Review Reports

PCA-based regression to predict the water content (hydration level) of the dough (Section 2.9, equations 43 – 44). However, the water content is an independent, controlled experimental variable—it is the input to the process, not an output of the electromechanical system.

Thank you for this note. The part about hydration regression models is removed from the paper.

2. Moreover, the list of features used in PCA includes “hydration” itself (Section 2.8). Using the variable to be predicted as a predictor constitutes a textbook case of data leakage. The resulting R² values (0.64–0.96) are therefore entirely artificial and provide no evidence of a valid predictive model. The subsequent “optimization” to find an optimal hydration level using this model is circular reasoning and yields no scientific insight beyond re-confirming the experimental set points (52%, 58%, 63%). This single mistake invalidates the entire data-driven component of the study.

Thank you for this note. The part about hydration regression models is removed from the paper. The description is changed to “Within the studied levels (52–63%), the minimum load was observed at 58%”.

3. The claim of an “optimal hydration range of 57–58%” is not supported by any independent validation. The regression model simply interpolates between three discrete hydration levels (52%, 58%, 63%). Finding that 58% is better than 52% or 63% is trivial when 58% was already one of the tested levels and nothing proves that 57% has any significance beyond interpolation artifacts. Using fmincon on a quadratic surface to claim an optimum is a hollow exercise when the underlying model is both contamined by leakage and extrapolates without any physical justification.

Thank you for this note. The part about hydration regression models is removed from the paper.

4. Notably, some expressions should be added with corresponding references, as they are not the first time proposed, such as:

Thank you for this note. The expressions are added in the descriptions of the equations (28-31). All equations are checked for the description of the variables used.

5. Some already existing methods, which are widely utilized these days to solve the control problems, state filtered disturbance rejection control, multilayer neuroadaptive reinforcement learning of disturbed nonlinear systems via actor-critic mechanism and so on. Please compare and discuss in detail the methods proposed above in this manuscript, in order to further highlight the advantages and innovation of this work.

Thank you for this note. A new Table 10 with these comparisons is added in the discussion part.

6. Large portions of the paper (PID controller tuning, hardware description, GUI screens, market analysis) are irrelevant to the core contribution and read like a technical report rather than a focused scientific article. These sections inflate the paper with superfluous detail while the methodological core remains deeply flawed.

Thank you for this note. The technical details are moved to Appendix A.

7. Only three hydration levels were tested, and the exact dough mass was fixed at 300g flour. The limited experimental space makes the regression model essentially a curve fit rather than a robust predictive tool. Additionally, the viscoelastic load simulation uses a farinograph-like torque profile that is manually tuned, with no sensitivity analysis or uncertainty quantification.

Thank you for the comment. We acknowledge that the experimental space is limited to three hydration levels and a fixed dough mass, which constrains the regression model and makes it more representative of curve fitting within the tested range. This limitation is now explicitly stated in the manuscript. We also clarify that the viscoelastic load profile was manually tuned and that no formal sensitivity or uncertainty analysis was performed. These aspects are now identified as important directions for future work.

Technical Issues

1. The regression model for whole wheat flour achieves only R^2=0.64, meaning 36% of variance is unexplained. The authors still categorize this as high accuracy.

Thank you for this note. The part about hydration regression models is removed from the paper.

2. The paper identifies 57–58% as the optimal hydration range, but the actual experimental conditions tested are 52%, 58%, and 63% only. No measurements were taken at 57%. Claiming an optimum between two tested points without finer resolution or analytical proof is scientifically unjustifiable.

Thank you for this note. The part about hydration regression models is removed from the paper.

3. The authors assert the Maxwell model outperforms Kelvin-Voigt for dough, but present no comparative fitting error, no residual analysis, and no rheological parameter identification from actual dough measurements.

Thank you for this note. A new description of Maxwell model is added. Dough exhibits both viscous and elastic behavior, which makes it suitable for representation through classical viscoelastic models. Among these, the Maxwell model is widely used to describe stress relaxation, a phenomenon strongly expressed in bread dough under mechanical deformation. In the Maxwell formulation, stress decays over time, reflecting the experimentally observed tendency of dough to continue deforming and flowing under sustained loading from the mixing element [35]. Maxwell model was selected as a qualitative approximation [36].

4. Table 13 reports MAE=2.17 Nm for torque, while the experimentally measured torque range is 3.33–4.37 Nm. The relative error therefore reaches roughly 50–65% of actual values.

Thank you for this note. The inappropriate regression models are removed from the paper. Also, the error values are checked and recalculated.

5. Nine out of twenty features were selected using a weight threshold of 0.6 with no sensitivity analysis.

Thank you for this note. Cross-validation is added. Three methods are described in Material and methods section and the results are presented in Table 7. (The methods are: K Fold (10), Hold Out (70/30), and Leave One Out)

6. The manuscript lacks citations to recent relevant works. CausalSR: Structural causal model-driven super-resolution with counterfactual inference

Thank you for this note. A new table 10 with these comparisons is added in the discussion part.

Writing Issues

7. The abstract states regression models achieve high accuracy (R^2=0.64–0.96) but R^2=0.64 is a mediocre fit by any standard in engineering modeling.

Thank you for this note. The part about hydration regression models is removed from the paper.

8. The title refers to Electromechanical Systems broadly, implying general applicability, while the entire study is conducted on a single consumer-grade bread machine model.

Thank you for this note. The title is corrected according to the reviewer note.

9.  Some references are commercial market reports, not peer-reviewed work.

Thank you for this note. All inappropriate references are removed. Only scientific works are cited.

10. Multiple figures (Figures 16–21) present experimental data curves without stating sample size per condition, measurement duration, or whether curves represent single runs or averages.

Thank you for this note. After correction of the paper the figures now are 12-17. The presented characteristics are averaged. It is noted in the figure title and descriptions.

The manuscript “Integrated Modeling and Data-Driven Analysis of Electromechanical Systems with Hydration-Dependent Viscoelastic Load” is devoted to the integrated modeling of the electromechanical system of an automatic bread maker with a viscoelastic load dependent on the hydration of the dough. The authors combine MATLAB/Simulink, Simscape, experimental measurements, a sensor system based on inexpensive components, principal component analysis, and regression modeling. The manuscript states that increasing hydration from 52% to 63% reduces the torque by 22%, 46%, and 34% for different flour types, and the regression models achieve R² = 0.64–0.96. The topic generally corresponds to the profile of Applied Sciences, but the manuscript requires significant revision due to methodological inconsistencies, overstated claims of novelty, problems with model validation, and numerous shortcomings in the presentation of the material.

We thank the reviewer for the assessment, and all noted issues regarding methodology, novelty, validation, and presentation have now been fully addressed in the revised manuscript.

Critical Comments

1. The Abstract needs correction, as it presents the model as generally reliable, although the given range of R² = 0.64–0.96 indicates uneven prediction quality for different flour types.

We thank the reviewer for the remark, and the Abstract has been revised to accurately reflect the uneven prediction quality indicated by the R² range. The regression models achieve moderate predictability (R²=0.64–0.96), model performance varies across flour types.

2. The title, keywords, and Abstract need stylistic refinement. The keyword “Machine Learning in Bread making” is formulated in an unacademic manner and does not correspond to typical indexing vocabulary. It should be replaced with “data-driven modeling”, “bread dough rheology” or “electromechanical monitoring”. The title is also too broad, as it actually studies one type of household bread maker, rather than a broad class of electromechanical systems.

Thank you for this note. The title, abstract and keywords are corrected according to the reviewer notes.

3. The Introduction contains a statement about the lack of previous integrated studies, but does not prove this through systematic comparison. The authors claim that no previous work offers a similar model, but no comparison table is provided with works on sensor monitoring, torque, current diagnostics, and viscoelastic load modeling. This weakens the justification for novelty.

We thank the reviewer for this observation, and the revised manuscript now includes a comparison Table 10 and an explicit discussion demonstrating how the proposed approach differs from existing work on sensor monitoring, torque/current diagnostics, and viscoelastic load modeling.

4. The introduction is significantly repetitive. The research gap is actually formulated several times on pages 2–4: first by dividing the previous works into three groups, then by reiterating the separate analysis of the engine, transmission, and dough, and then by almost identically formulating the contribution. The authors should condense this section, remove duplication, and conclude it with a clear plan of the paper structure.

Thank you for this note. The introduction has been condensed, repetitions removed, and the section now ends with a clear and streamlined paper outline.

5. The description of the experimental data is insufficient to reproduce the study. The authors mention more than 6,000 measurements for each combination, but do not indicate the sampling frequency, the duration of one cycle, the number of independent batches, the averaging procedure, signal filtering, and sensor calibration. Without these details, it is impossible to assess whether the 6000 points are independent observations or an autocorrelated time series of a single process.

Thank you for this note. The revised manuscript now provides full experimental detail (subsection 2.1), including sampling frequency, cycle duration, number of independent batches, filtering, averaging, and sensor calibration, ensuring that the data structure and independence of observations can be properly assessed.

6. The Materials and Methods section is overloaded with equations, some of which are not used directly in subsequent results. Pages 8–11 contain many integral and discrete dependencies, but it is not always clear which of them are used in the code and which are given only theoretically. Authors should add a generalized block diagram of the computational pipeline and link each formula to a specific stage of data processing.

Thank you for this note. The Materials and Methods section has been streamlined, a unified computational block diagram has been added, and each equation is now explicitly linked to its corresponding stage in the processing pipeline (Figure 4).

7. The formulation of the PCA regression problem is contradictory. The Abstract and Contributions state that the model predicts the load and parameters of the electromechanical system, but in section 2.9, the dependent variable is the amount of water in the dough. These are different statements of the problem: predicting hydration from electromechanical features and predicting load from hydration. Authors should clearly define the target variable and the modeling logic.

Thank you for this note. The revised manuscript now clearly defines the target variable, resolves the inconsistency between hydration‑prediction and load‑prediction formulations, and provides a unified modeling logic.

8. The energy analysis contains inconsistencies between the text and the table. On page 30, the authors write about a stable efficiency of 0.675 and a power factor of 0.93–0.94, while Table 10 shows cos(φ) of about 0.85. The conclusions also claim a reduction in energy consumption of up to 6%, but this is not shown as a separate statistically verified result. The figures should be recalculated and the text should be reconciled with the table.

Thank you for this comment. The energy values have been recalculated, the efficiency and power‑factor inconsistencies have been corrected, and the text has been fully aligned with the updated table, with the energy‑reduction claim revised accordingly.

9. The validation of the simulation model requires a full check. Table 13 shows RMSE = 2.42 and MAE = 7.65 for angular speed, which is mathematically impossible for the same data, since RMSE is usually not less than MAE. For torque, RMSE = 2.20 is also very large relative to the experimental range of 3.33–4.37 Nm. The authors should check the error formulas, units, and scaling.

Thank you for this note. The error metrics have been recalculated, the RMSE–MAE inconsistency has been corrected, and the formulas, units, and scaling have been verified to ensure mathematically valid model‑validation results.

10. The quality of the figures is insufficient for journal publication. On pages 16, 18, 26–28, the graphs and Simulink diagrams have small captions, unclear legends, and poor readability of the axes. Figures 16–21 are difficult to interpret without magnification. All graphs should be converted to vector format, fonts, axis captions, units, and legends should be standardized.

Thank you for this note. The figures have been regenerated and the font sizes have been increased to ensure clear readability throughout the manuscript.

In summary, although the manuscript fits the scope of Applied Sciences, it requires major revision before it can be considered for publication.

Thank you for this summary. The manuscript has undergone major revision, and all issues highlighted by the reviewer have now been fully addressed in the revised version.