difficult technique, no method using a sensing system has yet been established in spite of great effort by many researchers. In this paper, the authors propose a new gas/odor sensing system using a gas blender and a nonlinear numerical optimization algorithm by which the concentration of each component in an unknown vapor can be quantified. The component vapors are internally blended and the mixture ratio is modified by the system so that the sensor array output pattern of the blended vapor can be made equal to that of the unknown one. After several iterations, convergence is obtained and the vapor concentration of each component is determined from the mixture composition of the blended vapor. Although the conventional system is passive, this system is considered as an active one as it performs exploratory behavior prior to recognition. Here, gasoline vapor concentration is measured under the condition that one or two interference vapors exist together. Gasoline vapor has been adopted as an example of odors in the passenger compartment of a car, since it sometimes smells unpleasant. The measurement is essential for designing a car in order to keep it comfortable for passengers. The sensors used here are three semiconductor gas sensors and two electrochemical sensors, which are chosen in order to obtain high sensitivity to gasoline. The nonlinear numerical optimization techniques used are the simplex method and the gradient descent method and these two methods are compared here. It is found that the quantification error is within ten ppm for two- or three-component vapors.