【作者】 孙丽萍；

【导师】 李坚； 曹军；

【作者基本信息】 东北林业大学 ， 木材科学与技术， 2008， 博士

【摘要】 木材是全球应用最广的工程材料,是当今世界四大原材料(钢材、水泥、木材、塑料)中唯一可以再生和循环利用的绿色材料。而木材干燥是木材加工利用的基础研究之一。在木材干燥过程中外界的环境参数和木材本身的物理参数都可能影响木材含水率的变化,如何快速有效的建立各种参数与含水率之间的对应关系,是木材干燥的重要基础研究内容之一,也是实现木材干燥全自动控制,提高干燥质量,减少能量消耗,缩短干燥时间的先决条件,木材干燥是一个复杂的强耦合、非线性的动力系统,其木材干燥的关键是木材含水率参数的检测。近年来,随着多传感器信息融合理论及其应用技术研究的快速发展,多传感器融合技术已经成为一个重要的研究领域。针对木材含水率检测问题的特点,多传感器信息融合克服单一传感器在线检测时准确度低,测量数据变异大,全面性和系统性差等缺点。本研究即以此为基点,探讨了多传感器信息融合方法及其在木材干燥含水率在线检测的应用。在吸取发达国家和国内先进技术的基础上,以木材含水率检测为依据,在比较和选取传感器的同时,设计了以单片机为核心的木材含水率检测电路和控制系统的主控制器。采用双CPU结构的设计简化了硬件电路,完成构建木材含水率测试系统,实现了木材干燥含水率在线检测系统的硬件设计。深入分析木材干燥机理及木材干燥过程中的强耦合关系,依据多传感器信息融合技术,构建出符合木材干燥过程的木材含水率在线检测的分层融合体系。利用多传感器数据进行目标的状态估计,通过数学方法寻求与观测数据最佳拟合的状态向量,对多传感器的数据进行多级别、多方面、多层次的处理,这个过程是对多源数据进行检测、结合、相关、估计和组合,以达到精确的状态估计和身份估计。干燥窑内高温、高湿以及风机运行等恶劣的环境因素直接影响木材含水率传感器的检测值,导致出现粗大数据检测值和实际值的偏差,存木材含水率融合体系中,数据层融合利用底层传感器数据的检测(温度传感器和湿度传感器)和状态估计,采用与常规滤波的涵义及方法不同的最优估计方法Kalman滤波和被誉为调和分析发展史上的“数学显微镜”小波包变换方法。从Kalman滤波和小波包变换方法的仿真分析中可见,小波包在检测突变信号方面具有很大的优势,而Kalman滤波在处理实时数据上具有很好的效果。在特征层处理中把经过数据层处理的数据利用改进的最小二乘支持向量机和偏最小二乘回归的数据融合算法,得到木材干燥含水率的预测,并对预测模型进行了仿真和误差分析,结果表明偏最小二乘回归可以实现很好的预测效果。对于类似木材干燥的非线性回归问题,可用基于核函数的支持向量机算法。支持向量机具有依据有限样本训练获得良好的泛化能力,并且是一个通用的学习机,它将原空间映射到一个高维的特征空间里,使原分类问题在特征空间里成为一个线性可分的问题。由于木材含水率的变化与多种参数相关,通过支持向量机的改进算法建立影响木材含水率变化的输入输出模型,同时剔除对木材含水率变化影响甚小或者不影响的参数。最后,依据木材干燥在线检测的分层融合体系基本框图,基于Matlab图形用户界面(GUI)建立了木材含水率存线检测试验仿真平台。数据融合技术与人工智能的发展,是提高木材含水率实时在线检测的一个新的有效途径,通过对影响木材含水率环境参数和物理参数的研究,建立一种木材含水率存线检测的分层融合体系和功能模型,将数据融合技术、人工智能理论和支持向量机算法有机的结合,对提高我国木材干燥控制系统的智能化程度,提高木材干燥质量和干燥速率,具有重要现实的意义和科学价值。更多还原

【Abstract】 The lumber is the widely broadest engineering applied material,the only green material that may regenerate and circulate now in the four big raw materials(steel products,cement, lumber,plastic) of the world.Wood drying is one of basic researches in wood machining and using.The world forest resources reduce day by day,which brings the questions on environmental protection and ecology,in the wood drying process the outside environment parameter and the lumber itself physical parameter all possibly affects the change of lumber moisture content(LMC).To establish corresponding relations fast and effectively between each parameter and moisture content,that’s one of important basic research contents,and also that’s the precondition to realize the completely automatic control,improve the dry quality, reduce the energy consumption and reduce the drying time.Wood drying is a complex strong coupling,non-linear dynamic system,lumber moisture content parameter detection is the key. In recent years,the multi-sensor fusion technology already became an important research area along with its theory and the application engineering research fast development.In view of question of LMC detection,the multi-sensor information fusion overcomes low accuracy,big data variation,inferior integrity and systematic in the sole sensor online detection.This research namely takes this as a point,which has studied the multi-sensor information fusion method and application on the moisture content online detection.On the foundation of absorbing developed country and domestic advanced technology, taking the lumber moisture content examination as the basis,comparing and selecting the sensors,has designed the monolithic integrated circuit as core lumber moisture content examination electric circuit and control system hardware systems.The structure is designed by using the double CPU,which simplifies the hardware electric circuit,completes the construction of LMC test system and realizes the hardware design of moisture content online examination system.It analyses the wood drying mechanism and the strong coupling relations in the process, depends on the multi-sensor information fusion technology and constructs on-line detection hierarchical fusion system of the lumber moisture content accord with the wood drying process. Uses the multi-sensor data to carry on the goal the state estimation,seeks best-fit state vector with the observation data through mathematics method,Carries on the multi-ranks,multi-aspect, multi-level processing to the multi-sensor data,this process carries on the examination, the union,the correlation,the estimate and the combination to the multiple source data, achieves the precise state estimation and the status estimate.In drying kiln bad factor such as high temperature,humidity as well as air blower movement directly influence LMC sensor examination values,which can cause deviation between the thick data examination value and the actual value,in the LMC fusion system,the data level fusion uses detection data from the bottom layer sensor(temperature sensor and humidity sensor) and the state estimation,it’s proposed that superior estimate method Kalman filter which is different from the conventional filter implication and the method and the wavelet packet transformation method by the reputation for the harmonic analysis history in "mathematics microscope" is proposed.From the simulation analysis of Kalman filter and wavelet packet transformation method,obviously, the wavelet packet has a very big superiority advantage to examine sudden change signal,but also the Kalman filter has the very good effect in the processing for real-time data.In characteristic level processing the data after the data level processing uses the data fusion algorithm include the improvement least square support vector machines(LSSVM) and partial least squares(PLS) regression,obtains the forecast of wood drying moisture content, and carries on the simulation and the error analysis to the forecast model,finally indicates that PLS regression can be possible to realize the very good forecast effect.Regarding non-linear regression question similar with wood drying,support vector machines algorithm based on nuclear is available.The SVM has the basis limited sample training to obtain good exude ability,and it’s an general learning machine,it maps the original space into high dimensional characteristic space,makes the original classified question become a linear separable question in the characteristic space.Because the change of LMC and many kinds of parameters are related,input and output model influenced lumber moisture content change is establish by the improvement SVM algorithm,simultaneously,the parameter which is really small to the LMC change influence or does not affect is rejected.Finally,based on system basic diagram of the online detection lamination fusion system in wood drying process and Matlab graphical user interface(GUI),a LMC online detection experiment simulation platform is built.The development of data fusion technology and artificial intelligence is a new effective way enhanced the LMC real-time online detection,through the research of environment parameter and physical parameter affected the lumber moisture content,a hierarchical fusion system and function model on LMC online detection are established,The organic combination of data fusion technology,artificial intelligence theory and SVM algorithm has the important reality significance and scientific value to enhance the control system intellectualized degree and the wood drying quality and drying rate in our country and the lumber use factor.更多还原