Project's Infos & Updates

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. Jari Vepsäläinen (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