Gas Identification Method Based on Transfer Function Ratios Toward Artificial Olfaction

Abstract

Olfaction is the only human sense that has not been realized as a practical sensor. To realize practical artificial olfaction, it is essential to establish a simple measurement protocol that people without expertise can follow. In this study, a novel data analysis method for odor identification is proposed. The key concept of the analysis method is transfer function ratios (TFRs), which are intrinsic to the combination of sensors and odors. Based on TFRs, one can identify odors without controlling or monitoring gas flow because TFRs are independent of gas flow and can be calculated only from sensing signals. To demonstrate the feasibility of the analysis method based on TFRs, odor identification through the free-hand measurement was performed; the odor of a sample was measured by manually moving a small sensor chip—membrane-type surface stress sensors (MSS)—near the sample (Fig. 1a). TFRs were calculated from the measurement data obtained through the free-hand measurements on three spices: rosemary, garlic and red chili pepper. Principal component analysis (PCA) was performed on the dataset of the TFRs, resulting in the formation of three clusters corresponding to the spices. Machine learning models were developed from the dataset of TFRs. By using random forest as a classifier, the odors can be identified with an accuracy of 0.89±0.04. These results indicate that the data analysis method based on TFRs provides the easy measurement protocol for odor identification, leading to the realization of practical application of artificial olfaction. Figure 1