A metal oxide-based E-nose employing backpropagation artificial neural network for the quantitative recognition of common fuel gases
The safe use of fuel gases in industrial production and daily life requires prompt and accurate detection of their concentrations in the environment. However, due to the strong reducing properties of the flammable gases used as fuel, traditional chemoresistive gas sensors face challenges in simultaneously detecting multiple flammable gases. To address this issue, we have developed a metal oxide-based electronic nose capable of detecting common fuel gases, including CH 4 , H 2 , and CO. By controlling the composition and structure of metal oxides, as well as introducing noble metal modifications, we optimized the response differences of these gas sensitive materials to CH 4 , H 2 , and CO. Four sensitive materials were selected from 8 candidates using principal component analysis to construct gas sensor array on a compact four-channel circuit board. The response sensitivities from the sensor array to various concentrations of CH 4 , H 2 , CO, and their mixtures were utilized to train a backpropagation artificial neural network. A new component-identification outlet layer was integrated with the traditional concentration-recognition to improve the training speed and enable simultaneous identification of gas components and concentrations. Through combined material optimization and dual-output training, the backpropagation artificial neural network achieved 90% training accuracy within just 50 iterations, with both sparse cross-entropy loss and mean squared error reduced below 0.005 after 300 training epochs. The resulting E -nose can accurately identify gas components and determine their concentrations in single-, binary-, and ternary-component mixtures of CH 4 , H 2 , and CO, with detection accuracies exceeding 91%. This study provides a practical approach for E -nose construction, enabling reliable monitoring of flammable gases.
危险品化学品经营许可证（带存储）