Bay-Fi |

Project's Infos & Updates

Mon, 27 Oct 2025 00:00:00 +0000

Abstract

Structural Health Monitoring (SHM) and operational modal analysis rely on accurate identification of modal frequencies from vibration signals. However, in many practical scenarios, acquiring high-frequency signals is constrained by hardware limitations, energy consumption, or data transmission bandwidth — making Sub-Nyquist sampling a key challenge.

Bay-Fi (Bayesian Frequency Identification) addresses this challenge by introducing a novel general method that enables direct identification of the main modal frequency without requiring:

Signal reconstruction
Any prior tuning of hyperparameters

The proposed approach employs a Bayesian framework to optimize a curve-fitting algorithm applied to highly under-sampled and randomly acquired vibration signals. This allows robust frequency estimates even when conventional methods would fail due to aliasing or insufficient data.

Methodology

The core idea of Bay-Fi is to reformulate frequency identification as a Bayesian optimization problem. Rather than attempting to reconstruct the full signal from sparse measurements, the method directly estimates the dominant frequency by fitting a parametric model to the available samples in a probabilistic framework.

Key features of the method:

Parameter-free: no prior knowledge or manual tuning required
Robust to under-sampling: works with signals sampled well below the Nyquist rate
Random sampling compatible: handles non-uniform and randomly acquired samples
Scalable: applicable to single-channel and multi-channel measurements

AI-generated Podcast

Ever wondered what Bay-Fi would sound like if two enthusiastic hosts explained it over coffee? Wonder no more. (Aalto University, Finland) fed the paper to an AI and out came this surprisingly engaging podcast episode — Bayesian inference, Sub-Nyquist sampling, and all. Give it a listen!

Updates

Date	Update
Oct 2025	Project launched, initial development of the Bayesian framework

Publications and Talks

Mon, 27 Oct 2025 00:00:00 +0000

Journal Papers

J1. Bay-Fi: Approximate Bayesian main Frequency Identification for extremely undersampled signals

Conference Papers

C1. Bayesian Approach for Main Frequency Identification on Extremely Under-Sampled Signals — EVACES 2025, Porto

Talks and Presentations

#	Event	Location	Date	Status
T1		Porto, Portugal	Jul 2–4, 2025	✅ presented
T2		Naples, Italy	Jun 29 – Jul 2, 2027	🔜 upcoming

Code Repository

Mon, 27 Oct 2025 00:00:00 +0000

Bay-Fi

The Bay-Fi source code is openly available on GitHub under the MIT license.

Bay-Fi is a Bayesian framework for extracting main vibration frequencies from extremely sparse, under-sampled signals — without requiring signal reconstruction or parameter tuning. It achieves higher compression ratios than traditional compressed sensing approaches.

Repository Structure

Folder	Description
`BayFi/`	Core algorithm — `LIB.py` main library
`Classic-CS/`	MATLAB routines for convex optimization (comparison baseline)
`Examples/`	Application examples including a simply supported beam dataset
`Plotting/`	Visualization scripts
`Papers/`	Related conference and journal papers
`methodology.pdf`	Visual documentation of the Bayesian framework

Dependencies

numpy · scipy · tqdm · pyswarms

Python 3.13 — install via pip install -r requirements.txt

Authors

Chiara Nardin · Stefano Zorzi · Federica Zonzini · Marco Broccardo

💡 Star the repo on to follow updates!