Deep Learning-Enriched Stress Level Identification of Pretensioned Rods via Guided Wave Approaches

Round 1

Reviewer 1 Report

The paper "Deep-Learning Enriched Stress Level prediction of Pretensioned Rods towards Structural Health Monitoring via Simulation of Guided Wave Approaches" is interesting and could be considered, however some minor corrections are needed:

 

(a) The title is too long and confusing to readers;

(b) The abstract should be more concise, show some quantitative results and the scientific innovation of the document;

(c) The general format of the references is not clear, and it is outside the one used in the MDPI journals, please review all the formatting of the document;

(d) The authors do not delve into a review of the state of the art and literature on the topic, note that some current studies are absent, and the list of references is short for potential readers;

(e) A greater detail of the experimental procedure is necessary, mainly in the description of the models used;

(f) Figure 7 is composed, so it should be subdivided into (a), (b), (c) and other items;

(g) The conclusion can be more objective and according to the document in general, some future perspective and gaps in the themes should be quickly mentioned;

(i) "This represents that the features in this label were much easier to be separated from others. However, the clusters in red diamond (Base state) and yellow circle (20% UTS) located on the right side were overlapped. At least 1/4 data was mixed and difficult to separate. In addition, the rod samples prestressed in 60% UTS (blue lower triangle) and 80% UTS (purple star) were tangled together. Fig. 8(b) represented the feature maps from the last layer of CNN. Clearly, after 8 layers' processing, most of the sample were separated expect one outlier in base class and a small overlap between the samples in 60% UTS and 80% UTS. Results demonstrated that features became more sensitive after operating the whole CNN process." This part of the paper is unclear, the authors should explain further.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper uses the CNN method to simulate the prestress loss of PC structures based on the guided wave. In general, the implementation of the CNN has been detailed and the paper is well-written, but the following comments are suggested to consider before it is accepted.

 

1. Which method the authors used for the guided wave analysis in the frequency domain? It should be clearly presented.
2. CNN is used for classification rather than regression. Thus, is it suitable to claim prediction in the title?
3. The authors did not mention the hyperparameter determination, and it seems that the hyperparameters are empirically determined. Bayesian optimization is recommended to determine the critical hyperparameters. The following references regarding Bayesian optimization or other hyperparameter determination are suggested to enhance the corresponding section: Image-based post-disaster inspection of reinforced concrete bridge systems using deep learning with Bayesian optimization; and Probabilistic framework with Bayesian optimization for predicting typhoon-induced dynamic responses of a long-span bridge.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors made all the suggested corrections.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The authors have addressed my previous comments. A minor suggestion is listed as follows: for ordinary structural health monitoring systems, the anomalies (such as outliers and missing data) inevitably exist in the monitoring data (Structural Health Monitoring, 2022, 14759217211053779). Does this issue exist in Guided Wave? If not, it can be an advantage of the presented method and can be explained in the Introduction.

Author Response

Please see the attachment.

Author Response File: Author Response.docx