基于气体传感器阵列的混合气体检测系统

【作者】 王玮；

【导师】 陈明；

【作者基本信息】 西北工业大学 ， 检测技术与自动化装置， 2002， 硕士

【摘要】 随着科学技术的进步和工业生产的发展，对多组份气体的检测和分析的要求不断提高，但是现有的检测手段越来越不能满足需要，因而发展低成本、高性能的结合了气体传感器阵列与模式识别技术的智能电子嗅觉系统(电子鼻)已成为气体检测的新趋势。 本文分析研究了电子嗅觉系统的基本原理和系统组成，设计构建了一套气体传感器阵列和人工神经网络模式识别技术相结合的混合气体检测系统，并利用这套系统对目前在电子嗅觉系统中使用较广泛的几种信号预处理算法和人工神经网络模型的处理能力和辨识效果进行了分析和比较，最后得出了以下结论： 1) 气体传感器阵列与模式识别技术相结合能够很好地分析和辨识混合气体组份及其浓度。气体传感器不拘泥于某一种气体传感器，只要是具有宽响应范围的气体传感器均可使用，这大大降低了电子嗅觉系统的成本。对传感器阵列响应的后续信号处理分析的原理基本相同，因此，模式识别算法一定程度上具有通用性。 2) 信号预处理是提高电子嗅觉系统性能的一个必要步骤。在使用较多的几种信号预处理算法中，阵列归一化算法消除了传感器信号中的浓度因素，因此，在对气体浓度不感兴趣但要求准确识别气体类别的时候特别有用。 3) 人工神经网络以其非线性映射能力、高容错性和鲁棒性，有效地解决了由气体传感器普遍存在的交叉敏感性所带来的非线性严重等问题，并能在一定程度上抑制传感器的漂移或噪声，有助于气体检测精度的提高。反向传播(BP)神经网络和径向基函数(RBF)神经网络各具特点，辨识能力稍有差别。径向基函数神经网络的训练时间远小于反向传播神经网络，且不存在局部极小问题。另外，利用两个反向传播神经网络级联所形成的两级网络辨识气体的组份及其浓度的能力比单级反向传播神经网络高。 人工神经网络与气体传感器相结合，用于识别、分类、诊断和预测，将进一步提高气体检测系统的智能水平。现在，这类系统大都用微型计算机或单片机实现神经网西北工业大学硕士学位论文摘要络的功能，还处于实验室的研究阶段。随着人工智能和人工神经网络的发展，特别是神经网络芯片集成度和速度的提高，应用人工神经网络的电子嗅觉系统必将得到迅速发展和广泛应用。更多还原

【Abstract】 As the advancement of science and technology and the development of industry, the requirements of the detection and analysis of multicomponent gases are on great rise. But existing detection methods cannot meet them. Therefore, the intelligent electronic olfactory system (Electronic Nose) based on gas sensor array and pattern recognition has become the new tendency of gas detection because it can meet the main requirements such as high sensitivity, good selectivity, long-term stability, low cost and so on.In this thesis, the fundamental principle and system constituent of the electronic olfactory system are analyzed and studied; a set of detection system of gas mixture, combined gas sensor array with artificial neural network pattern recognition technology, is designed and constructed. Employing this system, the processing ability and identification results of several preprocessing algorithms and artificial neural network models are compared and analyzed. And finally the following conclusions are arrived:1) Gas sensor array coupled with pattern recognition technology has the good ability to identify the gas species and quantify its concentration. Any type of gas sensor that can respond broadly to a range or class of gases rather than to a specific one can be employed in the electronic olfactory system, which greatly reduces the cost. The principles of processing and analysis of sensor array responses are essentially similar, thus to some extent, the algorithm of pattern recognition is universal.2) Signal preprocessing is a necessary step to improve the performance of the electronic olfactory system. Among the various preprocessing algorithms, array normalization can remove the concentration factor in sensor responses, so it is particularly useful when the gas concentration is of no interest but fine discrimination is required.3) Artificial neural network (ANN), which has the ability of nonlinear mapping, high tolerance and robustness, can more effectively solve problems such as serious nonlinearitybrought by the cross sensitivity of gas sensors, and to a degree, can compensate the sensor drift and environmental noise, helping to improve the precision of gas detection. Back Propagation (BP) neural network and Radial Basis Function (RBF) neural network have their own advantages respectively and their identification capabilities are little different. The learning process of RBF network is much faster and easier than that of BP network, and not existing local minimum areas. Furthermore, it is also found that, compared with the Single-Stage BP network, the Two-Stage BP network (composed of two BP networks) has high ability to identify gas species and quantify its concentration.Artificial neural network together with gas sensors, which is applied to identification, classification, diagnosis and prediction, will further improve the intelligence of gas detection system. At present, most of this type system use micro-computers or microprocessors to perform the function of neural network, and are still under research in the laboratory. But with the development of artificial intelligence and artificial neural network, especially the sharp increase of integration level and speed of neural network chip, the electronic olfactory system employed artificial neural network will develop quickly and be applied widely.更多还原