부산대학교 도서관

Sensor devices can be made smarter, faster and highly selective if multiple technologies converge together; such as device simulation, novel material synthesis, device engineering and smarter machine learning algorithms. In this context, sensors for high-energy material, namely HMX (High Melting explosives), which are hazardous moiety and which cannot be much experimented for detection studies, are explored. The molecular similarity of HMX with other equally hazardous molecules, which all possesses active NOx groups, makes its detection and differentiation further challenging. In this context, resonator-based sensor operating at 433 MHz is used to derive power and frequency shifts experimentally. Machine learning algorithm, k-nearest neighbours (k-NN) is used for classification with overall accuracy of 92.9 % in the range of 400-1000 ppm. For class-wise detection of HMX, along with TNT (Trinitrotoluene), RDX (Research Department explosive), Dinitrobenzene, and Nitrobenzene, the accuracy obtained was > 98.25%. The data-driven technique based on statistical collection of sensor data and use of artificial intelligence (AI) to unlock performance predictions of the sensor defines high accuracy and selectivity of the developed device.