Chemical Identification with Chemical Sensor Arrays

An electronic nose composed of small, inexpensive chemical sensors would have many potential applications in areas such as breath analysis, hazardous gas detection and industrial process monitoring. An electronic nose, in anology to a biological nose, consists of an array of non-specific chemical sensors. With pattern recognition, the system can recognize individual chemicals or scents. Capacitive micromachined ultrasonic transducer (CMUT) chemical sensors are a particularly promising chemical sensor technology. They are micromachined devices, so they can be inexpensively batch fabricated. In addition, they have extraordinary sensitivity; A limit of detection of 50.5 parts-pertrillion (ppt) and sensitivity of 34.5 ppt/Hz to DMMP, a sarin gas simulant have been demonstrated [2]. In this work, I demonstrate the application of machine learning to an array of four CMUT chemical sensors to discriminate between six different chemicals present in low concentrations in nitrogen. Typically, only the equilibrium response of chemical sensors is used in chemical identification. I show that using the response of the sensor over time provides extra information that improves classification performance. I also study the robustness of the system with respect to variations in sensor sensitivity, and test its ability to determine the concentration of the detected chemical once it is identified. The data used in this work has been previously been shown at a conference, with some machine learning tested on it [8], but the feature selection and machine learning work in this paper is new, and shows significant improvement over the previous results.