Frequency-Domain Ultrasonic Signal Dataset for Battery Electrode Thickness Prediction

📘 Dataset Description

This dataset contains frequency-domain ultrasonic signals of the backwall echo from lithium-ion battery electrodes, measured before and after the calendering process.
It is curated for research in ultrasonic signal processing, machine learning, and battery manufacturing, with a particular focus on predicting electrode thickness and analyzing how calendering affects ultrasonic responses.

⚙️ Dataset Overview

Electrode Types

🔵 Anode (Graphite): Samples labeled as GA_1, GA_2, GA_3, etc.

🟢 Cathode (NMC622): Samples labeled as NMC622_1, NMC622_2, NMC622_3, etc.

Calendering States

Each electrode sample includes two conditions:

before-calendering.json

after-calendering.json

Signals

Each sample includes the frequency-domain representation of the backwall echo (trimmed from the 7.5–10 µs portion of the original ultrasonic signal).

The signals have been transformed using the Fast Fourier Transform (FFT) and include two key arrays:

fft_frequency (in MHz): Represents the spectral frequency components of the signal.

fft_magnitude (normalized a.u.): Represents the amplitude or strength of each frequency component.

Only the positive frequency range (0.75–15 MHz) is retained to capture meaningful physical information while reducing redundancy.

🎯 Target Variable

The dataset also provides the average electrode thickness (µm), measured across three regions.

This value serves as the ground truth for regression and machine learning tasks.

🧠 Applications

🔹 Predicting electrode thickness using frequency-domain ultrasonic features.

🔹 Studying the impact of calendering on ultrasonic signal responses.

🔹 Benchmarking machine learning and deep learning models for electrode property prediction.

🔹 Exploring spectral characteristics, feature extraction, and frequency-domain signal analysis.

💾 Data Structure

Each .json file contains:

Metadata — Sample ID and process parameters (Web Speed (m/min), Roll Gap (μm), Coat Weight (gsm), Calendering Speed (mm/sec), Thickness (μm), and Density (g/cm³))

fft_frequency — Frequency values (MHz).

fft_magnitude — Corresponding normalized FFT amplitude.

Example filenames:

Anode/GA_1/before-calendering.json

Anode/GA_1/after-calendering.json

Cathode/NMC622_1/before-calendering.json

Cathode/NMC622_1/after-calendering.json

🧩 Supporting Scripts

📊 visualisation.py — A helper script to visualise FFT magnitude spectra and compare before and after calendering states.

🤖 Machine Learning Script (under Machine Learning/ directory) — Contain complete workflow for feature extraction, 5-fold cross-validation, and model training/testing using the FFT data and measured thickness.

📂 Intended Use

This dataset supports:

Academic research in battery electrode characterization,

Development of data-driven ultrasonic analysis models, and

Benchmarking of machine learning pipelines for industrial process monitoring.