72个金属氧化物气体传感器数据集，可用于十类气体鉴别

Data Set Information: 实例数： 从基于72个金属氧化物气体传感器阵列的化学检测平台记录18000次系列测量。 属性（特征）的数量： 每次测量包含在260秒内记录的72个时间序列，每个时间序列以100 Hz的采样率（每秒采样数）采集。 数据集还包含时间、温度和相对湿度信息。 最终生成的数据集包括由26000个点组成的75个时间序列。 该档案包含从72个金属氧化物气体传感器阵列收集的18000个时间序列测量记录，这些传感器构成了我们的传感平台，用于在复杂环境条件下检测和识别潜在危险的化学气体物质，如下文相关手稿所述。我们的主要目的是使我们的数据集可供化学传感研究和机器学习社区以及其他感兴趣的社区免费在线访问，以开发与开放采样环境中的气体传感识别任务相关的替代竞争解决方案，如本文所追求的解决方案和/或导航。数据集只能用于研究目的。商业用途完全排除在外。 该数据集是从2010年12月到2012年4月（16个月）在加利福尼亚大学圣地亚哥分校的BioCiCuIt研究所2.5米（1.2米）-0.4米风洞研究试验台设施中收集的。具体而言，我们的定制研究设施配备了基于计算机监控的质量流量控制器的连续流量气体输送系统，通过不断将外部湍流空气吸入和穿过隧道并将其排回外部，以推进开放循环模式运行，从而产生相对较少的湍流气流，向试验场末端下游移动，特别适用于需要注入化学毒物或爆炸性混合物的应用，因为这样可以防止饱和。由一个完全计算机化的环境操作，一个“由玩家在一个PC上编程的玩家/阶段机器人服务器软件控制，配备了适当的串行卡胡”，并以最小的人为干预，设计的风洞试验台设施提供了多功能性，可在整个试验过程中以高精度和高重复性的方式，以所需浓度释放感兴趣的化学物质，同时保持适当的环境条件，以产生显示出湍流模式。本研究中考虑的设计风洞试验台设施的图示，以及问题的几何特征，以及化学分析物源和化学传感平台的准确位置，如下文引用的手稿图2所示。所设计风洞的实际图片也显示在随附手稿的补充材料图S.1中。 由此产生的数据集引发了十类气体鉴别问题，包括十种不同纯化学气体的记录，即丙酮、乙醛、氨、丁醇、乙烯、甲烷、甲醇、一氧化碳、苯和甲苯。目标是识别和区分相关浓度下的上述化学危害，而不考虑带注释的风洞研究设施内传感系统平台的位置以及环境和参数条件（更多详情请参见手稿）。见Vergara等人的表1。2013年（以下手稿），了解化学分析物危害的识别及其在气源出口处的标称浓度值（单位：百万分体积（ppmv））的详细信息。有关风洞试验台设施的更多详情，以及在创建上述数据集期间遵循的收集程序和使用的操作和环境参数的详细信息，请参考以下手稿。 Attribute Information: 传感器平台的响应以电阻的形式读出，电阻穿过集成传感器阵列的72个气体传感器的有源敏感膜；因此，每次测量产生72个通道的时间序列，每个时间序列由以每秒100个样本（Hz）的采样率采集的260秒时间序列表示，反映了评估情景中的所有环境变化。有关时间序列处理的更详细分析和讨论以及时间序列的图形说明，请分别参考以下手稿的第2节和第3节以及图4。 为了便于操作，数据被组织到11个文件夹中，每个文件夹包含上述和手稿表2中所述的每个化学类别标识和标称浓度的测量数量。例如，名为a€？甲苯_200a€的文件夹？表示气体标识的名称为甲苯，其剂量为200 ppmv。每个文件夹包含6个文件夹，每个文件夹代表风洞试验区域内的测线位置（位置1，L1，至位置6，L6，L1是距离气源最近的点），从中记录时间序列集。在每个文件夹中有300个文件，每个文件对应于隧道中每个位置记录的测量数量。每个文件的名称包含整个实验期间执行的每个测量的精确日志信息，其组织如下。文件名的前12位数字（例如2011060617）表示从年、月、日和时间开始收集每个特定测量值的日期和时间。名称文件的以下19个字符中的最后4位数字（例如，board_setPoint_500V）表示固定的工作温度值，由应用于化学传感器中嵌入式加热元件的电压值表示，应用于整个传感平台，可采用4至6 V的标称值，分辨率值为0.5 V。请注意，示例中的500V值是应用于传感器加热器的5V值的图形表示。有关我们平台中使用的化学传感器工作原理的更多详细信息，请参阅手稿第2节。文件名以下16个字符中的最后3位数字（例如，fan_设定值_060）表示用于在风洞中诱导不同人工气流速度的多步进电机驱动排气扇标称转速的设定值。在这种情况下，只采用了三个值：a€？000a€？在文件名中，表示最低转速（1500 rpm），值a€？060a€？表示风扇的中点转速值（3900rpm），值a€？100a€？表示风扇的最快感应转速5500 rpm。以下27个字符的字符串的最后14个字符（例如，mfc_设定值_甲苯_200ppm）描述了每个特定测量的分析物特性和浓度值。因此，刚才提到的示例表示与化学分析物标识a€？甲苯a€？对应的类别？以200 ppm的标称浓度值给药。最后，名称中的最后2或3位数字（例如a€？p7a€？）描述了该行。 Creators: Alexander Vergara (vergara '@' ucsd.edu) BioCircutis Institute University of California San Diego San Diego, California, USA Donors of the Dataset: Alexander Vergara (vergara '@' ucsd.edu) Jordi Fonollosa (fonollosa '@' ucsd.edu) Marco Trincavelli (marco.trincavelli '@' oru.se) Nikolai F. Rulkov (nrulkov '@' ucsd.edu) Ramon Huerta (rhuerta '@' ucsd.edu)

Data Set Information:
The dataset contains 18,000 sequential measurement records collected from a chemical detection platform based on a 72-element metal oxide gas sensor array. Number of attributes (features): Each measurement consists of 72 time series recorded over 260 seconds, with each time series sampled at 100 Hz (samples per second). The dataset also includes time, temperature, and relative humidity information. The final generated dataset includes 75 time series composed of 26,000 points each.
This archive contains 18,000 time series measurement records collected from a 72-element metal oxide gas sensor array, which forms our sensing platform for detecting and identifying potentially hazardous chemical gaseous substances under complex environmental conditions, as described in the associated manuscript below. Our primary objective is to make our dataset freely and publicly accessible to the chemical sensing research, machine learning, and other interested communities to develop alternative competitive solutions for gas sensing recognition tasks in open sampling environments, such as those pursued and/or navigated in this paper. The dataset may only be used for research purposes; commercial use is strictly prohibited.
This dataset was collected between December 2010 and April 2012 (16 months) at the 2.5 m (1.2 m)–0.4 m wind tunnel research testbed facility of the BioCircuits Institute, University of California San Diego (UCSD). Specifically, our custom research facility is equipped with a continuous-flow gas delivery system based on computer-monitored mass flow controllers, operating in open-loop mode by continuously drawing ambient turbulent air into and through the tunnel and exhausting it back to the outside, generating relatively low-turbulence airflow that travels downstream toward the end of the test field. This setup is particularly suitable for applications requiring the injection of chemical toxicants or explosive mixtures, as it prevents saturation. Operated by a fully computerized environment, controlled by a "player/stage robot server software" programmed on a PC and equipped with appropriate serial cards, the designed wind tunnel testbed facility offers versatility to release target chemical substances at desired concentrations with high precision and repeatability throughout the entire experimental process, while maintaining appropriate environmental conditions to produce airflow exhibiting turbulent patterns. Illustrations of the designed wind tunnel testbed facility considered in this study, along with the geometric characteristics of the setup, the exact positions of the chemical analyte source and the chemical sensing platform, are shown in Figure 2 of the referenced manuscript below. Actual photographs of the designed wind tunnel are also shown in Supplementary Figure S.1 of the accompanying manuscript.
The resulting dataset poses a ten-class gas discrimination problem, with records of ten distinct pure chemical gases: acetone, acetaldehyde, ammonia, butanol, ethylene, methane, methanol, carbon monoxide, benzene, and toluene. The goal is to identify and distinguish the aforementioned chemical hazards at relevant concentrations, regardless of the position of the sensing system platform within the annotated wind tunnel research facility and the environmental and parametric conditions (see the manuscript for more details). See Table 1 in Vergara et al. 2013 (the accompanying manuscript) for detailed information on the identification of the chemical analyte hazards and their nominal concentration values at the analyte source outlet, expressed in parts per million by volume (ppmv). For more details on the wind tunnel testbed facility, as well as the collection procedures followed during the creation of the aforementioned dataset and the operational and environmental parameters used, please refer to the accompanying manuscript.

Attribute Information:
The response of the sensing platform is read out as the resistance across the active sensitive membranes of the 72 gas sensors in the integrated sensor array; thus, each measurement yields 72-channel time series, with each time series representing a 260-second duration sampled at 100 samples per second (Hz), reflecting all environmental variations in the evaluation scenario. For a more detailed analysis and discussion of time series processing and graphical illustrations of the time series, please refer to Sections 2 and 3 and Figure 4 of the accompanying manuscript, respectively.
For ease of organization, the data is grouped into 11 folders, each containing measurement counts for each chemical class identifier and nominal concentration as described above and in Table 2 of the manuscript. For example, a folder named "toluene_200" indicates the gas identifier is toluene with a dose of 200 ppmv. Each of these 11 folders contains 6 subfolders, each representing a measurement line position within the wind tunnel test area (Position 1, L1, through Position 6, L6; L1 is the point closest to the analyte source) where the time series sets were recorded. Within each subfolder, there are 300 files, each corresponding to a single measurement count recorded at that position in the tunnel.
The name of each file contains precise log information for each measurement performed during the entire experiment, organized as follows: The first 12 digits of the filename (e.g., 2011060617) indicate the date and time when each specific measurement was collected, starting with year, month, day, and time. The last 4 digits of the subsequent 19 characters of the filename (e.g., board_setPoint_500V) represent the fixed operating temperature value, expressed as the voltage applied to the embedded heating elements within the chemical sensors, applied across the entire sensing platform. Nominal values range from 4 to 6 V, with a resolution of 0.5 V. Note that the 500V value in the example is a graphical representation of the 5 V value applied to the sensor heaters. For more details on the operating principles of the chemical sensors used in our platform, please refer to Section 2 of the manuscript. The last 3 digits of the subsequent 16 characters of the filename (e.g., fan_setPoint_060) represent the nominal rotational speed setpoint of the multi-step motor-driven exhaust fan used to induce different artificial airflow velocities in the wind tunnel. In this study, only three values were used: "000" in the filename indicates the lowest rotational speed (1500 rpm), "060" indicates the mid-point rotational speed value of the fan (3900 rpm), and "100" indicates the fastest sensed rotational speed of the fan, 5500 rpm. The last 14 characters of the subsequent 27-character string (e.g., mfc_setPoint_toluene_200ppm) describe the analyte characteristics and concentration value for each specific measurement. Thus, the aforementioned example corresponds to the class identified with the chemical analyte "toluene", administered at a nominal concentration value of 200 ppm. Finally, the last 2 or 3 digits in the filename (e.g., "p7") describe the test run.

Creators:
Alexander Vergara (vergara '@' ucsd.edu)
BioCircuits Institute, University of California San Diego, San Diego, California, USA
Dataset Donors:
Alexander Vergara (vergara '@' ucsd.edu), Jordi Fonollosa (fonollosa '@' ucsd.edu), Marco Trincavelli (marco.trincavelli '@' oru.se), Nikolai F. Rulkov (nrulkov '@' ucsd.edu), Ramon Huerta (rhuerta '@' ucsd.edu)

搜集汇总

背景与挑战

背景概述

该数据集包含72个金属氧化物气体传感器阵列记录的18000次测量，用于鉴别十类化学气体（如丙酮、乙醛等）。数据采集于特定研究设施，时间跨度为16个月，适用于气体传感识别任务的研究。