Four-Channel Vibration Dataset for Cross-Speed Rotating-Machinery Fault Diagnosis at 600, 900, and 1200 RPM

This dataset contains 3,545 synchronized one-second vibration samples acquired from an experimental rotating system under normal, imbalance, and misalignment conditions at rotational speeds of 600, 900, and 1200 RPM. Each sample contains four radial vibration channels sampled at 25.6 kHz, comprising two orthogonal measurement directions at each of two sensing locations. The dataset includes 1,200 imbalance samples, 1,176 misalignment samples, and 1,169 normal samples. Each MATLAB file contains the four-channel vibration array together with the sampling frequency, operating speed, condition label, acquisition-level identifier, and within-acquisition second identifier. The release is organized into nine condition-speed archives. It includes a complete sample manifest, class-speed distribution, archive inventory, SHA-256 checksums, and predefined acquisition-disjoint partitions for three multi-source leave-one-speed-out scenarios and six single-source cross-speed scenarios. These partitions prevent samples originating from the same acquisition from appearing in both source training and validation sets. The dataset supports research on rotating-machinery fault diagnosis, cross-speed domain generalization, tacholess diagnosis, multisensor vibration analysis, interpretable machine learning, signal processing, and predictive maintenance. Reference implementation and evaluation code are available at https://github.com/yakoopqasim/SRO-Net.