LIMAN-C: Three-Phase Induction-Motor Current Measurements under Multiple Fault and Load Conditions
Description
LIMAN-C is a current-only experimental dataset for condition monitoring of three-phase induction motors under multiple source-defined motor states and nominal load settings. It contains 480 complete three-phase parent recordings (1,440 phase-specific CSV files): 128 from one healthy motor group, 136 from one motor group with a rotor-bar defect, 139 from one motor group with a bearing defect, and 77 from one motor group with inter-turn short circuits. Measurements were collected on 3 October 2024 in one campaign. The healthy, rotor-bar-defect, and bearing-defect groups cover nominal load settings 0, 20, 40, 60, 80, and 100; the inter-turn-short-circuit group covers 60, 80, and 100. Load was varied through generator parameter adjustments while motor speed was kept approximately constant. Each parent recording comprises three phase channels. Each phase file contains 16,384 samples over approximately 4 s at an effective sampling rate of approximately 4,096 Hz, as derived from the recorded time coordinate. Raw CSV bytes are unchanged; rare empty current-value cells are fully inventoried, and the frozen mapping and gap-repair policy used in the associated analysis is documented. The release provides the measurements, parent-level metadata, a data dictionary, channel and phase-convention documentation, file checksums, and the exact analysis roster used in the associated study. The four condition groups come from four source motor groups, with one declared state per group. Consequently, the dataset supports reference-versus-nonreference, source-motor-group out-of-distribution evaluation, benchmarking of condition-aware anomaly detection, motor-current signal processing, and studies of load-dependent spectral variation. It must not be interpreted as a same-motor before-and-after intervention, as causal evidence that a fault alone produced every between-group difference, or as an estimate of population-level generalization across motors. LIMAN LLC provided the experimental measurements and authorized their public release. The release was curated and documented by Aleksandr Khizhik, Saraa Ali, Artem Ryzhikov, and Denis Derkach at HSE University. The release is licensed under CC BY 4.0.
Files
Steps to reproduce
1. Download the complete release and, from its root directory, run python3 scripts/validate_release.py . The validator checks checksums.sha256, the inner current-file inventory, archive membership, metadata, and the exact inventoried empty cells. 2. Read metadata/recordings.csv, metadata/data_dictionary.md, and metadata/missing_current_values.csv. A recording_id denotes one complete acquisition and maps to exactly three phase-channel CSV files; the channels are dependent measurements, not independent samples. 3. Assemble each recording using its common time coordinate and apply the dataset-wide mapping in metadata/phase_mapping.yaml. Do not select a new phase order for individual records. 4. Raw empty current-value cells are intentionally preserved. To reproduce the associated study, apply analysis/publication_v1_preprocessing_policy.json: perform the locked mapping first, linearly repair internal mapped gaps of at most three samples, use the stated edge rule if applicable, and reject longer gaps. Do not replace values ad hoc. 5. Use load_condition only as a dataset-local ordered operating context. Do not interpret it as measured torque or as a percentage of rated load. 6. Preserve source_group_id in all derived tables and splits. The healthy and three nonreference states belong to four source motor groups; do not construct same-motor healthy/fault pairs. 7. To reproduce the associated study population, use analysis/publication_v1_roster.csv and analysis/publication_v1_fold_assignments.csv; keep all channels and derived windows from one recording in the same fold. Run the associated versioned analysis code without changing the roster or phase convention.
Institutions
- National Research University Higher School of EconomicsMoscow, Moscow