A Novel Monitoring Method of a Multi-Stage Planetary Gearbox Low Frequency Dynamic Behaviour Using Piezoelectric Strain Sensor

Abstract Background Condition-Based Monitoring (CBM) of planetary gearboxes has traditionally relied on vibration analysis via accelerometers. However, strain sensing, particularly using modern piezoelectric strain sensors, remains underutilized despite its potential to directly measure structural dynamics, especially in low-frequency ranges critical for multi-stage gear systems. Purpose This study introduces and validates a novel monitoring approach using a piezoelectric strain sensor to capture the low-frequency dynamic behaviour of a multi-stage planetary gearbox, demonstrating its advantages over conventional accelerometer and Lazer vibrometer measurements. Methods A three-stage planetary gearbox test rig was instrumented with a piezoelectric strain sensor (RHM240A02), an ICP accelerometer, and a Lazer vibrometer under no-load operating conditions. Data were collected across a range of input speeds. The strain signal was analysed in both time and frequency domains, with performance quantitatively benchmarked against the other sensors using metrics such as Low-Frequency Energy Ratio (LFER) and Identification Score (IS). Results The time-domain strain signal enabled the direct visual identification of 10 out of 15 key characteristic frequencies, including carrier and planet-pass frequencies, a result not achievable with accelerometer data. Subsequent spectral analysis identified two additional frequencies, yielding a total identification rate of 80%. The strain sensor exhibited a 12.6 times higher low-frequency energy content compared to the Lazer vibrometer and demonstrated excellent SNR and repeatability. Conclusions Piezoelectric strain sensing is established as a superior and complementary tool for profiling the low-frequency dynamics of multi-stage planetary gearboxes. It provides a rich, high-fidelity signal that simplifies feature extraction and offers valuable empirical data for model validation, though it is less suitable for high-frequency fault diagnostics such as bearing defects.