Lee Byeongju, Kang Mingu, Lee Kichul, Chae Yujeong, Yoon Kuk-Jin, Lee Dae-Sik, Park Inkyu
Korea Advanced Institute of Science and Technology, Daejeon 34141, Republic of Korea.
Electronics and Telecommunications Research Institute (ETRI), Daejeon 34129, Republic of Korea.
ACS Sens. 2025 Feb 28;10(2):954-964. doi: 10.1021/acssensors.4c02715. Epub 2025 Jan 20.
Semiconductor metal oxide (SMO) gas sensors are gaining prominence owing to their high sensitivity, rapid response, and cost-effectiveness. These sensors detect changes in resistance resulting from oxidation-reduction reactions with target gases, responding to a variety of gases simultaneously. However, their inherent limitations lie in selectivity. Despite attempts to address this through new sensing materials and filters, achieving perfect selectivity remains challenging. This study addresses the selectivity issue by implementing temperature-modulated operation of a single SMO gas sensor utilizing an anodic aluminum oxide (AAO) microheater platform. The AAO-based sensor ensures a high thermal and mechanical stability during prolonged temperature modulation. A staircase waveform featuring six temperature conditions was applied to the microheater platform, and gas response data were collected for acetone, ammonia, ethanol, and nitrogen dioxide. Leveraging a convolutional neural network (CNN), gas patterns were trained and used to predict gas types and concentrations. The results demonstrated a high classification accuracy of 97.0%, with mean absolute percentage errors (MAPE) for concentration estimation of acetone, ammonia, ethanol, and nitrogen dioxide at 13.7, 19.2, 19.8, and 19.4%, respectively. The proposed method effectively classified four spices and accurately distinguished similar odors, which are difficult for human olfaction to differentiate. The results highlight that the combination of temperature modulation and deep learning algorithms proves to be highly effective in precisely determining gas types and concentrations.
半导体金属氧化物（SMO）气体传感器因其高灵敏度、快速响应和成本效益而日益受到关注。这些传感器检测与目标气体发生氧化还原反应所导致的电阻变化，能够同时对多种气体作出响应。然而，它们的固有局限性在于选择性。尽管人们尝试通过新型传感材料和过滤器来解决这一问题，但实现完美的选择性仍然具有挑战性。本研究通过利用阳极氧化铝（AAO）微加热器平台对单个SMO气体传感器进行温度调制操作来解决选择性问题。基于AAO的传感器在长时间温度调制过程中确保了高的热稳定性和机械稳定性。将具有六个温度条件的阶梯波形应用于微加热器平台，并收集了丙酮、氨、乙醇和二氧化氮的气体响应数据。利用卷积神经网络（CNN）对气体模式进行训练，并用于预测气体类型和浓度。结果表明分类准确率高达97.0%，丙酮、氨、乙醇和二氧化氮浓度估计的平均绝对百分比误差（MAPE）分别为13.7%、19.2%、19.8%和19.4%。所提出的方法有效地对四种香料进行了分类，并准确地区分了人类嗅觉难以区分的相似气味。结果突出表明，温度调制和深度学习算法的结合在精确确定气体类型和浓度方面被证明是非常有效的。
