NRF & TÜBİTAK Funded · 2027–2028

KTBIX
Battery Intelligence
eXchange

Korea–Türkiye joint research on Physics-Informed Variational Information Bottleneck (PI-VIB-ResNet) for lithium-ion battery State-of-Health estimation via Electrochemical Impedance Spectroscopy.

⚡ About KTBIX

PI-VIB-ResNet Pipeline

📡

Raw EIS Data

Nyquist spectra from T-EV & TOGG EVs

▼

🌊

Wavelet Denoising

Multi-level noise suppression

▼

⚖️

Feature Scaling

Normalization & impedance mapping

▼

🧠

VIB Latent Extraction

Physics-informed information bottleneck

▼

🔗

ResNet-BPNN

Residual deep neural regression

▼

🔋

SOH Estimate

RMSE ≤1.2% — BMS ready output

Nations

Universities

Months

Work Packages

≥2

Q1 Papers

≤1.2%

Target RMSE

≥15%

SOH Improvement

About the Project

Advancing Battery Diagnostics Through AI & Physics

KTBIX (Korea–Türkiye Battery Intelligence eXchange) is a 24-month bilateral mobility project funded by NRF Korea and TÜBİTAK Türkiye. It unites Sejong University (Seoul) and Istanbul Technical University (Istanbul) to tackle a critical challenge in next-generation electric mobility: accurate, robust battery State-of-Health estimation under real-world field conditions.

"We transition from our published AE-BPNN baseline (Scientific Reports, 2025) to a Physics-Informed Variational Information Bottleneck architecture — directly addressing the noise-entanglement and generalizability limitations of purely data-driven EIS models."

The project leverages real EV field data from Turkish T-EV platforms and TOGG vehicles (WP1), combines it with a novel PI-VIB-ResNet model developed at Sejong University's AINTLab (WP2), and validates against internationally recognized Oxford and NASA battery datasets (WP3).

Baseline Paper (2025)

Nyquist Plot — EIS Signature

~1.8%

Baseline RMSE

AE-BPNN (2025)

≤1.2%

Target RMSE

PI-VIB-ResNet

Datasets

Oxford, NASA, T-EV

Chemistries

NMC, LFP, etc.

Research Highlights

Six Pillars of KTBIX

From physics-constrained deep learning to open industrial datasets — KTBIX advances every layer of the battery SOH estimation stack.

🔬

Core Innovation

Physics-Informed VIB

The Variational Information Bottleneck disentangles measurement noise from structural EIS features while a Loss_Physics term enforces electrochemical consistency via Randles circuit constraints.

🏗️

Architecture

ResNet-BPNN Architecture

Residual skip connections prevent gradient vanishing in deep BPNN layers, enabling 15+ layer depth for complex impedance-to-SOH mappings across battery chemistries and aging states.

🗃️

WP1 — ITU

Industrial Field Data

Unlike lab-only datasets, KTBIX acquires EIS measurements from real Turkish EV fleets (T-EV platforms) and TOGG vehicles, covering diverse duty cycles and climate conditions.

🧩

WP3 — Joint

Explainable AI (XAI)

SHAP attributions and gradient saliency maps reveal which impedance features (R_ct, W, CPE) drive SOH predictions — essential for BMS engineers and regulatory validation.

⚡

WP3 — Joint

Benchmark Validation

Rigorous comparison against >5 state-of-the-art baselines including SVR, LSTM, Transformer-SOH, and conventional EIS equivalent circuit fitting on Oxford and NASA datasets.

🌐

WP4 — Joint

Bilateral Knowledge Exchange

Two international workshops (Istanbul 2027; Seoul 2028), researcher exchange visits, joint PhD co-supervision, and open-source code and dataset releases via GitHub and Zenodo.

Project Structure

Work Packages

Four coordinated work packages over 24 months with measurable milestones, dual-institution leads, and concrete exchange deliverables.

WP1ITU — Istanbul Technical University

Field EIS Data Acquisition

Deploy impedance analyzers on Turkish T-EV platforms and TOGG electric vehicles. Acquire EIS spectra across diverse temperature ranges and aging states. Deliver curated open dataset.

⏱ M1 – M8

WP2SJU — Sejong University · AINTLab

PI-VIB-ResNet Model Design

Design and train the Physics-Informed Variational Information Bottleneck encoder paired with a deep ResNet-BPNN regressor. Ablation studies and hyperparameter optimization.

⏱ M4 – M14

WP3Joint — ITU + SJU

Benchmark Validation & XAI

Cross-validate PI-VIB-ResNet against Oxford, NASA, and T-EV datasets. Compare with 5+ baselines. Apply SHAP and gradient saliency for explainability. Publish ≥2 Q1 papers.

⏱ M10 – M22

WP4Joint — ITU + SJU

Knowledge Exchange

Researcher exchange visits (ITU↔SJU, min. 2 weeks each direction), two international workshops (Istanbul M10, Seoul M22), PhD co-supervision, open-source code releases.

⏱ M1 – M24

Research Team

Principal Investigators

A cross-continental partnership grounded in a prior exchange visit (Prof. Syafrudin's visit to Türkiye, funded by Sejong University) and a joint 2025 publication.

🇰🇷

Principal Investigator · Korea

Prof. Muhammad Syafrudin

Sejong University (SJU) · AINTLab · Seoul, Korea

Deep LearningApplied IntelligenceeXplainable AI (XAI)Battery AI

SJU Profile ↗

🇹🇷

Principal Investigator · Türkiye

Prof. Muhammet Tahir Güneşer

Istanbul Technical University (ITU) · Istanbul, Türkiye

Electrical EngineeringElectric VehiclesEnergy StorageSignal Processing

ITU Profile ↗

📄

Foundation Publication (2025)

"AE-BPNN: autoencoder and backpropagation neural network-based model for lithium-ion battery state of health estimation." — Scientific Reports, Nature Publishing Group.

Read Paper →

Knowledge Exchange

International Workshops

Two flagship workshops — one on each side of the collaboration — open to the global battery research community. *The exact date to be updated later.

2027

Istanbul, Türkiye · M10

Integrating Industrial Field Data into Battery Management Systems

Host: Prof. Muhammet Tahir Güneşer · ITU

📍Istanbul Technical University (ITU), Istanbul

🗓️October 2027 (Month 10 of the project)*

👥International participants — Korea, Türkiye, EU partners

📢Open registration via ktbix.org

2028

Seoul, Korea · M22

eXplainable AI (XAI) for Second-Life Battery Prediction

Host: Prof. Muhammad Syafrudin · Sejong University / AINTLab

📍Sejong University (SJU), Seoul, Korea

🗓️October 2028 (Month 22 of the project)*

👥International participants — Korea, Türkiye, ASEAN partners

📢Open registration via ktbix.org

Project Timeline

24-Month Roadmap

Three strategic phases connecting data to discovery to dissemination.

Phase I · 2027 H1

Data Acquisition & Model Foundations

Deploy EIS hardware on T-EV/TOGG platforms. Collect baseline measurements. Begin PI-VIB encoder design. Exchange visit: ITU researchers to SJU (min. 2 weeks). M3: WP1 Protocol Report.

Phase II · 2027 H2–2028 H1

Model Training, Validation & Istanbul Workshop

Full PI-VIB-ResNet training on T-EV data. Cross-validation on Oxford & NASA datasets. SHAP explainability. Istanbul Workshop (M10). Exchange visit: SJU researchers to ITU. Submit first Q1 paper.

Phase III · 2028 H2

Dissemination, Seoul Workshop & Phase 2 Proposal

Final benchmark comparison (5+ baselines). Seoul Workshop (M22). BMS hardware integration demo. Open-source release on GitHub + Zenodo. Submit second Q1 paper. Prepare Phase 2 grant proposal.

Partners & Funders