【作者】 王阳；

【导师】 孟庆浩；

【作者基本信息】 天津大学 ， 检测技术与自动化装置， 2013， 博士

【摘要】 移动机器人气味源定位研究始于二十世纪九十年代，主要研究利用配备有气味浓度及其它传感器的移动机器人搜寻事先未知的气味释放源头的问题。该研究涉及传感及信息处理、移动机器人学、计算智能、流体力学和仿生学等多个研究领域，在有毒/有害气体泄露检测、火源探测、灾后搜救和深海热泉勘察等方面具有广阔的应用前景。从气味源释放的气味分子在空气中扩散形成的空间分布称为烟羽。对气味烟羽进行计算机仿真一方面可以帮助理解浓度分布情况，另一方面可以作为气味源定位实验的辅助手段。本文围绕移动机器人气味源定位和气味烟羽仿真问题，重点开展了以下研究工作。采用多尺度分析方法研究了室外近地面单点风速在不同时间尺度下的脉动特征。室外环境中烟羽的扩散受风场的控制，但目前尚不清楚不同尺度风速脉动具有哪些规律。针对此问题，使用经验模态分解方法将风速序列分解得到多个时间尺度的风速脉动信号，并分析了这些脉动信号的自相关性、样本熵和湍动能，以及各尺度脉动信号对风稳定度的影响。结果表明，风速脉动信号在小尺度上具有不规则性，在大尺度上则表现出周期性，而时间尺度在8s至240s范围内的风速信号是造成风向变化的主要原因。为了研究风场中风向分布的均匀性，提出了区域内风向一致性指标。通过比较风向一致性指标与风速之间的关系表明，风速越大则区域内的风向越均匀，风速越小则区域内风向的差别越大。采用动力学异同性分析和相关性分析方法研究了不同位置的风速序列之间的关系。结果表明风场中距离越近、地形越相似位置的风速序列的动力学属性越相近，相关性越强。针对现有的烟羽仿真模型的不足，建立了基于实测风场数据的烟羽仿真环境。仿真风场由实测风场通过时间插值和空间插值获得，烟羽模型则采用经过修改的基于细丝的大气扩散模型，并建立了金属氧化物半导体传感器模型。仿真烟羽与实测烟羽的对比结果表明两者间具有相似的瞬时和统计特征。提出了基于模拟退火的气味源定位算法。该算法使用模拟退火策略选择浓度场的最大值，也就是气味源。通过室外场景下的仿真以及室内场景下的实验验证了本文提出的算法可以准确地定位气味源，并能够有效地克服风场和浓度波动造成的影响，且可避开局部最优。更多还原

【Abstract】 Mobile robot odor source localization (OSL) research, which aims to find one ormultiple previously unknown odor sources using one or multiple mobile robotsequipped with odor concentration and other auxiliary sensors, began from1990s. Thisresearch is related to the fields such as sensing and information processing, mobilerobotics, computation intelligence, hydrodynamics and bionics. It is expected that theOSL research will have a wide application in toxic/harmful gas leak detection, firesource monitor, post-disaster rescue and deep-sea hydrothermal vent exploration.Odor molecules released from their source and diffused in the air form an odorplume. The simulation of the odor plume on the one hand can help to understand theconcentration distribution; on the other hand can be a good auxiliary for the OSLexperiments. Focusing on the mobile robot odor source localization and odor plumesimulation problems, the achievement of this dissertation can be concluded asfollows.The multi-scale analysis method is adopted to study the pulsant characteristics ofsingle-point wind velocities under multiple time scales. In outdoor environments, thedispersion of the odor plume is dominated by the wind. However, the characteristicsof the pulsant wind velocities of different time scales are not clear yet. To address thisproblem, the multi-scale pulsant signals of the wind velocities are derived byempirical mode decomposition method, and the autocorrelation, sample entropy andturbulent energy of these pulsant signals are analysed. The impact of these pulsantsignals on wind steadiness is also investigated. Results show that the small-scalepulsant signals are more irregular than the large-scale ones. The pulsant signals underthe time scales from8s to240s are the main causes of the wind direction fluctuation.To study the uniformity of the wind direction in the wind field, the spatialwind-direction consistency index is proposed. The comparison between the spatialwind-direction consistency index and the wind speed shows that the wind direction ismore uniform when there is higher wind speed and vice versa.The dynamic conformity and correlation analysis methods are used to investigatethe relationship of the wind velocities in different locations. Results show that thecorrelation and the similarity of the dynamics properties of the wind velocitysequences recorded at the positions being close to each other and having similar terrain are more significant.In view of the drawbacks of current odor plume simulation models, a real winddata based odor plume simulation environment is built. The simulation wind field isderived using temporal interpolation and spatial interpolation of the real wind fielddata. The modified filament based atmospheric dispersion model is used as the plumemodel. The metal oxide semiconductor gas sensor model is also built. The comparisonof the simulated plume and real plume shows that they have similar instantaneous andstatistical characteristics.A simulated annealing based odor source localization algorithm is put forward.The algorithm obtains the maximum of the concentration field, which is exactly theodor source, employing a simulated-annealing strategy. The simulation on the openoutdoor scene and the experiments in the indoor scene verify that the proposedalgorithm is able to locate the odor source accurately. At the same time, the influencesof wind and concentration fluctuation can be overcome, and the local extrema can beavoided.更多还原