# A dataset of acoustics emissions recordings of woodboring insects in wood and cultural objects, context images and remarks

## TL;DR

This paper introduces a dataset of acoustic emissions from woodboring insects in wood and cultural objects, including recordings, images, and analysis tools for machine learning and conservation research.

## Contribution

The novel contribution is a comprehensive dataset combining acoustic emissions, contextual images, and metadata from multiple institutions for heritage conservation and machine learning applications.

## Key findings

- The dataset includes 440.9 hours of acoustic emission recordings from four institutions with infestation labels.
- It supports supervised machine learning model development through statistical features and binary classification.
- Reference sensor data enables ambient noise analysis and noise filtering method development.

## Abstract

This dataset presents acoustic emission (AE) recordings collected from woodboring insect-infested and non-infested wood samples and cultural heritage objects. Data acquisition was conducted across four institutions: Haute École Arc (HE-Arc), Switzerland; Canadian Museum of History (CMH), Canada; National Gallery of Canada (NGC), Canada; and Musée National de l'Automobile (MNA), France; from April to July 2025.

The recordings were captured using Vallen VS900-M sensors with AEP5 preamplifiers set to 34dB gain and AMSY-6 4-channel chassis, employing continuous acoustic emission monitoring at 2 MHz sampling rate. Each experiment utilized three sensors positioned on test objects and one reference sensor facing up to record ambient noise conditions. The dataset comprises approximately 440.9 hours of recordings distributed across the four collection sites.

The dataset includes four main components: raw Vallen AE database files (.tradb format), processed statistical data exported as CSV files, contextual images documenting setups and sensor placements, and Python script for statistical data processing. Each experiment is documented with duration, material specifications, coupling methods (renaissance wax, cyclododecane, or mechanical fastening), environmental conditions, and infestation labels.

The dataset's structure enables multiple research applications. The time-series statistical features and binary classification labels (infested/non-infested) provide a foundation for supervised machine learning model development. The diverse experimental conditions across four geographic locations, varying coupling methods, and different ambient environments offer opportunities to evaluate model generalization and robustness. Reference sensor recordings captured simultaneously with each experiment allow for ambient noise characterization studies and development of noise filtering methodologies. The combination of raw acoustic data and contextual documentation makes this dataset suitable for comparative studies of different signal processing approaches and feature extraction techniques in acoustic emission analysis for heritage conservation applications.