AI-Based Waste Battery and Plasma Convergence System for Adaptive Energy Reuse and Real-Time Process Optimization

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Comments:

• The manuscript is generally well-written with professional academic tone.
• The proposed AI-WBPCS system effectively combines waste battery reuse with plasma energy control, forming a novel hybrid electrochemical-plasma system. This convergence of AI, energy storage, and plasma processing is original and relevant to the circular economy and green technology transition.
• The study provides clear mathematical formulations for models D₁ (plasma prediction), D₂ (battery state analysis), and D₃ (battery-plasma matching), each backed by equations (1)-(5). The performance evaluation metrics (SOH accuracy, ΔP fluctuation, η efficiency) are well-defined and supported by data.
• Achieving 93.7% SOH prediction accuracy and ±2.1% plasma output fluctuation demonstrates strong technical feasibility and reproducibility. Experimental data from 2P6S EV modules lend credibility.
• The use of hybrid STM32-Jetson Nano architecture ensures real-time computation. The combination of CNN, GRU, and ResNet layers is appropriate for spatio-temporal and residual prediction.
• The inclusion of SHAP analysis to validate AI decision logic enhances trust and explainability, aligning with industrial AI safety practices.
• The manuscript lacks details on the dataset size, type of sensors, and data preprocessing (normalization, augmentation, or filtering).
• While laboratory-scale tests are clear, the scalability to industrial or field deployment such as grid-level or large plasma systems is not discussed.
• The study could be strengthened by including an energy flow diagram, efficiency breakdown, or cost–benefit assessment.
• The environmental impact reduction (CO₂ savings, waste mitigation) is mentioned conceptually but not quantified.
• The paper compares the AI system to a rule-based baseline only. Please add comparisons with other AI models (LSTM-only, XGBoost) which would improve scientific depth.
• Please provide more details on data acquisition, sampling intervals, number of cycles, and preprocessing to ensure reproducibility.
• Please add Environmental and Economic Impact Section:
• Quantify how much energy or emissions are saved through AI-WBPCS reuse such as CO₂ reduction, extended battery life.
• Please use consistent color themes (AI vs. Rule-based) and increase figure resolution. Please add descriptive captions explaining trends.
• It is recommended to demonstrate system performance in a practical setting such as wastewater plasma treatment or agricultural sterilization.
• It is recommended to add p-values or confidence intervals in Table 5-7 to confirm statistical significance.
• It is recommended to discuss potential integration with green hydrogen systems or microgrid frameworks to align with circular energy ecosystems.
• It is recommended to shorten overly long sentences, fix minor grammatical issues, and improve transitions between sections.
• Please focus less on repetition of numerical results and more on the broader implications (industrial scalability, future work, policy relevance).
• Some sentences are overly long. Please try to split into shorter sentences.
• References are up-to-date and well-balanced.
• Reference formatting should be consistent as per MDPI Applied Sciences guidelines.

Comments for author File: Comments.pdf

Author Response

Please find attached our detailed point-by-point response document

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study focuses on the environmental and industrial challenges posed by retired lithium-ion batteries from electric vehicles and energy storage systems. It innovatively proposes and experimentally validates an AI-based Waste Battery and Plasma Convergence System (AI-WBPCS), which integrates residual energy recovery from waste batteries with adaptive plasma control, providing a practical technical pathway for circular economy-oriented energy reuse. The research holds certain engineering significance, yet there are still some shortcomings that need improvement.

1. The font styles in the tables need to be standardized, the figure layouts require adjustment, and the line spacing in the main text should be unified. For instance, this applies to Table 7 and Figure 7.
2. The experimental validation in the paper utilized 2P6S retired battery modules, but critical background information such as the source, usage history, and initial state variations of these batteries was not provided. Please supplement this information.
3. While SHAP values measure the contribution of features to the model's output, this does not fully equate to causal impact on the actual physical system. How do the authors ensure that the SHAP values of the identified important features (e.g., SOH and DCIR) genuinely reflect their physical causal roles in the real electrochemical-plasma system, rather than merely representing statistical correlations in the data?
4. Have comparisons been conducted with recently published battery-energy management systems based on other AI methods, such as reinforcement learning or Transformer models?
5. Although the conclusion section mentions that the system can be applied in multiple scenarios, it fails to discuss the challenges in practical implementation.

Author Response

Please find attached our detailed point-by-point response document

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This manuscript presents a novel and interesting approach to a significant environmental and industrial challenge: the reuse of retired electric vehicle batteries. The authors propose an AI-based Waste Battery and Plasma Convergence System (AI-WBPCS) that integrates energy recovery from these batteries with adaptive plasma control for various applications. The concept of creating a self-optimizing, intelligent energy reuse framework is timely and holds considerable promise. A major revision is recommended to address critical gaps.

* The manuscript would be significantly strengthened by providing a more comprehensive and detailed explanation of the architectures for the three AI sub-models (D1, D2, and D3). For the D1 model, for example, it is suggested that the authors specify the key hyperparameters of the Multi-task Deep Neural Network (MT-DNN), including the number and configuration of CNN and GRU layers, the number of neurons in each layer, the specific activation functions used, and details of the optimization algorithm beyond its name (e.g., learning rate, momentum). This level of detail is crucial for the replicability of the research.

* The authors are encouraged to elaborate on the rationale for selecting a ResNet-18 architecture for the D2 battery state analysis model. A justification explaining why this particular deep residual network is more suitable for estimating State of Health (SOH) and Direct Current Internal Resistance (DCIR) compared to other potential architectures (e.g., LSTMs, other CNN variants) would provide valuable context and reinforce the methodological choices made in the study.

* The manuscript states that the weighting coefficients in the composite loss function (Equation 2) were "empirically tuned." To enhance the scientific rigor of this process, it would be beneficial to provide more details on the tuning methodology. For instance, did the authors employ a grid search, random search, or another optimization technique? Furthermore, including a sensitivity analysis that shows how the model's performance varies with changes in these coefficients would add robustness to the findings.

* While the use of SHAP (SHapley Additive exPlanations) for model interpretability is a valuable addition, the discussion could be expanded. It is suggested that the authors provide a brief overview of how the SHAP values were computed in the context of their models. More importantly, the manuscript could more explicitly connect the quantitative SHAP results (e.g., SOH having an importance of 46%) to the underlying electrochemical principles governing the system, thereby more concretely demonstrating the physical relevance and trustworthiness of the AI's decisions.

* The reported 13.1% improvement in energy matching efficiency is a key finding. To better highlight the impact of this result, the authors could include a brief analysis of the potential economic and environmental implications in a real-world scenario. Translating this percentage into tangible benefits, such as reduced operational costs, lower carbon footprint, or extended operational lifetime for a specific application, would make the contribution more impactful.

* The experimental validation was conducted using specific 2P6S retired EV battery modules. It would be valuable for the authors to discuss the potential scalability and generalizability of the proposed AI-WBPCS. A discussion on how the system might perform with different battery chemistries (e.g., LFP, NCA), varying levels of degradation, or different pack configurations from other manufacturers would provide a more complete picture of the technology's applicability.

* The paper reports that the AI controller maintains plasma output stability within ±2.1%. To help readers appreciate the significance of this achievement, the authors are encouraged to provide context by comparing this stability metric to established industry standards or operational requirements for the target applications mentioned, such as biomedical sterilization or smart agriculture.

* The feature importance analysis rightly identifies SOH and DCIR as dominant predictors. The manuscript could be improved by briefly discussing other potentially influential parameters that were considered or could be relevant (e.g., cell temperature uniformity, voltage distribution) and justifying why SOH and DCIR were ultimately prioritized as the key input features for the matching model.

* For clarity on the control system's design, it is suggested that the authors specify whether the feedback control loop for the plasma generation is driven exclusively by the AI model's predictions. If any conventional controllers (e.g., PID) were used in parallel as a fail-safe or for fine-tuning, this architecture should be described, as it has important implications for system robustness and safety.

* The manuscript proposes several compelling applications. To move from a general proposal to a more grounded discussion of deployment, it would be beneficial if the authors elaborated on the specific adaptations or model re-calibrations the AI-WBPCS would require to be effectively implemented in one or two of these diverse fields, highlighting the engineering steps needed to bridge the gap from lab to field.

* Given that the system is designed for real-time control, a discussion of the computational load is warranted. The authors could report the average inference time of the AI models on the STM32-Jetson Nano platform. This would allow an assessment of whether the computational performance is sufficient for the stated 2 Hz sampling rate while leaving adequate processing overhead for other system tasks.

* The comparison against a baseline is central to the paper's claims. The manuscript would be more compelling if the "conventional rule-based methods" used as a baseline were described in greater detail. A clear definition of the rules and logic employed in this baseline system is necessary for a transparent and comprehensive comparison.

* To ensure reproducibility, the methodology section should be expanded to include a clear description of the software environment. Please reference the programming language (e.g., Python version) and the specific names and versions of key libraries used for model development, training, and analysis (e.g., TensorFlow, PyTorch, Scikit-learn, SHAP).

* Section 3.3, "Plasma Output Control and Stability," would benefit from a clearer exposition of the model's operation. It is suggested that the authors explicitly restate the specific inputs provided to the D1 model and the outputs it generates in the context of the results being presented. Furthermore, please provide details on how this model was constructed and which software, packages, and programming languages were employed in its implementation.

Author Response

Please find attached our detailed point-by-point response document

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have  revised the manuscript in response to the reviewers' comments. Recommendation: Accept for publication.

Reviewer 3 Report

Comments and Suggestions for Authors

The reviewer thanks the authors' the effort taken on the review.
The manuscript is recommended for publication.