A Strain-Data-Driven Factor-Wise Inverse Identification Approach for Blown-Sand Erosion Parameters of GFRP Strips

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Manuscript ID: materials-4471723

Type: Article

Title: A Strain-Data-Driven Inverse Identification Approach for Blown-Sand Erosion Parameters of GFRP Strips Within the Tested Range

First of all, I would like to emphasize that the authors have presented an interesting and timely study on the degradation of GFRP composites subjected to sand erosion. This topic is of particular importance in regions exposed to sandstorms and is not limited exclusively to desert environments. It is also noteworthy that the sand used in the experimental programme was collected from the Kubuqi Desert in Inner Mongolia. This demonstrates the authors’ careful attention to detail and their effort to ensure that both the testing material and the simulated blown-sand erosion conditions accurately reflected natural environmental conditions.

I highly appreciate the comprehensive research approach adopted in this study, which combines advanced experimental techniques, including Digital Image Correlation (DIC) and Scanning Electron Microscopy (SEM), with artificial intelligence tools such as neural networks and the Particle Swarm Optimization (PSO) algorithm. The analyses provided valuable insights into the mechanisms of material degradation and the influence of key erosion parameters on the mechanical performance of the investigated composite. It should be noted, however, that the experimental programme was conducted on a specific GFRP composite system consisting of glass fibres and an epoxy resin matrix with well-defined material characteristics.

This raises an important question: would GFRP composites manufactured from different fibre and resin systems, and therefore characterized by different material properties, exhibit a comparable resistance to degradation under sand erosion conditions?

The authors’ effort to develop an inverse identification model deserves particular recognition, as such an approach has significant potential for practical applications in the diagnostics and durability assessment of composite structures operating in erosive environments. Nevertheless, expanding the experimental database and conducting additional validation over a broader range of erosion parameters would likely improve the model's generalization capability and further enhance its practical applicability.

After addressing the minor editorial comments and standardizing the notation used throughout the manuscript, the paper will constitute a valuable contribution to the field of research on the durability of GFRP composites exposed to sand erosion.

Notes for the Authors:
verse 199: 2.7 g/cm3/ should be corrected to 2.7 g/cm³.
verse 267: 28.0%/ should be corrected to 28%.
verse 308: The quality of the graphical objects should be improved.
verse 591: The readability of the graphical objects should be enhanced.
Sections 2.3, 3.4, 4.2, and 4.3: The notation used for the load parameters (Pu, Pmax/P_u, P_max) should be standardized throughout the manuscript. The notation P_u and P_max is recommended for consistency and clarity.

Author Response

1. Summary

    We sincerely thank the reviewer for the careful reading of our manuscript and for the constructive editorial suggestions. We have checked the manuscript carefully and made the corresponding corrections. The point-by-point responses are provided below.

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: [verse 199: 2.7 g/cm3/ should be corrected to 2.7 g/cm³.]

Response 1: [Thank you for pointing out this typographical issue. The unit of particle density has been corrected from “2.7 g/cm3” to “2.7 g/cm³” (Line 202) in Section 2.2 of the revised manuscript.] 

Comments 2: [verse 267: 28.0%/ should be corrected to 28%. ]

Response 2: [Thank you for identifying this inconsistency. The value “28.0%” has been corrected to “28%”(Line 277) throughout the revised manuscript.] 

Comments 3: [verse 308: The quality of the graphical objects should be improved.]

Response 3: [Thank you for this suggestion. The graphical presentation has been revised to improve quality and consistency. In particular, Figure 5 has been reformatted using consistent axis ranges, line styles, symbols, labels, and subplot dimensions, and Figure 14 has been redrawn to present the inverse-identification workflow more clearly.] 

Comments 4: [verse 591: The readability of the graphical objects should be enhanced.]

Response 4: [Thank you for this suggestion. The graphical presentation has been revised to improve readability. In particular, the layout and arrangement of the subfigures in Figure 15 have been reorganized, and the figure caption, labels, and unit formatting have been revised for greater clarity and consistency.] 

Comments 5: [Sections 2.3, 3.4, 4.2, and 4.3: The notation used for the load parameters (Pu, Pmax/Pu, Pmax) should be standardized throughout the manuscript. The notation Pu and Pmax is recommended for consistency and clarity.]

Response 5: [Thank you for this suggestion. The load notation has been standardized throughout the revised manuscript. The symbol Pᵤ is now used consistently to denote the specimen-specific ultimate tensile load, and the corresponding normalized load levels are expressed consistently as fractions of Pᵤ.] 

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript combines tensile testing, SEM, DIC and a PSO-based inverse procedure to study blown-sand erosion of GFRP strips. The experimental part is useful and the leave-one-condition-out validation is a welcome addition. However, I have several concerns regarding the scope and interpretation of the inverse model.

1. The experiments follow essentially a one-factor-at-a-time approach. This is adequate for studying individual effects, but I am not convinced that it supports a four-parameter coupled mapping. The authors should either justify this point more clearly.
2. The difference between the in-sample error (about 4.2%) and the held-out error (about 31.8%) is substantial. Since the held-out conditions are still within the experimental range, the statement that the model is “reliable within the tested parameter range” is not fully supported by the cross-validation results. Calibration, interpolation and extrapolation should be clearly distinguished.
3. The Introduction refers to a neural-network model, whereas Section 4 describes linear interpolation. This inconsistency should be corrected. The reference to a “two-variable inverse problem” should also be checked.
4. The authors note that different erosion conditions may produce similar strain responses. This is central to the inverse problem and deserves more discussion. Some assessment of parameter sensitivity or non-uniqueness would strengthen the work. Experimental variability should also be shown; repeating PSO runs does not account for measurement uncertainty.
5. More details on the DIC procedure and strain uncertainty are needed. I would also present the proposed in-service use based on 0.2Pu–0.6Pu as a possible future application rather than as a demonstrated practical procedure.
6. The manuscript also requires careful editorial revision. For example, Figure 3 is captioned “Strain gauge arrangement”, although it shows the blown-sand erosion test setup; Section 2.3 appears after Sections 3.1 and 3.2; Table 2 lists an erosion angle of “91°” instead of 90°; and Table 7 contains a duplicated “Erosion factor” heading. Typographical and wording errors such as “50mine” (in Fig.15) and “elasticity modulus of elasticity” should also be corrected.

 

Comments on the Quality of English Language

The English could be improved to more clearly express the research.

Author Response

1. Summary

   We sincerely thank the reviewer for the careful and constructive assessment. The comments were particularly helpful in clarifying the scope, validation, uncertainty, and practical interpretation of the inverse-identification procedure. We have revised the manuscript accordingly, as detailed below.

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: [The experiments follow essentially a one-factor-at-a-time approach. This is adequate for studying individual effects, but I am not convinced that it supports a four-parameter coupled mapping. The authors should either justify this point more clearly.]

Response 1: [Thank you for this important comment. We agree that the experimental program follows a one-factor-at-a-time (OFAT) design and therefore does not support a simultaneous four-parameter coupled mapping. The manuscript has been revised accordingly. The proposed method is now consistently described as a factor-wise inverse-identification approach, in which one erosion factor is varied and identified at a time while the remaining factors are fixed at their prescribed values. The title, Abstract, Introduction, Section 4.1, Figure 14, Equations (1)–(2), and the related discussion in Sections 4.2–4.4 have been revised to reflect this scope. The revised manuscript therefore no longer presents the method as a coupled four-parameter prediction or identification model.] 

Comments 2: [The difference between the in-sample error (about 4.2%) and the held-out error (about 31.8%) is substantial. Since the held-out conditions are still within the experimental range, the statement that the model is “reliable within the tested parameter range” is not fully supported by the cross-validation results. Calibration, interpolation and extrapolation should be clearly distinguished. ]

Response 2: [Thank you for this important comment. We agree that the low in-sample error should not be interpreted as predictive accuracy for unseen conditions. The revised manuscript now clearly distinguishes in-sample reconstruction from generalization to held-out conditions. The approximately 4.2% error is described as an in-sample reconstruction error, whereas leave-one-condition-out validation gives an overall mean error of approximately 31.8%. Section 4.4 also clarifies that a held-out parameter level is an unseen condition for the corresponding factor-wise interpolation model, even when it lies within the nominal experimental bounds. In addition, conditions involving two or more simultaneously off-baseline factors have not been experimentally validated and are not treated as interpolation points of the present model. The corresponding claims in the Conclusions have therefore been moderated.] 

Comments 3: [The Introduction refers to a neural-network model, whereas Section 4 describes linear interpolation. This inconsistency should be corrected. The reference to a “two-variable inverse problem” should also be checked.]

Response 3: [Thank you for pointing out this inconsistency. The description of the proposed method in the Introduction has been revised to match the methodology used in Section 4. The present model is now consistently described as a factor-wise interpolation-based inverse-identification approach combined with PSO, rather than as a neural-network model. The abbreviation ANN is retained only when referring to artificial neural networks reported in the cited literature and is defined at its first occurrence. The previous reference to a “two-variable inverse problem” has also been removed, and the inverse problem is now described as a factor-wise one-parameter identification problem.] 

Comments 4: [The authors note that different erosion conditions may produce similar strain responses. This is central to the inverse problem and deserves more discussion. Some assessment of parameter sensitivity or non-uniqueness would strengthen the work. Experimental variability should also be shown; repeating PSO runs does not account for measurement uncertainty.]

Response 4: [Thank you for this important comment. We have expanded the discussion of identifiability and non-uniqueness in Section 4.4. The factor-specific leave-one-condition-out errors are now used as an indication of relative parameter identifiability: the larger errors for erosion angle and sand flow rate are discussed in relation to non-monotonic or weak strain–parameter relationships, for which different parameter values may produce similar strain responses and reduce the uniqueness of the inverse solution. We also clarify in Section 4.1 that the 20 repeated PSO runs quantify only the stochastic variability of the optimization algorithm and do not represent specimen-to-specimen experimental variability or measurement uncertainty. Three parallel specimens were tested at each parameter level, and the reported mechanical-property values are their mean values. Because specimen-to-specimen variability was not quantitatively evaluated, unsupported error bars or uncertainty estimates were not introduced; this limitation is explicitly acknowledged in Section 4.4 and identified as an issue for future uncertainty analysis.] 

Comments 5: [More details on the DIC procedure and strain uncertainty are needed. I would also present the proposed in-service use based on 0.2Pu–0.6Pu as a possible future application rather than as a demonstrated practical procedure.]

Response 5: [Thank you for this important comment. The DIC procedure has been expanded in Section 2.3. The revised manuscript now specifies the VIC-3D SR system, image resolution, speckle preparation, image-acquisition frequency, working distance, analysis software, axial-strain definition, and the nominal strain resolution of the system. We also explicitly state that no separate experimental calibration of DIC strain uncertainty was performed, so the nominal resolution is not presented as a measured experimental uncertainty. In addition, the use of strain data at 0.2Pᵤ–0.6Pᵤ is now presented as a laboratory proof of concept rather than as a demonstrated in-service procedure. Section 4.3 discusses possible future formulations based on a known non-destructive reference load P_ref or on absolute load–strain pairs.] 

Comments 6: [The manuscript also requires careful editorial revision. For example, Figure 3 is captioned “Strain gauge arrangement”, although it shows the blown-sand erosion test setup; Section 2.3 appears after Sections 3.1 and 3.2; Table 2 lists an erosion angle of “91°” instead of 90°; and Table 7 contains a duplicated “Erosion factor” heading. Typographical and wording errors such as “50mine” (in Fig.15) and “elasticity modulus of elasticity” should also be corrected.]

Response 6: [Thank you for carefully identifying these editorial and formatting issues. The corresponding corrections have been made throughout the revised manuscript. Specifically, the caption of Figure 3 has been corrected, Section 2.3 has been placed appropriately in the Methods section, the erosion angle in Table 2 has been corrected to 90°, the duplicated heading in Table 7 has been removed, and typographical and wording errors such as “50mine” and the incorrect elastic-modulus expression have been corrected. Table 2 has also been clarified by explicitly labeling the total number of specimens in each OFAT series.] 

Reviewer 3 Report

Comments and Suggestions for Authors

The work is dealing with the Blown-Sand Erosion of GFRP Strips and proposes a methodology for Strain-Data-Driven Inverse Identification Approach of the problem.

The work is interesting, however, it suffers from some major scientific problems that is necessary to be fixed prior of its publication. In the following the identified problems are listed:

1) There is a sequential approach to the problem. First erode then test, which is not representative of the actual Blown-Sand Erosion. 

The applied test protocol is strictly sequential: specimens are eroded unloaded and only afterwards loaded quasi-statically to failure (Section 2.2–2.3, Fig. 4). No mechanical load is applied during the erosion phase. This does not represent the real service condition the paper targets. The stated application, GFRP strips as tensile reinforcement/fixed formwork on in-service bridges (Section 1), and back-analysis of the "in-service erosion environment of GFRP strengthening members" (Section 4.3), involves strips that are already under sustained operational stress (dead load, prestress, traffic load, possibly cyclic) for months to years while erosion accumulates, i.e. erosion and loading act simultaneously and continuously, not sequentially and briefly as in the test programme.

For a notch-/flaw-sensitive brittle composite this is a substantive difference, not a cosmetic one:

* Sand impact on an already-strained surface may open or propagate micro-defects (fiber–matrix debonding, resin micro-cracking) differently than impact on an unstressed surface, so eroded pits could act as stress-raisers that grow during exposure, not just fixed damage assessed afterward.

* Sustained/cyclic load during erosion opens the possibility of load–erosion synergy (e.g. stress-assisted or fatigue-assisted crack growth at eroded sites) that a single post-erosion quasi-static pull-to-failure test cannot capture at all.

* The erosion-center strain localization reported in Section 3.4/Fig. 10–13 is measured on a specimen with zero prior stress history; whether the same localization forms in a pre-stressed member is untested.

The reported degradation percentages (16.8% at 90°, 28.0% at 31 m/s, 35% at 50 min, etc.) and the inverse-identification tool built on them should therefore be presented, at most, as an upper-bound / best-case estimate for erosion-only exposure, not as a general "reference for material selection, protective design and durability assessment" (Abstract, Conclusion) applicable to members loaded throughout their service life.

2) The authors claim multi factor coupling, but this never applied during testing. 

Title/Abstract/Introduction repeatedly claim study of erosion under "multi-factor coupling," but Table 2 is a one-factor-at-a-time design around a single baseline (90°, 26 m/s, 55 g/min, 30 min. No specimen combines two or more off-baseline factors, and no interaction/response-surface analysis is reported. This language should be removed or substantiated. This OFAT structure is also the likely root cause of the large leave-one-condition-out errors in Table 7 (up to 43.8%): a field condition combining several off-baseline factors is effectively an extrapolation for a model trained only along single-factor axes through one center, not a true interpolation. This is worth stating explicitly in Section 4.4.

3) There is a circularity argument in the proposed work flow. 

The inverse model needs strain at fixed fractions of Pu (0.2–0.6 Pu), where Pu is the specimen's own ultimate load, precisely the unknown quantity the tool is meant to estimate in service. Please clarify what Pu value (nominal/design, historical, iteratively estimated) would be used operationally without a destructive test.

There are also several minor inconsistencies that the authors will correct during revision (number of test samples, error bars that are missing, etc)

Based on the above my proposal is MAJOR REVISION prior of the publication of the work.

Comments on the Quality of English Language

A professional English review is needed

Author Response

1. Summary

  We sincerely thank the reviewer for the detailed scientific assessment. The comments helped us substantially revise the scope and interpretation of the study, particularly regarding the sequential erosion–tension protocol, the OFAT experimental design, the use of Pᵤ, and the limitations of the inverse-identification framework. Our point-by-point responses are provided below.

2. Point-by-point response to Comments and Suggestions for Authors

Comments 1: [(a) There is a sequential approach to the problem. First erode then test, which is not representative of the actual Blown-Sand Erosion. The applied test protocol is strictly sequential: specimens are eroded unloaded and only afterwards loaded quasi-statically to failure (Section 2.2–2.3, Fig. 4). No mechanical load is applied during the erosion phase. This does not represent the real service condition the paper targets. The stated application, GFRP strips as tensile reinforcement/fixed formwork on in-service bridges (Section 1), and back-analysis of the "in-service erosion environment of GFRP strengthening members" (Section 4.3), involves strips that are already under sustained operational stress (dead load, prestress, traffic load, possibly cyclic) for months to years while erosion accumulates, i.e. erosion and loading act simultaneously and continuously, not sequentially and briefly as in the test programme.

For a notch-/flaw-sensitive brittle composite this is a substantive difference, not a cosmetic one.

(b) Sand impact on an already-strained surface may open or propagate micro-defects (fiber–matrix debonding, resin micro-cracking) differently than impact on an unstressed surface, so eroded pits could act as stress-raisers that grow during exposure.

(c) Sustained/cyclic load during erosion opens the possibility of load–erosion synergy (e.g. stress-assisted or fatigue-assisted crack growth at eroded sites) that a single post-erosion quasi-static pull-to-failure test cannot capture.

(d) The erosion-center strain localization reported in Section 3.4/Fig. 10–13 is measured on a specimen with zero prior stress history; whether the same localization forms in a pre-stressed member is untested.

(e) The reported degradation percentages and the inverse-identification tool built on them should therefore not be presented as generally applicable to members loaded throughout their service life; the reviewer suggests presenting them, at most, as an upper-bound / best-case estimate for erosion-only exposure.]

Response 1: [(a) We agree that the present protocol is sequential and does not reproduce simultaneous mechanical loading and erosion. Section 2.3 now explicitly states that no mechanical load was applied during the erosion stage and that tensile testing was conducted only after the prescribed erosion exposure. The revised text clarifies that the tests characterize the post-erosion residual behavior of initially unloaded GFRP strips.

(b) We agree that erosion acting on an already stressed GFRP member may lead to damage evolution different from that observed in an initially unloaded specimen. The present experiments do not quantify stress-assisted resin cracking, fiber–matrix debonding, or growth of erosion-induced defects under sustained mechanical stress; therefore, no such mechanism is claimed in the revised manuscript.

(c) We agree that possible synergistic effects between sustained/cyclic loading and blown-sand erosion cannot be captured by the present sequential protocol. Section 2.3 now explicitly states that the tests do not reproduce simultaneous mechanical loading and erosion or the possible load–erosion interaction occurring in service. Such coupled loading–erosion behavior remains outside the scope of the present study and should be investigated in future work.

(d) The strain localization in Figures 10–13 is interpreted only as the post-erosion tensile response of specimens that had no mechanical load during the erosion stage. Whether the same localization pattern develops in a pre-stressed member under simultaneous erosion and loading remains unverified in the present study.

(e) We agree that the reported degradation percentages and inverse-identification results should not be generalized directly to members continuously loaded during service. The Abstract, Section 2.3, Sections 4.3–4.4, and the Conclusions have therefore been revised to restrict the findings to the investigated erosion-only laboratory conditions and to present the inverse-identification method as a laboratory proof of concept. We have not labelled the present results as an “upper-bound” or “best-case” estimate because the sequential tests do not establish a quantitative ordering relative to simultaneous load–erosion conditions; instead, their applicability is explicitly limited and general field-performance claims are avoided.] 

Comments 2: [The authors claim multi factor coupling, but this never applied during testing.Title/Abstract/Introduction repeatedly claim study of erosion under "multi-factor coupling," but Table 2 is a one-factor-at-a-time design around a single baseline (90°, 26 m/s, 55 g/min, 30 min). No specimen combines two or more off-baseline factors, and no interaction/response-surface analysis is reported. This language should be removed or substantiated. This OFAT structure is also the likely root cause of the large leave-one-condition-out errors in Table 7 (up to 43.8%): a field condition combining several off-baseline factors is effectively an extrapolation for a model trained only along single-factor axes through one center, not a true interpolation. This is worth stating explicitly in Section 4.4. ]

Response 2: [Thank you for this important observation. We agree that the experiment is an OFAT design and does not provide evidence for multi-factor interaction or simultaneous multi-parameter identification. The manuscript has therefore been systematically revised to describe the method as a factor-wise inverse-identification approach. The title, Abstract, Introduction, Section 4.1, Figure 14, Equations (1)–(2), Sections 4.2–4.4, and the Conclusions now consistently state that one erosion factor is varied or identified at a time while the remaining factors are fixed. Section 4.4 also explicitly states that conditions involving two or more simultaneously off-baseline factors have not been experimentally validated and cannot be regarded as interpolation points of the present model.] 

Comments 3: [There is a circularity argument in the proposed work flow.The inverse model needs strain at fixed fractions of Pu (0.2–0.6 Pu), where Pu is the specimen's own ultimate load, precisely the unknown quantity the tool is meant to estimate in service. Please clarify what Pu value (nominal/design, historical, iteratively estimated) would be used operationally without a destructive test.]

Response 3: [Thank you for identifying this important limitation. We agree that using the specimen-specific Pᵤ introduces a circularity for direct in-service application because Pᵤ is known only after destructive tensile testing. Section 4.3 now explicitly states that the present Pᵤ-normalized procedure is a laboratory proof of concept and cannot be directly applied to an in-service GFRP member whose ultimate load is unknown. For future engineering applications, the method could be reformulated using a known non-destructive reference load P_ref, such as a prescribed proof or service load, or by directly using absolute load–strain pairs. Such a field-oriented formulation would require an expanded experimental database and independent validation.] 

Comments 4: [There are also several minor inconsistencies that the authors will correct during revision (number of test samples, error bars that are missing, etc).]

Response 4: [Thank you for pointing out these inconsistencies. The experimental grouping has been clarified: Table 2 now reports the total number of specimens in each OFAT series, and the Methods section states that three parallel specimens were tested at each parameter level. The reported strength and modulus values are the means of the three parallel specimens. Because specimen-to-specimen variability was not quantitatively evaluated in the present analysis, we have not introduced unsupported error bars or standard-deviation estimates. This limitation is acknowledged in Section 4.4. The manuscript also explicitly distinguishes this experimental variability from the stochastic variability associated with repeated PSO runs.] 

Reviewer 4 Report

Comments and Suggestions for Authors

Please consider the attacched file

Comments for author File: Comments.pdf

Author Response

1. Summary

  We sincerely thank the reviewer for the detailed comments on the Abstract, Introduction, Methods, and Results. We have carefully checked the corresponding sections and revised the manuscript where appropriate. For the two graphical-scale suggestions that were not adopted exactly as proposed, our reasons are explained transparently below.

2.Point-by-point response to Comments and Suggestions for Authors

Comments 1: [In the following sentence “Tensile strength decreases by approximately 16.8% at a 90° erosion angle, and 28.0% at an erosion velocity of 31 m/s.” the authors are considering both the effect of the angle and the effect of the velocity. So when they indicate 16.8% at a 90° erosion angle they should also mention the velocity, and in the same way, when they indicate 28.0% at an erosion velocity of 31 m/s they should mention the erosion angle.]

Response 1: [Thank you for this comment. The Abstract has been revised to specify the fixed erosion conditions associated with the reported tensile-strength reductions. For the 90° erosion-angle result, the erosion velocity, sand flow rate, and erosion time are now stated as 26 m/s, 55 g/min, and 30 min, respectively. For the 31 m/s erosion-velocity result, the erosion angle, sand flow rate, and erosion time are now stated as 90°, 55 g/min, and 30 min, respectively.] 

Comments 2: [Specify the acronym PSO. ]

Response 2: [Thank you for the suggestion. The acronym PSO is now defined at its first occurrence in the Abstract as “particle swarm optimization (PSO)”.] 

Comments 3: [“other fantastic mechanical properties” is not a scientific formal language! Please rephrase it in a more professional sentence.]

Response 3: [Thank you for pointing this out. The informal expression has been removed. The revised Introduction now describes GFRP strips in formal terms, referring to their low density, high strength, corrosion resistance, and other favorable properties.]

Comments 4: [Specify the acronym ANN.]

Response 4: [Thank you for the suggestion. The acronym ANN is now defined at its first occurrence in the Introduction as “artificial neural network (ANN)”.] 

Comments 5: [The caption of Figure 3 is wrong.]

Response 5: [Thank you for identifying this error. The caption of Figure 3 has been corrected to “Blown-sand erosion test setup.”] 

Comments 6: [Line 266: The sentence “when it increases to…., it decreases…” is not easy to understand. Please explain the concept behind it better.]

Response 6: [Thank you for this comment. The description of the erosion-velocity effect in Section 3.1 has been rewritten for clarity. The revised text now separately reports the residual tensile-strength reduction at 26 m/s and 31 m/s and explains the progressive degradation with increasing erosion velocity. ] 

Comments 7: [In figure 5 it would be better if the same scale bar is used.]

Response 7: [Thank you for this helpful suggestion. Figure 5 has been revised so that all four subplots use identical left- and right-axis ranges, enabling more direct comparison of the residual tensile strength and elastic modulus under the four investigated erosion factors. The graphical formats have also been standardized. ] 

Comments 8: [The sentence related to the magnification of sem analysis can be written at the beginning of the paragraph instead of repeat it for each figure of the paragraph.]

Response 8: [Thank you for the suggestion. The SEM magnification is now stated once at the beginning of Section 3.2 for Figures 6–9, and the repeated magnification descriptions have been removed from the subsequent discussion.] 

Comments 9: [In order to better evidence the effect of the different values of all investigated parameters it would be more suitable to use the same scale range in the color bar of DIC analysis for each table from table 3 to table 6.]

Response 9: [Thank you for this helpful suggestion. We considered using an identical color-bar range for all DIC maps. However, the strain magnitudes vary considerably among different erosion conditions and load levels, and using a single range would substantially compress the lower-strain maps and reduce the visibility of localized strain variations. Therefore, the individual color ranges have been retained to preserve the visibility of strain localization and its development. The color maps are used primarily for qualitative visualization, while quantitative comparisons are based on the measured strain values and the strain-distribution results.] 

Comments 10: [The previous comment is also valid for figures 10,11,12,13.]

Response 10: [Thank you for this suggestion. The units and graphical formats of Figures 10–13 have been standardized. However, slightly different y-axis ranges were retained because the strain magnitudes vary among the investigated conditions. Applying the same maximum range to all panels would compress the lower-strain curves and reduce the visibility of local strain variations. We therefore retained the present y-axis ranges to preserve the clarity of the strain-distribution trends.] 

Comments 11: [Why in Figure 11c, the value of strain gauge increases at a distance of 60mm from the center? Discuss more on this.]

Response 11: [Thank you for pointing out this local feature. We re-examined the strain distribution in Figure 11(c) and added a discussion in Section 3.4. The local increase at approximately +60 mm is confirmed in the experimental data. The revised manuscript explains that this non-monotonic response indicates an asymmetric local strain distribution and may be associated with the spatially heterogeneous distribution of erosion-induced resin and fiber damage. Because the strain gauges provide measurements only at discrete locations, this isolated increase is not interpreted as evidence of a systematic secondary stress-concentration zone.]

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

The authors have addressed almost all of my previous comments, and the manuscript has clearly improved. I have only one remaining minor point. Since three parallel specimens were tested at each parameter level, it would be useful to show the experimental variability for the main mechanical-property results, for example as individual data points or mean ± standard deviation.  After this minor revision, I would have no further comments.

Comments on the Quality of English Language

A final careful revision of the English and wording would also be advisable, as some awkward expressions remain.

Author Response

Comments 1: [The authors have addressed almost all of my previous comments, and the manuscript has clearly improved. I have only one remaining minor point. Since three parallel specimens were tested at each parameter level, it would be useful to show the experimental variability for the main mechanical-property results, for example as individual data points or mean ± standard deviation. After this minor revision, I would have no further comments.]

Response 1: [Thank you for this valuable suggestion. We agree that presenting the experimental variability can improve the reliability and clarity of the mechanical-property results. In the revised manuscript, the variability of the main mechanical properties has been incorporated by adding error bars to Figure 5. The plotted values represent the mean values obtained from three parallel specimens at each parameter level, and the error bars indicate the standard deviation of the measurements. The caption of Figure 5 has been revised accordingly.]