Multi-Sensor Condition-Monitoring Dataset of a Brushed DC Servo Motor
Description
***Disclaimer: Still in Draft - missing some files! Raw multi-sensor recordings from a brushed permanent-magnet DC servo motor (3PI12.12) driven by a 4-quadrant thyristor (SCR) converter. Each sensor data was recorded separately under the same operating conditions — matched load level and mechanical condition — using four sensors: armature current, an AV-160B vibrometer probe, a budget Android phone microphone, and vibrometer spot readings. Hypothesis. The dataset is built to test whether ordinary smartphone audio can replace invasive or specialised diagnostic equipment (current probes, contact vibrometers) for motor condition monitoring. With the phone recorded at ~1 m under the same conditions as the instrument-grade references, researchers can compare models trained on phone audio against those trained on current and vibrometer signals — i.e. whether a phone alone can estimate load and tell apart normal operation, direction reversal, and a loose foundation. Published as recorded (raw, untransformed). It also suits load estimation, foundation-looseness detection, direction-reversal analysis, and converter/commutation signature studies. The readme file gives ML pipeline suggestions only as guidance.
Files
Steps to reproduce
The 3PI12.12 motor (see Section 2) was driven by a 4-quadrant thyristor (SCR) converter with armature voltage/current control. Mechanical load was applied with a coupled load unit and set to 1, 2, 5, 10, 20, 30, 40, 50, 60, 70, 80, 90 and 100 % of rated power. Conditions. The full load sweep was repeated for each condition in Section 3 (normal without reversal, normal with periodic direction reversal, and a deliberately loosened foundation without reversal). Acquisition (each sensor recorded separately, under the same operating conditions — not simultaneously): Armature current — Rigol MSO5074 oscilloscope, saved as native binary waveforms (.bin); sample rate and scaling are in each file header (~20 s per capture). Vibration waveform — AV-160B vibrometer external piezoelectric probe; the 2.0 V AC analog output recorded as 44.1 kHz / 16-bit stereo WAV (~20.5 s). Vibration spot readings — the same AV-160B in display mode (velocity, acceleration, displacement per ISO 2954), logged to XLS. Acoustic — a budget Android smartphone microphone ~1 m from the machine, saved as M4A (AAC). Procedure. For each condition and load level the motor was brought to steady state, then each sensor was recorded in turn. Files are named load{NNN}_{sensor} and organised under data/<condition>/<sensor>/; see metadata.csv for the full inventory with sample rates and durations.
Institutions
- Technical University of SofiaSofia-Capital, Sofia