Experimental vibration analysis and digital signal processing work carried out during my MSc in Mechanical Engineering at Politecnico di Torino (Vibration Mechanics).
Both projects follow the same pipeline that underpins industrial condition monitoring: acquire accelerometer data → pre-process → transform to the frequency domain → extract modal parameters → validate against a physical model.
Experimental modal analysis of a scaled bridge structure (simply-supported aluminium beam, 186 × 15 × 1.5 cm).
- Test design: 8 accelerometers positioned off the modal nodes, with off-axis placement to discriminate torsional modes from bending modes
- Three impulse-response acquisitions at 256 Hz and 512 Hz sampling frequency, 20 s each
- Time-domain inspection of all 8 channels, windowing of the decaying response, channel selection by noise level
- Power Spectral Density (per-channel and summed) for preliminary identification of eigenfrequencies — 8 candidate resonance peaks
- Complex Exponential Method implemented in MATLAB for modal parameter estimation (eigenfrequencies, damping ratios, poles)
- Stabilization diagram to separate physically stable modes from computational modes caused by numerical error, noise and model over-ordering
- Mode shape reconstruction from amplitude and phase across the measurement grid
- Cross-validation against the analytical Euler–Bernoulli solution
Identified modes: 9.91 / 41.22 / 77.60 / 119.27 Hz, consistent with the PSD peak estimates (10.07 / 41.13 / 77.66 / 119.37 Hz).
Free vibration of a cantilever beam with a tip mass — SDOF identification from acceleration data.
- Aluminium cantilever (3 × 30 mm section) with lumped tip mass, two accelerometers, sampling frequency 2048 Hz
- Logarithmic decrement applied over 10 periods on the log-envelope of the acceleration signal to estimate the modal damping ratio
- Damped and undamped natural frequency estimation from the identified period and damping ratio
- FFT of the free response to confirm the dominant frequency and identify measurement noise contributions
- Parametric campaign over three configurations (tip mass 0.2 / 0.46 kg, free length 37 / 45 cm) quantifying the effect of mass and stiffness on the first natural frequency
- Comparison with continuous Euler–Bernoulli beam theory including the added-mass boundary condition
| Area | Techniques |
|---|---|
| Acquisition | Multi-channel accelerometer setup, sampling frequency selection, sensor placement w.r.t. modal nodes |
| Time domain | Impulse/free-response analysis, log-envelope, logarithmic decrement, record trimming |
| Frequency domain | FFT, Power Spectral Density, resonance peak identification, noise discrimination |
| System identification | Complex Exponential Method, stabilization diagrams, pole/damping estimation, model-order selection |
| Modal analysis | Eigenfrequencies, damping ratios, mode shapes from amplitude and phase |
| Validation | Analytical Euler–Bernoulli model, experimental vs. theoretical comparison |
Tools: MATLAB
Extracting damping, resonances and mode shapes from noisy accelerometer signals is the foundation of vibration-based condition monitoring and predictive maintenance for rotating machinery and civil/industrial structures. The same chain — acquire, condition, transform, extract features, validate against physics — is what I now apply to data-driven and AI-based diagnostics of industrial assets.
Giuseppe Luciano Scrimali MSc Mechanical Engineering, Politecnico di Torino · Specializing Master in Artificial Intelligence (in partnership with NVIDIA)