基于ESP32-S的小型智能气体识别系统设计

Design of small intelligent gas identification system based on ESP32-S

摘要:

智能气体识别系统主要包括硬件电路、用户界面、信号处理和深度神经网络模型各模块。其采用 ESP32-S作为主控芯片，设计硬件电路对金属氧化物半导体气体传感器加热端电压动态调制，实验选择调制传感器温度范围为150~250 ℃，对应调制电压范围2.5~3.8 V。以20 Hz的频率采集热调制下的传感器特性变化曲线，通过信号传输功能传递至上位机，将所得热调制响应信号经信号处理特征提取后喂入自行搭建的I-5×5-2×2-200-200-P结构卷积神经网络训练构建模型，使用训练后深度神经网络模型前向传播实现对目标气体的快速识别检测。在静态气敏测试平台中的测试结果显示，该系统能有效实现对多种类VOC气体的识别功能，对乙醇、丙酮和异丙醇的42种测试气氛的识别准确率达到100%。

Abstract:

The intelligent gas identification system mainly includes hardware circuit, user interface, signal processing and deep neural network model modules. The ESP32-S is used as the main control chip, and the hardware circuit is designed to dynamically modulate the voltage at the heating end of the metal oxide semiconductor gas sensor. The modulated temperature range of the sensor is 150~250 ℃, and the corresponding modulation voltage range is 2.5~ 3.8V in the experiment. The sensor characteristic change curve obtained by thermal modulation is collected at a frequency of 20 Hz and transmitted to the upper computer through the signal transmission function, and the obtained thermal modulation response signal is fed into a self-built convolutional neural network with I-5×5-2×2-200-200-P structure to build a model after signal processing feature extraction. Fast identification detection of target gases is achieved using forward propagation of trained neural network model parameters. Test results in a static gas-sensitive test platform show that the system can effectively achieve the identification function for multiple types of VOC gases, and the recognition accuracy of 42 kinds of test atmospheres of ethanol, acetone and isopropyl alcohol reaches 100%.