【作者】 张宇；

【导师】 段国韬；

【作者基本信息】 华中科技大学 ， 电子信息（专业学位）， 2024， 硕士

【摘要】 金属氧化物半导体气体传感器因其快响应、高灵敏、低成本和长寿命等特点被广泛采用。半导体气体传感器通常采用在恒定工作温度下检测测试电极阻值变化的工作方式。该方法能瞬时感知环境气体信息变化,有利于快速响应危险气体的泄漏;但其测量参数少,存在多种气体交叉干扰问题,不利于复杂环境中目标气体的选择性识别。针对这些问题,本论文以硫化氢为目标气体,研究传感器工作温度变化对敏感材料电学特性的影响,探究不同加热模式与气敏动态响应的关联性,提出了一种半导体传感器变温气敏测量方法,利用其动态响应特征高精准识别目标气体,进而对其实际应用开展了基础研究。取得的主要研究结果如下:(1)设计并制作了一套气体传感器测试系统,基于市售传感器,通过调控加热区间、加热周期、加热占空比三个参数探究了27种不同矩形波变温加热模式下的气敏动态响应,利用动态响应曲线分析设计出对硫化氢具有特征峰值高、加热周期短以及响应特征显著的传感器变温气敏测量方法和工作模式。(2)基于气敏动态响应数据设计了对气体种类和浓度具有强关联的12个时域特征参数,并提出了一种基于气体传感器动态响应的皮尔逊相关系数特征提取和高精度气体识别方法,该方法利用皮尔逊相关系数组合出不同维度的特征向量;针对不同维度的特征向量,利用K近邻、支持向量机和人工神经网络三种常用算法分析了单一气体识别和浓度预测的准确性,利用反向传播神经网络算法分析了双组分混合气体成分预测的准确性,通过结果比对优化出针对硫化氢高准确识别的特征向量算法组合,混合气体下硫化氢识别准确率可达到99%,浓度预测误差为0.834 ppm。(3)基于传感器变温气敏测量方法开发了以ESP32单片机为控制和信号处理中心的硫化氢分析系统,并利用硫化氢高准确识别特征向量算法组合和TensorFlow Lite框架将已训练人工神经网络模型压缩部署至系统微控制器终端,成功实现了模拟应用环境中硫化氢高精度识别和浓度预测。更多还原

【Abstract】 Metal oxide semiconductor gas sensors are widely utlized for their fast response,high sensitivity,low cost,and long life.Typically,semiconductor gas sensors function by detecting the resistance value alterations of the test electrode at a constant operating temperature.This method enables instantaneous detection of changes in ambient gas information,facilitating rapid response to hazardous gas leaks.Nonetheless,it presents limitations in measurement parameters,susceptibility to cross-interference from various gases,and lacks the capability to selectively identify target gases in complex environments.To address these issues,this thesis focuses on hydrogen sulfide as the target gas,investigates the impact of variations in sensor operating temperature on the electrical characteristics of the sensitive material,explores the relationship between different heating modes and the dynamic response of gas sensitivity,proposes a semiconductor sensor variable temperature gas-sensitive measurement method,and makes use of its dynamic response characteristics to achieve high-accuracy discrimination of the target gas,and then carries out the fundamental research on its practical application.The main research results are as follows:(1)A gas sensor test system was designed and fabricated.Through control of three parameters—namely,heating interval,heating period,and heating duty cycle—investigation into the dynamic response of gas sensitivity under 27 distinct rectangular wave heating modes was conducted using commercially available sensors.Analysis of dynamic response curves facilitated the design of a variable temperature gas-sensitive measurement method and operational mode for hydrogen sulfide,characterized by high characteristic peak,short heating periods,and the significant response characteristic.(2)Based on the gas-sensitive dynamic response data,12 time-domain feature parameters highly correlated with gas species and concentration were devised,and a Pearson correlation coefficient feature extraction and high-precision gas identification method based on the dynamic response of gas sensors was proposed,which combined different dimensional feature vectors by the Pearson correlation coefficient method.For the different dimensions feature vectors,three commonly used algorithms,namely K-nearest neighbour,support vector machine and artificial neural network,were used to analyse the accuracy of single gas discrimination and concentration prediction,and the back propagation neural network algorithm was used to analyse the accuracy of the prediction of the composition of the two-component mixed gases,and the results were compared to obtain the combination of feature vector and algorithms for the high-accurate discrimination of hydrogen sulphide,and the accuracy of hydrogen sulphide identification under two-component mixed gases could reach 99%,with a concentration prediction error of 0.834 ppm.(3)A hydrogen sulfide analysis system was developed based on the sensor variable temperature gas-sensitive measurement method,with the ESP32 microcontroller as the control and signal processing center,and the trained artificial neural network model was compressed and deployed to the system microcontroller terminal utilizing a combination of highly accurate hydrogen sulphide discrimination feature vector algorithms and the TensorFlow Lite framework,which had successfully achieved high-accurate discrimination and prediction of the concentration of hydrogen sulfide in simulated application environments.更多还原