Signal processing in Vibration analysis

What this presentation is about and why it matters

This talk surveys the signal processing methods used to monitor rotating machinery by measuring vibration. Christophe Blouet connects practical problems—safety, maintenance planning, and passenger comfort—to the DSP tools engineers use to detect faults in shafts, gears, and bearings. The material is practical: it explains how simple sensors (accelerometers, magnetic pickups, optical sensors) feed algorithms that estimate speed, isolate shaft-synchronous vibration, reveal gear mesh components, and expose bearing defects using statistical indicators.

Why this matters: vibration analysis is often the earliest and most reliable indicator of mechanical failure. For aircraft and rotorcraft (but also turbines, pumps, electric motors and industrial drives) the difference between timely maintenance and catastrophic failure can be a matter of life, cost, and operational availability. The talk highlights methods used in real systems (onboard engine monitoring units and recorder systems) and ties signal-processing choices to operational constraints like variable speed, limited sensors, and limited onboard compute.

Who will benefit the most from this presentation

• DSP engineers who want concrete examples of algorithms applied to rotating machinery (order-tracking, TSA, demodulation).
• Condition-monitoring engineers and maintenance engineers who need to understand how vibration features are computed and interpreted.
• Students and researchers learning practical spectral methods and statistical fault indicators.
• Systems engineers designing onboard monitoring electronics or data pipelines for offline analysis.

What you need to know

The talk assumes a basic familiarity with time-domain and frequency-domain signal processing. Below are the compact background topics that will make the presentation clearer.

Signals and sensors

• Accelerometer vs velocimeter: accelerometers measure acceleration (useful at higher frequencies), velocimeters measure velocity (useful at mid-band), and displacement analysis is used at very low frequencies.
• Sensors used for rotational reference: optical or magnetic pickups provide a once-per-rev timing mark for synchronization and speed estimation.

Sampling and spectral analysis

• FFT/DFT: converting a sampled time signal into frequency bins. Be aware of windowing and spectral leakage when measuring narrow tones.
• Nyquist and aliasing: choose sampling rates high enough to capture the highest frequency content of interest (gear meshes and bearing impacts can reach many kHz).

Filters and tracking

• Band-pass, high-pass and low-pass filters are used for wide-band vibration indicators. FIR filters preserve linear phase (useful when you must keep phase relationships).
• Tracking filters and resampling: when shaft speed varies, you either resample the vibration into an angle domain (order domain) or use filters that follow the changing rotational frequency.
• Demodulation: multiply-by-sinusoid and low-pass to extract amplitude and phase of specific harmonics (efficient when only a few orders are needed).

Statistical and time-synchronous methods

• Time Synchronous Averaging (TSA): align and average signal revolutions to enhance shaft-synchronous components and suppress asynchronous noise.
• Kurtosis and spectral kurtosis: kurtosis detects impulsive events (bearing impacts). A compact definition is kurtosis = \(\dfrac{\mathbb{E}[(X-\mu)^4]}{(\mathbb{E}[(X-\mu)^2])^2}\), and spectral kurtosis finds frequency bands with impulsive energy to design targeted filters.

Glossary

• Accelerometer: sensor that measures acceleration, commonly used for vibration acquisition.
• FFT (Fast Fourier Transform): algorithm to compute DFT efficiently; converts time signals to frequency bins.
• Order / Order-Tracking: components of vibration expressed as multiples of shaft rotation (1×, 2×, …); order-tracking extracts these even when speed changes.
• Time Synchronous Averaging (TSA): resampling/averaging technique that aligns revolutions to enhance synchronous components.
• Zero-crossing: simple method to timestamp a periodic reference (used to estimate shaft speed); interpolation improves angular accuracy.
• Envelope detection: rectify and low-pass a high-frequency signal to reveal amplitude-modulated patterns (useful for bearings).
• Kurtosis: statistical indicator sensitive to impulsive events; high kurtosis often indicates bearing faults.
• Spectral kurtosis / Kurtogram: frequency-dependent kurtosis map used to find optimal bands for filtering impulsive faults.
• Demodulation: multiply-by-sinusoid and low-pass to recover amplitude/phase of a harmonic or order.
• Resampling / Angular sampling: changing the effective sampling grid so data is read at uniform angular intervals rather than uniform time intervals.

Final note — why this is worth your time

Christophe Blouet gives a compact, experience-grounded tour of the signal-processing toolbox that professionals use for vibration-based condition monitoring. The talk balances practical sensors and system constraints with algorithms (from simple band-pass filters to TSA, order-tracking, and spectral kurtosis). If you want to understand how DSP choices translate to real-world detection and maintenance decisions—especially in safety-critical domains like aerospace—this presentation will give you a useful map and motivate deeper reading into the specific methods most relevant to your application.

This overview is AI-generated from the session transcript. Spot an issue? Let us know.