电子鼻优化设计及其在坚果品质检测中的应用

发布时间：2018-01-03 02:10

本文关键词：电子鼻优化设计及其在坚果品质检测中的应用　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,坚果品质检测愈发受到社会各界的关注。然而,因其坚硬外壳的阻挡,使得坚果内部品质信息很难被直接准确地获取。作为一种嗅觉仿生仪器,电子鼻基于样品挥发性气体的"指纹图谱"信息进行内部品质的快速判定,令其在坚果品质检测领域具有较好的应用前景。目前,电子鼻的研发主要朝着自动化、智能化方向发展。而现有电子鼻主要分为便携式与台式两种,其中便携式电子鼻主要存在人机交互性差、数据处理效率低等问题;台式电子鼻则缺少自动进样机构,自动化程度低下等缺点。针对上述问题及不同应用环境,本文分别研发了基于DSP的嵌入式便携电子鼻和基于DELTA并联机械臂的全自动电子鼻系统,并以山核桃为研究对象,进行系统实际检测能力的验证。本文主要研究内容和结论如下:(1)研制了基于DSP的嵌入式便携电子鼻系统,包括气体采集响应模块、信号采集及调理模块、DSP控制处理模块和信息显示交互模块。系统采用FLUENT流体分析进行气室的优化设计,并基于PS-LCD进行人机交互界面开发,以实现检测过程可视化。同时,通过将数据预处理、特征值提取、模式识别算法移植进DSP使电子鼻具有独立运算能力。(2)研制基于DELTA并联机械臂的全自动电子鼻系统,包括插拔式传感器响应模块、自动进样模块、信号采集及进样控制模块和上位机交互界面。系统通过传感器安装底座的设计实现不同类型传感器的快速插拔,并经FLUENT流体分析确定传感器分布及进气口高度最佳方案。自动进样模块则采用DELTA 3自由度并联机械臂结构进行设计。同时,根据系统功能模块,基于Matlab GUI开发环境为系统提供一个高集成度的人机交互界面。(3)针对不同检测对象,基于非搜索性特征选择策略,采用均值分析、变异系数分析、聚类分析、相关性分析及方差膨胀因子分析对由平均微分值、稳定值、面积值构成的初始特征矩阵进行逐步优化,并根据最终获得的优化矩阵选择相应的传感器阵列进行应用。(4)进行了未加工山核桃陈化时间、加工山核桃货架期及掺杂比例检测。结果表明,不同陈化时间的未加工山核桃基于PCA算法能获得很好的区分效果;LDA较PCA能更好的对加工山核桃货架期及不同掺杂比例进行检测。同时,上述实验分别基于经PCA降维处理的数据进行了 PLSR和BPNN预测模型的构建。其中,加工山核桃货架期评定实验中,PLSR模型预测精度较低,建议采用精度更高的BPNN模型。而其他两组实验所构建的PLSR和BPNN模型对样品均具有较好的预测性能。
[Abstract]:In recent years, nut quality detection more attention of the society. However, the barrier for its hard shell, the internal quality of information is hard to obtain nuts directly and accurately. As a kind of bionic olfaction instrument, electronic nose based on samples of volatile gas "fingerprint" information for internal quality rapid determination, with its application good prospects in the field of nut quality detection. At present, the main research and development of electronic nose toward automation, intelligent direction. But the existing electronic nose is mainly divided into two kinds of desktop and portable, the portable electronic nose are poor human-computer interaction, data processing and low efficiency; the desktop electronic nose is the lack of automatic sampling mechanism the low degree of automation and other shortcomings. To solve the above problems and different application environment, this paper developed the embedded portable electronic nose based on DSP and DELTA based and online Automatic electronic nose system of the robot arm, and pecan as the research object, to verify the actual system detection ability. The main research contents and conclusions are as follows: (1) the development of embedded portable electronic nose system based on DSP, including the gas collection response module, signal collecting and processing module, DSP control module and the information interactive display module. The system adopts FLUENT fluid analysis optimization design of the gas chamber, and based on the PS-LCD development of man-machine interface, to realize visualization of the detection process. At the same time, the data preprocessing, feature extraction, pattern recognition algorithm is transplanted into DSP so that the electronic nose has independent operation ability. (2) research automatic electronic nose system based on DELTA parallel manipulator, including response module plug type sensor, automatic sampling module, signal acquisition and sampling control module and the host computer interaction system through the interface. The design of sensor mounting base to achieve quick plug from different types of sensors, and the FLUENT fluid analysis to determine the sensor distribution and inlet height best. Automatic sampling module is designed based on DELTA structure of the 3 DOF parallel manipulator. At the same time, according to the system function module, providing a high degree of integration of the man-machine interface system based on Matlab GUI development environment. (3) according to the different detected objects, non search feature selection strategy based on the analysis of the mean, coefficient of variation analysis, cluster analysis, factor analysis and stability of value by the average differential value, correlation analysis and variance inflation, the initial feature matrix of the area values were gradually optimized, and according to the application the final selection of the optimal matrix corresponding sensor array. (4) the unprocessed pecan processing period and aging time, the proportion of doping pecan shelf Detection. The results show that different aging time of raw pecans based on the PCA algorithm can achieve good distinguish effect; LDA is better than PCA to detect the processing of Pecan shelf life and different doping ratio. At the same time, the experiments were based on the PCA dimension of the data in the model PLSR and BPNN forecast. The processing of Pecan shelf life evaluation experiment, PLSR model prediction accuracy is low, the proposed use of BPNN model with higher precision. PLSR and BPNN models constructed by the other two groups of samples have better prediction performance.

【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP212;S664.1

【相似文献】