CWRU Domain Generalization Version

The Case Western Reserve University (CWRU) is a widely adopted dataset for rolling bearing elements (Smith and Randall 2015). Accelerometer sensors were deployed at both the drive-end and fan-end of the housing motors. Vibration signals were collected with 12 KHz sampling rate under eight different operating conditions. Particularly, we have four different operating conditions with different loading torques collected from the drive end, denoted as domain A, B, C, and D. Similarly, we have other four operating conditions collected from the fan-end of the motor, denoted as domain F, G, H, and I. For each operating condition, there is one healthy state and three faulty states, namely, inner fault (IF), outer fault (OF), and bearing fault (BF). Each faulty state has three levels of severity with dimensions of 7, 14, 21 mil. In total, we have 10 classes with 1 healthy class and 9 faulty classes. To prepare the data for our experiments, we partitioned the sensor readings into smaller samples using sliding windows with a fixed length of 4,096 and the shifting step of 290, which is widely used in previous studies (Zhang et al. 2017; Ragab et al. 2021). Overall, we can generate 4,000 samples for each domain.凯斯西储大学（Case Western Reserve University, CWRU）发布的滚动轴承元件数据集是目前被广泛采用的公开数据集（Smith与Randall，2015）。实验中在驱动端与风扇端的电机外壳均部署了加速度传感器，以12kHz的采样率采集了8种不同工况下的振动信号。具体而言，从驱动端采集得到4种不同加载扭矩的工况，分别记为域A、B、C、D；同理，从电机风扇端采集得到另外4种工况，记为域F、G、H、I。每种工况均包含1种健康状态与3种故障状态，即内圈故障（Inner Fault, IF）、外圈故障（Outer Fault, OF）与滚动体故障（Bearing Fault, BF）。每种故障状态包含3种严重程度等级，故障尺寸分别为7、14、21密耳（mil）。总计包含10个类别：1个健康类别与9个故障类别。为适配本实验的数据需求，我们采用固定长度为4096、滑动步长为290的滑动窗口将原始传感器读数分割为更小的样本，该预处理方式已被此前的多项研究广泛采用（Zhang等，2017；Ragab等，2021）。最终每个域可生成4000个样本。

搜集汇总

背景与挑战

背景概述

该数据集是凯斯西储大学(CWRU)轴承故障数据的域泛化版本，专为故障诊断研究设计。数据包含从电机驱动端和风扇端采集的振动信号，覆盖八个不同工况，每个工况有10个类别（1个健康状态和9个故障状态，故障分为三种类型和三种严重程度）。数据经过滑动窗口预处理，每个域生成4000个样本，适用于域泛化方法验证。