Synthetic Reaction Wheel Datasets for Nominal and Faulty Conditions as Used for Training the One-Dimensional Sliding Window Residual Network Proposed

All of the datasets provided here have been generated in MATLAB. The datasets describe the measurement outputs of the continuous operation of a reaction wheel based on Bialke's high fidelitiy RW model onboard a satellite that undergoes different faults.

If you import any of these datasets into MATLAB, a cell-type variable called "dataset" will show up.

You can also load them for use in Python and other usages.

Datasets #1-#3, #4-#6, and #7-#9 are grouped together and are describing the same signals from different perspectives, therefore,

they have the same number of cells. For example, the 10th cells of datasets #1-#3 describe the same signal.

More information about the datasets is available in Table III in the referenced paper.

For any of the MJAPF and SPF datasets, each cell is of the form 2x1001 doubles.

For these datasets, the first row is the descriptor vector of the signal and the second row corresponds to the signal

as perceived by MJAPF or SPF. Both signals have 1001 time steps. The time steps are 0.1 seconds apart, meaning that 1001 encompasses operational details of the reaction wheel over about 100 seconds.

The raw datasets' cells are of the form 5x1001 doubles.

The first row is, again, the descriptor row.

The second row is the reaction wheel's motor current as received by sensors (with sensor noise).

The third row is the reaction wheel's angular velocity as received by sensors (with sensor noise).

The fourth row is the reaction wheel's motor current with process noise but without the measurement noise (true signals).

The fifth row is the reaction wheel's angular velocity with process noise but without the measurement noise (true signals).

Therefore, the datasets are structured as follows:

1) For MJAPF- or SPF-treated datasets:
First row: Descriptors that describe the health state of the reaction wheel at each time step and have no units since it is a categorical variable.

Second row: The reaction wheel motor's electrical current (I_m), as estimated by MJAPF or SPF, has the unit of Amperes (A).

2) For raw datasets:
First row: Descriptors that describe the health state of the reaction wheel at each time step and have no units since it is a categorical variable.

Second row: The reaction wheel motor measures the electrical current (I_m) without undergoing any treatments and has the unit of Amperes (A).

Third row: The reaction wheel measures angular velocity (omega_m) without undergoing any treatments and has the unit of Radians per second (rad/s).

Fourth row: Reaction wheel motor's real or true process electrical current (I_m) without undergoing any treatments and has the unit of Amperes (A).

Fifth row: Reaction wheel's real or true process angular velocity (omega_m) without undergoing any treatments and has the unit of Radians per second (rad/s).

本文提供的所有数据集均基于MATLAB生成。这些数据集描述了卫星搭载的、基于比尔克（Bialke）高保真反作用轮（RW）模型的反作用轮持续运行时的测量输出，该反作用轮会经历多种故障工况。

若将任意此类数据集导入MATLAB，将出现一个名为"dataset"的元胞类型变量。

你也可将其加载至Python及其他环境中使用。

数据集#1~#3、#4~#6以及#7~#9分别归为一组，它们从不同视角描述同一组信号，因此元胞数量完全一致。例如，数据集#1~#3的第10个元胞对应同一信号。

有关该数据集的更多细节可参见参考文献中的表III。

对于MJAPF与SPF处理后的数据集，每个元胞均为2×1001的双精度数组。此类数据集的第一行是信号的描述向量，第二行对应经MJAPF或SPF处理后的信号。两类信号均包含1001个时间步，各时间步间隔为0.1秒，即1001个时间步可覆盖反作用轮约100秒的运行细节。

原始数据集的元胞则为5×1001的双精度数组，其第一行同样为描述行。第二行是传感器采集到的反作用轮电机电流（含传感器噪声）；第三行是传感器采集到的反作用轮角速度（含传感器噪声）；第四行是仅含过程噪声、无测量噪声的反作用轮电机真实电流；第五行是仅含过程噪声、无测量噪声的反作用轮真实角速度。

据此，数据集的结构如下：
1) 经MJAPF或SPF处理的数据集：
第一行：描述各时间步下反作用轮健康状态的描述符，为分类变量，无单位；
第二行：经MJAPF或SPF估计得到的反作用轮电机电流（I_m），单位为安培（A）。

2) 原始数据集：
第一行：描述各时间步下反作用轮健康状态的描述符，为分类变量，无单位；
第二行：未经任何处理的传感器采集到的反作用轮电机电流（I_m），单位为安培（A）；
第三行：未经任何处理的传感器采集到的反作用轮角速度（ω_m），单位为弧度每秒（rad/s）；
第四行：未经任何处理的反作用轮电机真实过程电流（I_m），单位为安培（A）；
第五行：未经任何处理的反作用轮真实过程角速度（ω_m），单位为弧度每秒（rad/s）。