Rogue Device Detection via Redundant Environmental Telemetry from Cargo Aircraft Operations

Overview This dataset contains one-second-resolution time-series telemetry recorded during real cargo aircraft flight operations (Antonov), augmented with two synthetic sensor spoofing scenarios. It is designed as a benchmark resource for unsupervised anomaly detection methods on univariate and multivariate sensor time series. The dataset captures four environmental parameters: altitude (H), atmospheric pressure (Pr), temperature (T), and relative humidity (Hum); each measured simultaneously by two independent sensors (Sensor 1, Sensor 2). The agreement between paired sensors is the core anomaly signal: under healthy operation, the redundant sensors track each other closely; under a fault condition, Sensor 2 diverges from its paired reference (Sensor 1). Sensor Architecture and Anomaly Signal Each parameter is instrumented with a redundant pair: Pair Parameter Unit Sensor 1 (reference) Sensor 2 (monitored) H Barometric altitude m H1 H2 Pr Atmospheric pressure Pa Pr1 Pr2 T Temperature °C T1 T2 um Relative humidity % Hum1 Hum2 Sensor 1 is the trusted reference throughout all scenarios. Anomalies are expressed as inter-sensor divergence: |S1 − S2| exceeds its normal operating bounds when Sensor 2 is under sensor spoofing. This redundancy-based detection principle supports the identification of rogue or malfunctioning sensor devices connected to the monitoring system. Files All files contain 11,425 rows (one row per second) and 9 columns. The files are unlabeled; no per-row anomaly flag column is included in this release. Column Description Unit time Unix timestamp s H1 Barometric altitude, Sensor 1 — reference m H2 Barometric altitude, Sensor 2 — monitored m Pr1 Atmospheric pressure, Sensor 1 — reference Pa Pr2 Atmospheric pressure, Sensor 2 — monitored Pa T1 Temperature, Sensor 1 — reference °C T2 Temperature, Sensor 2 — monitored °C Hum1 Relative humidity, Sensor 1 — reference % Hum2 Relative humidity, Sensor 2 — monitored % Scenarios flight001_normal.csv - Nominal operation (real data). Real flight telemetry with no injected faults. Both sensors in each pair agree throughout the full flight arc. This file represents healthy operation and serves as the training baseline for unsupervised anomaly detection models. flight001_mixed.csv - Intermittent sensor spoofing (synthetic). Sensor 2 channels (T2, Pr2, H2, Hum2) receive periodic burst noise in repeating on/off intervals, while Sensor 1 retains the original recorded signal. During active sensor spoofing ( fault windows), Sensor 2 oscillates across the full physical range of each channel (e.g., Pr2: 0–150,000 Pa; T2: 5–40 °C; Hum2: −20–80 %). Between bursts, Sensor 2 briefly returns to nominal and converges with Sensor 1. The burst pattern is synchronised across all four affected channels. This scenario models intermittent or transiently degrading sensor behaviour. flight001_rogue.csv - Complete sensor spoofing (synthetic). Noise of the same amplitude as the mixed scenario is applied continuously and without interruption throughout the entire flight. No recovery windows occur. Sensor 2 readings are fully corrupted from the first to the last sample; only Sensor 1 retains the true flight signal. This scenario models total, non-recoverable sensor failure. Noise Injection Methodology Noise was injected exclusively on Sensor 2 (T2, Pr2, H2, Hum2); Sensor 1 (T1, Pr1, H1, Hum1) retains the original recorded values in all three files. The injected values (noise) are drawn from a uniform distribution spanning the full physical range of each channel. The resulting noise amplitudes observed in the data are: Channel Normal operating range Noise amplitude under fault T2 20–27 °C 5–40 °C Pr2 51,000–101,000 Pa 0–150,000 Pa H2 0–5,200 m suppressed by the altitude scale Hum2 10–65 % −20–80 % In the mixed scenario, noise is applied in discrete time blocks, giving Sensor 2 a characteristic burst-and-recover pattern. In the rogue scenario, the same noise is sustained for the full duration with no recovery intervals. Supplementary Figures Three figures are included (fig_sensor_pairs_normal_cycle001.png, fig_sensor_pairs_mixed_cycle001.png, fig_sensor_pairs_rogue_cycle001.png), each showing the four redundant sensor pair time series (H1/H2, Pr1/Pr2, T1/T2, Hum1/Hum2) for the corresponding scenario over the full flight cycle. Normal: All four pairs overlap tightly. H1/H2 and Pr1/Pr2 are virtually identical. T1/T2 maintain a consistent natural offset of approximately 1 °C. Hum1/Hum2 co-vary with minor natural spread; humidity falls from ~65 % at ground level to ~12 % at cruise altitude and partially recovers on landing. Mixed: Burst noise is most visible in the Pr1/Pr2 and T1/T2 panels, where Sensor 2 alternates between wide-amplitude oscillations and brief periods of nominal agreement with Sensor 1. The Hum1/Hum2 panel shows the same burst pattern. H1/H2 divergence is less apparent because the noise amplitude is small relative to the 0–5,200 m altitude scale. Rogue: Sensor 2 forms a continuous wide-amplitude noise band in all four panels throughout the entire flight. Sensor 1 is visible only as a thin line within the noise envelope. No convergence occurs at any point.概览 本数据集收录了真实安东诺夫（Antonov）货运飞机飞行作业期间记录的秒级分辨率遥测时序数据，并补充了两类合成传感器欺骗场景，旨在作为单变量及多变量传感器时序数据上无监督异常检测方法的基准测试资源。 本数据集涵盖四类环境参数：海拔高度（altitude, H）、大气压强（atmospheric pressure, Pr）、温度（temperature, T）及相对湿度（relative humidity, Hum），每类参数均由两台独立传感器（传感器1、传感器2）同步采集。配对传感器间的一致性是核心异常信号：在健康运行状态下，冗余传感器的采集值高度贴合；当出现故障时，传感器2会偏离其配对参考传感器（传感器1）的采集结果。 传感器架构与异常信号 每类参数均配备一组冗余传感器对： | 配对参数 | 物理量 | 单位 | 传感器1（参考端） | 传感器2（监测端） | | ---- | ---- | ---- | ---- | ---- | | H | 气压高度 | 米 | H1 | H2 | | Pr | 大气压强 | 帕斯卡 | Pr1 | Pr2 | | T | 温度 | 摄氏度 | T1 | T2 | | Hum | 相对湿度 | 百分比 | Hum1 | Hum2 | 所有场景下，传感器1均作为可信参考端。异常表现为传感器间的偏差：当传感器2遭受欺骗时，|S1 − S2|会超出正常运行阈值。该基于冗余的检测原理可用于识别监控系统中异常或故障的传感器设备。 数据文件 所有数据文件均包含11425行（每秒对应一行）与9列数据，本版本发布的文件均未标注，未包含逐行异常标记列。 各列说明如下： | 列名 | 描述 | 单位 | | ---- | ---- | ---- | | time | Unix时间戳（Unix timestamp） | 秒 | | H1 | 传感器1采集的气压高度（参考端） | 米 | | H2 | 传感器2采集的气压高度（监测端） | 米 | | Pr1 | 传感器1采集的大气压强（参考端） | 帕斯卡 | | Pr2 | 传感器2采集的大气压强（监测端） | 帕斯卡 | | T1 | 传感器1采集的温度（参考端） | 摄氏度 | | T2 | 传感器2采集的温度（监测端） | 摄氏度 | | Hum1 | 传感器1采集的相对湿度（参考端） | 百分比 | | Hum2 | 传感器2采集的相对湿度（监测端） | 百分比 | 场景说明 1. flight001_normal.csv：正常运行场景（真实数据）。为未注入任何故障的真实飞行遥测数据，每对传感器的采集值在整个飞行周期内均保持一致。该文件代表健康运行状态，可作为无监督异常检测模型的训练基准。 2. flight001_mixed.csv：间歇性传感器欺骗场景（合成数据）。传感器2的各通道（T2、Pr2、H2、Hum2）会在周期性的通断间隔中接收突发噪声，而传感器1保留原始采集信号。在活跃欺骗时段（故障窗口），传感器2的采集值会在各通道的全物理量程内振荡（例如Pr2：0~150000帕；T2：5~40摄氏度；Hum2：-20~80%）。在噪声突发间隙，传感器2会短暂恢复至正常状态，与传感器1的采集值重新贴合。所有受影响的四个通道的突发模式保持同步。该场景模拟间歇性或瞬态退化的传感器行为。 3. flight001_rogue.csv：完全传感器欺骗场景（合成数据）。与混合场景相同幅值的噪声会在整个飞行过程中持续施加，无任何恢复窗口。从第一个采样点到最后一个采样点，传感器2的所有采集值均被完全篡改，仅传感器1保留真实飞行信号。该场景模拟完全且不可恢复的传感器故障。 噪声注入方法 噪声仅注入传感器2的各通道（T2、Pr2、H2、Hum2）；在所有三个数据文件中，传感器1（T1、Pr1、H1、Hum1）均保留原始采集值。 注入的噪声值取自覆盖各通道全物理量程的均匀分布，数据中观测到的故障噪声幅值如下： | 通道 | 正常运行量程 | 故障时噪声幅值 | | ---- | ---- | ---- | | T2 | 20~27摄氏度 | 5~40摄氏度 | | Pr2 | 51000~101000帕 | 0~150000帕 | | H2 | 0~5200米 | 受高度量程限制 | | Hum2 | 10~65% | -20~80% | 在混合场景中，噪声以离散时间块的形式施加，使传感器2呈现出典型的突发-恢复模式；在完全欺骗场景中，相同的噪声会在整个飞行周期内持续施加，无任何恢复间隙。 补充配图 本数据集附带三张配图（fig_sensor_pairs_normal_cycle001.png、fig_sensor_pairs_mixed_cycle001.png、fig_sensor_pairs_rogue_cycle001.png），每张图均展示了对应场景下全飞行周期内四组冗余传感器对的时序数据（H1/H2、Pr1/Pr2、T1/T2、Hum1/Hum2）。 1. 正常场景：四组传感器对的时序曲线均高度重合。H1与H2、Pr1与Pr2几乎完全一致。T1与T2保持约1摄氏度的恒定自然偏移。Hum1与Hum2存在小幅自然离散度：相对湿度从地面的约65%下降至巡航高度的约12%，并在着陆阶段部分回升。 2. 混合场景：Pr1/Pr2与T1/T2子图中的突发噪声最为明显，传感器2会在大幅振荡与短暂贴合传感器1的正常状态之间交替。Hum1/Hum2子图也呈现相同的突发模式。H1/H2的偏差相对不明显，因为相对于0~5200米的高度量程，噪声幅值较小。 3. 完全欺骗场景：在整个飞行周期内，所有四个子图中传感器2的采集值均呈现为连续的宽幅噪声带。传感器1仅表现为噪声包络内的一条细直线，全程未出现任何贴合现象。