Hybrid NARX Neural Network with Model-Based Feedback for Predictive Torsional Torque Estimation in Electric Drive with Elastic Connection
Round 1
Reviewer 1 Report
Comments and Suggestions for Authors1.The article should give all the parameters of the motor, such as resistance, inductance, moment of inertia, etc.
2.In the paper, the sampling frequency is 2kHz, and the PWM switching frequency is 7kHz, the sampling frequency is much lower than the PWM switching frequency, which will inevitably lead to distortion of the sampled signal, thus reducing the dynamic response of the system. Can the authors give the current waveform and the distortion rate?
3.The analysis of various observer methods in the introduction is too simple, for example, the chattering problem in the sliding mode observer can be suppressed by using higher order sliding mode, etc. In order to highlight the contribution of this paper, the authors should cite more relevant literature and analyse it.
4. The quality of the images in the text is too blurry and the resolution should be further improved.
5. Parameter uptake experiments should be included to demonstrate the robust performance of the proposed strategy.
6. A stability proof of IDOB should be given to enhance theoretical stringency.
Author Response
Please find the answers in the attached file.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for Authors- In the intro part, more references should be introduced on the various predictors, estimators, and filtering techniques to address the measurement or estimation in electric drivetrain systems.
-
Please revise the manuscript in a logical flow with shorter but clear sentences; some sentences are too long and complex, and some portions (e.g., the section discussing model-based feedback using IDOB) are dense and difficult to follow.
- Can the authors add more information about the experimental setup and system parameters for the dual-mass drive system?
Author Response
Please find the answers in the attached file.
Author Response File: Author Response.docx
Reviewer 3 Report
Comments and Suggestions for Authors1. The introduction and conclusion sections should more explicitly highlight the novel contribution of this work in comparison to existing methods. The distinction between your proposed hybrid NARX NN and existing approaches like Kalman filters or other neural network-based estimators should be clearer.
2. The literature review section is comprehensive, but it would benefit from a more critical analysis of the limitations of existing methods. For example, while you mention various observers and estimators, a more detailed comparison of their strengths and weaknesses in the context of your study would help readers understand why your approach is superior.
3. While the hybrid model (NARX NN + model-based feedback) is introduced, a more detailed explanation of how the model-based feedback improves the prediction performance and reduces noise sensitivity would be useful. The exact mechanism of integration between the model and feedback needs more elaboration.
4. The paper mentions several simplifications, such as neglecting nonlinear effects like backlash and hysteresis. It would be beneficial to discuss the potential impact of these assumptions on the accuracy and generalizability of your approach, particularly for real-world applications.
5. The experimental verification section could benefit from more details on the hardware setup, particularly in terms of how the real-time implementation was carried out. Were there any challenges in the deployment that could be addressed in future versions of the system?
6. While the results are promising, more statistical analysis (confidence intervals) would strengthen the results. Providing error bars or uncertainty measures for the performance metrics in the results would give readers a clearer idea of the reliability of your approach.
7. The noise handling capability is discussed, but further elaboration on how different types of noise (Gaussian, impulse noise) affect the performance of the estimator would be helpful. Additionally, a comparative study with more advanced noise reduction techniques could strengthen this section.
8. You discuss the robustness of the proposed method, but more rigorous testing under diverse operating conditions (different speeds, loads, and environmental factors) would better demonstrate the generalizability of the model. More validation using real-world data would also be valuable.
9. The comparison between simulation and experimental results is promising, but it would be useful to explain any discrepancies observed between the two. Are there particular factors in the experimental setup that might have led to differences in the results?
Author Response
Please find the answers in the attached file.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe author addressed my worries very well and I recommend accepting it!
Reviewer 3 Report
Comments and Suggestions for AuthorsNo more comments.
