【作者】 罗刚银；

【导师】 唐玉国； 王弼陡；

【作者基本信息】 中国科学院研究生院（长春光学精密机械与物理研究所） ， 光学工程， 2012， 博士

【摘要】 全自动血型分析系统作为高端血液分析设备的代表，其研究涉及到许多重要的关键技术，包括运动控制技术、微量移液技术、故障诊断技术和血型图像识别技术等。因此，解决好各个关键技术的研究工作，是实现全自动血液分析系统的前提。本文针对目前国内对全自动血型分析系统的需求，在分析目前血型分析技术和国外血型分析系统研究现状的基础上，以解决好血型分析系统各个关键技术为基础，研制了国内首台全自动血型分析系统。全自动血型分析系统的微量移液机构是一个多轴协调运动的复杂机械结构。本文研究了微量移液机构的机械结构、控制策略、系统模型、单轴运动控制器、多轴运动控制器和基于DSP芯片、FGPA芯片的运动控制主板。由于采用了改进的单神经元PID速度控制器和改进的模糊PID位置控制器，弥补了传统PID算法的不足，单轴的定位精度可达0.05mm，最大运动速度可达1.6m/s，最大加速度可达6m/s2，极大的提高了微量移液机构的运行速度和稳定性，最终实现了每小时80个样品的仪器检测速度。全自动血型分析系统的微量移液技术是微量移液机构实现精确加样的前提。本文研究了一种精密微量泵的机械模型、加样精度检测、取样针堵塞识别和液面分层探测等问题。通过应力分析和模态分析，证明了柱塞选择ZrO2材料的有效性，其一阶振型最大位移小于1um。通过对微量泵加样精度检测机构的光学改进，把荧光采集效率提高了48.19%。通过分析取样针导管压力数据的幅值分布和频谱特性，有效的识别了血块堵塞和遇到泡沫两种情况。最后，采用CD4046芯片设计的液面传感器，研究了红细胞和血清分层探测时的不同导电特性。全自动血型分析系统自动化要求的实现需要分析系统具有较低的故障率。本文研究了离心机的机械结构、故障检测平台、振动信号软件滤波、特征提取和小波BP神经网络故障诊断器。通过自适应率和遗传算法对小波BP神经网络的改进，使小波BP神经网络训练的迭代次数由463代降低到了112代，故障诊断正确率由82.5%提高到了95%，有效的提高了小波BP神经网络的收敛速度和故障诊断正确率。全自动血型分析系统运行的最终目的是为了正确地识别血型结果。本文研究了血型试剂卡的倾斜校正、图像分割、种类识别和结果识别四个问题。通过采用两次匹配算法，使血型试剂卡的模板匹配时间由9.342s降低到了0.946s，有效地减少了图像处理所需的时间。课题的研究工作得到了国家高技术研究发展计划（863计划）（2011AA02A104）、苏州市科技计划项目（ZXG201135）等基金的支持。更多还原

【Abstract】 The automatic blood type analysis system was one of the representative of thehigh-end blood analysis equipments,and the system’s research involves a number ofkey technology,such as motion control, micro pipette, fault diagnosis, imageprocessing,and so on. Therefore, the solution of these key technology is the premiseof the system’s research.To meet the current domestic demand of the automatic bloodtype analysis system,based on the study of the blood type analysis technology and theforeign blood type analysis system,we solved these key key technology and developedthe domestic first automatic blood type analysis system.The micro-pipette institutions of the automatic blood type analysis system is acomplex mechanical structures with multi-axis coordinated motion. Its mechanicalstructure, control strategy, system model, single-axis motion controller, multi-axismotion controller was studied,and its control circuit was designed by the chip DSPand FGPA. The improvement of the single neuron PID speed control algorithm andfuzzy PID position control algorithm met the deficiency of the traditional PIDalgorithm.The speed and stability of the micro-pipette institutions was greatlyimproved with its single axis positioning accuracy5mm, its maximum velocity1.6m/sand its maximum acceleration6m/s2,which led the instrument’s detection speed to80/h eventually.The micro-pipette technology is the premise to acheive the precise sample of themicro-pipette institutions. The mechanical model of a micro-pump, the outputaccuracy detction of the pump,the recognition of sampling needle’s block and theliquid level detection was studied. Choosing the material ZrO2was proved effectiveby stress analysis and modal analysis, for the maximum displacement of its first-ordermodes was less than1um. The optical design improved the fluorescence collection efficiency of the micro-pump precision’s testing organization by48.19%. The bloodclots blocking and serum bubbles was distinguished by analyzing the sampling needlecatheter pressure’s amplitude distribution and spectral characteristics. The differentelectrical properties of the red blood cells and serum was studied with the liquid levelsensor designed by chip CD4046.The automatic blood typing system needs lower failure rate to achieve itsautomation. We studied the centrifuge’s mechanical structure, fault detectionplatform,vibration signal software filtering, feature extraction and wavelet BP neuralnetwork fault diagnosis. The wavelet BP neural network’s convergence speed andfault diagnosis accuracy was improved by adaptive rate and genetic algorithms. Itstraining iterations reduced by the463generations to112generations, and its faultdiagnosis’s correct rate82.5%to95%.The automatic blood type analysis system is committed to the correct analysis ofthe blood group results. Four issues was studied,such as the blood card’s tilt correction,image segmentation,species identification and results identify.The time required forimage processing was reduced effectively by using two matching algorithms,whichreduced the template matching time of the blood card from9.342s to0.946s.The subject was supported by the "863" Project with the mumber2011AA02A104.And,it also was supported by the science and technology projects inSuzhou with the number ZXG201135.更多还原