【作者】 王昕；

【导师】 孙金玮； 魏国；

【作者基本信息】 哈尔滨工业大学 ， 仪器科学与技术， 2012， 博士

【摘要】 作为现代传感技术的主要发展趋势之一，多功能传感技术因其具有体积小、功耗低、能有效克服交叉敏感效应等优点，能够很好地满足各种测控系统在测试精度、稳健性和抗干扰能力等方面的要求，已越来越多地在科学研究和工程实际中被采用。同时，由于与传统的单一功能传感器相比，多功能传感器的输入输出特性更加复杂，因而对其信号重构方法也提出了更高的要求。本论文的目的是研究多功能传感器信号重构中的若干关键技术。首先，设计了一种多功能传感器，以实现对食品渗透脱水过程中，三元溶液浓度的在线测量。然后，以此传感器为基础，研究多功能传感器信号重构中的三个关键技术：信号重构的方法、信号重构中均方误差与计算量的平衡算法以及信号重构中的样本选择方法。本课题的研究，不但能够实现对三元溶液的测量，其信号重构的方法还可广泛地应用于多种多功能传感器。论文的主要研究工作如下：针对三元溶液两种溶质的浓度测量，设计了一种多功能传感器。该传感器通过测量超声波在水中的传播速度和水的电导率间接地反映三元溶液溶质的浓度。以该多功能传感器为基础，制作相应的实验电路并获得实验数据，用于验证本文提出的多功能传感器信号重构关键技术的可行性和有效性。研究了多功能传感器的信号重构问题，提出了基于B样条和扩展卡尔曼滤波的信号重构方法。本方法将B样条作为多功能传感器的逆模型，并利用扩展卡尔曼滤波估计B样条模型的控制系数。仿真和实验结果表明，本方法不但能够获得较高的信号重构精度，而且结构简单、计算复杂度低，适合在微处理器上实现，能够满足智能传感器在线信号重构的需求。研究了多功能传感器信号重构过程中重构精度和计算量的平衡算法。首先，研究B样条逆模型的结构对信号重构精度和计算量的影响，并给出信号重构精度和计算量的估计方法。然后，设计适应度函数，利用遗传算法寻找能够获得信号重构精度和计算量平衡的B样条模型结构。仿真和实验结果表明，经过平衡后的B样条模型，在保证重构效果的前提下，能够极大地降低建模过程中所需的计算量，即实现了重构精度和计算量的平衡，从而降低微处理器在对多功能传感器进行信号重构时的负荷。针对大批量多功能传感器的信号重构，提出了多功能传感器信号重构中训练样本的选择方法。利用核减法聚类对多功能传感器的高维样本数据进行分类和筛选，从而找出数量较少且能够保证信号重构精度的训练样本。仿真结果表明，经过了样本选择的训练样本，虽然在数量上明显减少，但是仍然能够获得较高的重构精度。因此，将本方法应用于大批量多功能传感器的信号重构中，能够在保证重构精度的前提下，大大减少采样所需的工作量。更多还原

【Abstract】 As one of the major trends of the modern sensor technology, multifunctionalsensing technology can satisfy the requirement of various monitoring systems in themeasurement accuracy, robustness and anti-jamming capability because of its smallsize, low power consumption, and the ability of overcome the cross-sensitivityeffect. At the same time, the input-output characteristics of multifunctional sensorsare more complex than traditional single-functional sensors, which require higherperformance of their signal reconstruction methods.The paper aims to study the key techniques of signal reconstruction formultifunctional sensors. First of all, a multifunctional sensor for onlinemeasurement of the two solute concentrations of ternary solution in the osmoticdehydration process is designed. Then, some key techniques for the signalreconstruction of multifunctional sensors based on this sensor, including the signalreconstruction method, the tradeoffs between the reconstruction accuracy andcomplexity, and the sample selection for multifunctional sensors are studied. Theresearch results of this subject can not only be used to achieve the measurement ofthe ternary solutions, but also be used in the signal reconstruction process for avariety of multifunctional sensors.The main study works are as follows:The multifunctional sensor for measuring the solute concentrations of ternarysolutions simultaneously is designed. The solute concentrations of ternary solutionsare achieved indirectly by measuring the temperature, the ultrasonic propagationvelocity and the conductivity of the solution. At the same time, to verify thefeasibility of the proposed methods, an experimental circuit based on themultifunctional sensor is designed to obtain the experimental data.The signal reconstruction of multifunctional sensors is studied and areconstruction method based on B-spline and Extended Kalman filter is presented.B-spline is used as the structure of the sensor inverse model and the modelparameters are estimated using the Extended Kalman filter. The results of thesimulations and experiments show that the proposed method can achieve highsignal reconstruction accuracy, and its reconstruction process is simple and suitableto be applied in the microprocessors which meet the requirement of online signalreconstruction for smart sensors.The method for the tradeoffs between the reconstruction accuracy andcomplexity in the signal reconstruction is studied. First, the impact of the B-splinestructure to the reconstruction accuracy and complexity is analyzed. Then the fittness function is designed and the genetic algorithm is adopted to optimize the B-spline model structure. The results of simulation and experiments show that theoptimized B-spline structure can reduce a lot of complexity without significantreduction of the signal reconstruction accuracy, and achieve the tradeoffs betweenthe reconstruction accuracy and complexity which can reduce the burden ofmicroprocessors.The sample selection method for the signal reconstruction of large batches ofmultifunction sensors is presented. Kernel subtracting clustering is adopted toclassify the high-dimension sample data of multifunctional sensors and select thetraining samples suitable for the signal reconstruction. The results of thesimulations show that although the number of training samples is reduced, thereconstruction method can still obtain high accuracy. Therefore, the presentedmethod can be used to reduce the workload of sampling in the signal reconstructionfor large batches of multifunctional sensors.更多还原