Defesa de Tese: Aprendizado de Máquina Aplicado ao Prognóstico e Estado de Saúde de Células de Armazenamento de Energia

Item: Defesa de Tese: Aprendizado de Máquina Aplicado ao Prognóstico e Estado de Saúde de Células de Armazenamento de Energia
Rating: 4
Author: PostSingularity Institute

by Giovane Ronei Sylvestrin

Applied Sciences & Engineering Energy THYEnergy, power generation, distribution and storage THYBBattery technology

🎙 Giovane Ronei Sylvestrin 👥 305 📅 October 31, 2025 ⏱ 62 min 👁 14 📄 original study 🧭 2026-08-16

Available in: English (current) Français

Keywords

batterystate of healthremaining useful lifemachine learningdeep learning

Summary

The thesis defense presents a comprehensive study on applying machine learning and deep learning to estimate the state of health (SOH) and remaining useful life (RUL) of lithium-ion battery cells, with a focus on second-life applications. The research is structured in three main axes: a systematic literature review using the Proknow-C methodology, the development of a scalable and reproducible feature engineering pipeline, and the prediction of RUL using a short history of cycles. The systematic review analyzed 534 articles, identifying 20 public datasets and mapping current trends in ML/DL for SOH estimation. The feature engineering pipeline extracted over 40,000 variables from seven groups, including charge/discharge, CC/CV, and incremental capacity analysis, and employed univariate and multivariate selection methods to identify 773 finalist attributes. For RUL prediction, the study evaluated 12 deep learning architectures (CNN-LSTM/GRU) and a tree-based boosting model, as well as a stacking ensemble, on the MIT Battery Dataset (LiFePO4/graphite). The results showed that the stacking ensemble achieved a MAPE of ~6.0% and RMSE of ~40.6 cycles, outperforming individual models and positioning the approach at the upper level of recent literature. The research also demonstrated that capacity prediction with multiple horizons achieved MAPE below 1%. The thesis highlights the practical applicability of the method for accelerating second-use screening and supporting maintenance decisions with reduced testing time.

216 words

Critical Evaluation

Value of the Information & Strength of the Argument

The value of the information is high, as the research addresses a relevant problem (battery second-life) and provides a systematic, reproducible methodology. The argumentation is solid, supported by a systematic review, a large-scale feature engineering approach, and rigorous evaluation on a public dataset. The presentation clearly explains the motivation, gaps, methodology, and results, and the conclusions are well-aligned with the evidence presented.

Scientific Rigor, Source Quality, Title Accuracy

The scientific rigor is high, with a systematic review methodology (Proknow-C) and the use of a public dataset (MIT Battery Dataset) for reproducibility. The sources cited are primarily academic papers identified in the review, and the presentation does not rely on unverified claims. The title accurately reflects the content, and the presentation is well-structured and transparent about the methods and limitations.

138 words

Title / Content Match

The title accurately reflects the content, which focuses on machine learning for battery state-of-health prognosis and estimation.

Quality & Reliability

8/10

The thesis defense presents a systematic and reproducible methodology, including a systematic review (Proknow-C), a large-scale feature engineering pipeline, and rigorous evaluation on a public dataset (MIT Battery Dataset). The results are quantified (MAPE <1% for capacity, ~6% for RUL) and compared with literature. The presentation is detailed and transparent about methods and limitations.

Key Moments

Markers derived by PSI from the transcript: the creator did not define chapters.

Cited Sources

MIT Battery Dataset — Public dataset used for training and evaluation of the models.

Concurring Sources

MIT Battery Dataset — Public dataset used for training and evaluation of the models.

Contribution & Novelties

The thesis contributes a systematic and reproducible pipeline for battery state-of-health estimation, integrating multiple feature domains and demonstrating that a stacking ensemble of boosting and deep learning can achieve competitive RUL prediction with a short history of cycles. The systematic review using Proknow-C provides a comprehensive overview of the field, and the feature engineering pipeline is a valuable resource for future research.

Pour aller plus loin :

Proknow-C methodology — The systematic review methodology used in the thesis.
Remaining Useful Life prediction — Overview of RUL prediction concepts.
State of Health (SOH) of batteries — Definition and estimation methods.

99 words

Radar Profile

The radar profile shows high scores across all dimensions, indicating a well-rounded and rigorous presentation. The quantitative and qualitative information are balanced, with a strong technical level and high reliability.

Reliability 8/10