Joint Diagnosis of Mechanical Faults Using Vibration, Acoustic, and Current Signals - Blog

Combining vibration, acoustic, and current signals for mechanical fault diagnosis is essentially a multi-source sensor information fusion problem. Each signal modality captures a different physical aspect of machine health, and their complementary strengths make joint analysis more reliable than any single source.

Vibration analysis remains the most mature baseline.

Accelerometers capture time-domain features such as kurtosis and peak values, as well as frequency-domain signatures via FFT and envelope spectra, making it effective for detecting bearing wear, rotor imbalance, and misalignment . However, vibration sensors are insensitive to electrical faults like winding insulation degradation and may miss early-stage defects before mechanical symptoms manifest.

Acoustic sensing, particularly acoustic emission (AE), detects high-frequency stress waves generated by crack propagation and friction. AE is highly sensitive to incipient damage, often flagging faults before vibration signatures become pronounced . Its limitations include susceptibility to industrial background noise and complex signal propagation paths, which complicate feature extraction.

Motor current signature analysis (MCSA) offers a non-invasive electrical perspective. Faults in bearings, rotor bars, and gearboxes modulate the stator current, producing characteristic sidebands around supply frequency . MCSA requires no sensors on the motor shaft, making it valuable for hazardous or enclosed environments, though interpretation demands expertise and is sensitive to load variations.

The fusion strategy typically proceeds from feature-level integration to decision-level learning.

Time-domain statistics, frequency spectra, and time-frequency representations are extracted from each modality, then combined via concatenation, attention mechanisms, or Bayesian inference frameworks. Recent research demonstrates that deep cross-modal fusion-such as Transformer-BiGRU architectures-can capture nonlinear correlations across heterogeneous signals and quantify diagnostic uncertainty . A multi-channel fusion approach using vibration-acoustic and vibration-current signal pairs achieved recognition accuracy exceeding 99% on rolling bearing and gearbox datasets.

The practical diagnostic logic follows a hierarchical pattern

Acoustic signals often provide the earliest warning of incipient cracks; vibration confirms mechanical degradation patterns; current analysis corroborates electrical and load-related manifestations. Joint decision-making reduces false alarms and missed detections, particularly when one modality is corrupted by noise or operating conditions. For engineering deployment, edge computing enables local feature extraction to manage the high data rates of AE sensors, while synchronized sampling across modalities ensures temporal alignment for reliable fusion.

VTZ-12 630A Vacuum Circuit Breaker