Population Based Structural Health Monitoring: Homogeneous Offshore Wind Model Development

Black, Innes Murdo; Häckell, Moritz Werther; Kolios, Athanasios

doi:https://doi.org/10.5194/wes-2022-93

Preprints

https://doi.org/10.5194/wes-2022-93

Preprints

10 Oct 2022

| 10 Oct 2022

Status: a revised version of this preprint was accepted for the journal WES and is expected to appear here in due course.

Population Based Structural Health Monitoring: Homogeneous Offshore Wind Model Development

Innes Murdo Black, Moritz Werther Häckell, and Athanasios Kolios

Abstract. This is a development of the preceding paper that introduced the idea and methodology of population-based structural health monitoring (PBSHM). PBSHM involves transferring knowledge from one structure to a different structure so that predictions about the structural health on each of the members in the population can be inferred. One of the most important aspects of PBSHM involves using the information on the source domain structure and the target domain structure to create an effective classifier. Domain adaptation is a subcategory of transfer learning that can create a general classifier using both the source and target domain structures to create an enhanced overall classifier of the entire population. This paper presents a novel domain adaptation model for PBSHM in offshore wind.

Received: 04 Oct 2022 – Discussion started: 10 Oct 2022

Download & links

Innes Murdo Black, Moritz Werther Häckell, and Athanasios Kolios

Status: closed

RC1:
'Comment on wes-2022-93', Anonymous Referee #1, 04 Nov 2022
This paper proposes a domain adaptation model for population-based structural health monitoring in an offshore wind farm. The idea is to predict the fatigue damage equivalent moments in the jacket support structure of a wind turbine based on the trained machine learning model on another wind turbine in the same wind farm. The proposed idea is novel. However, there are several issues in the paper that should be addressed:

The text is quite confusing for the reader to capture the idea of the proposed method because there are several unclear sentences and a large number of grammatical errors and typos throughout the paper.

The paper may not be well organized. For example, why is the data description included in subsection 2.1? The organization of the article is needed.

There is no literature review in the introduction section to state the novelty and advantage of this work over the previous research studies, e.g., https://doi.org/10.1016/j.renene.2021.01.143, https://doi.org/10.1016/j.renene.2020.10.121, https://doi.org/10.1016/j.ymssp.2021.108519, https://doi.org/10.3390/en15020558, https://doi.org/10.1016/j.renene.2022.07.117.

The motivation for the study should be further emphasized. In particular, the contribution should be highlighted in the introduction and conclusion sections.

The reference style is not consistent throughout the text, as well as the numbering of figures and tables (table 3.1 in line 71, figure 3 in line 195, and figure 4 in line 199, for example)

The description on the train and test data should be provided. For example, what percentage of source and target domains data is used for training and testing the models?

In your domain adaptation model, how do you consider the wind turbine position in the wind farm? Because based on different positions, different severity of loads may be imposed on wind turbines; hence, the trained machine learning model on a wind turbine may not give promising results in another wind turbine.
Citation: https://doi.org/10.5194/wes-2022-93-RC1
- AC1:
  'Reply on RC1', Innes Black, 30 Jan 2023
  Dear Referee,
  Thank you for taking the time to review our paper and providing your valuable feedback. We appreciate your comments and suggestions, which will help us to improve the quality of our work.
  Regarding your points:
  We will revise the text to make it more clear and concise, and also correct all the grammatical errors and typos.
  
  We will revise the paper's organization to make it more clear and concise.
  
  We will include a literature review in the introduction section.
  
  We will emphasize the motivation for the study and highlight the contribution in both the introduction section.
  
  We will revise the reference style to make it consistent throughout the text and the numbering of figures and tables.
  
  We will provide a description of the train and test data, including the percentage of source and target domains data used for training and testing the models.
  
  We have tried to create a general model, one where we do not break down the model into different directions. But it has been noted that one could make a less general model and split the data into wind directions to potentially increase the performance.
  
  The data has the wind direction, mean wind speed, and wind speed standard deviation. We conducted a case study where we compared the damage equivelant moments using the wind direction to split them and compare this against the turbulence intensity. We saw small deviations in the Fréchet number and the Persons correlation.The small deviations may provide improvements but this study was to create a general classifier.
  
  We are grateful for your constructive feedback and will take your comments into consideration in our revised submission.
  Best regards,
  Innes Black
  
  Citation: https://doi.org/10.5194/wes-2022-93-AC1
RC2:
'Comment on wes-2022-93', Anonymous Referee #2, 30 Dec 2022
The paper presents a novel application of domain adaptation - and a critical focus for the progression of SHM (sharing data/information between systems). However, the reviewer cannot recommend publication on the grounds of two concerns. The first requires more thorough reviewing/referencing, and the second concerns the interpretability of the method/notation.

While the authors reference earlier work in PBSHM, there appears to be an overlap that should be more clearly acknowledged. Especially in the introductory section 2. In the reviewer's opinion, the text doesn't justifiably reference the original framework papers - discussions around figure 1, for example. More referencing is required throughout, especially to definitions of the underlying theory of domain adaptation.

The second (more major concern) is that the paper is rather confusing in terms of notation, to the point that the methodology becomes unclear

Both feature data and the feature space appear to be defined as 'X'

The set notation throughout is abnormal - missing brackets? (Maybe a rendering issue.) If the authors are following some other convention it should be clearly defined

page 4: the summations are misleading to the reviewer, and the statement, Ds = X = Xi,s, Yi,s)^N needs more explanation

page 5: there appears to be a notation equating 'domains' for homogeneous transfer, but isn't the transformation (into the encoded space) required for the assumption of (approximate) equivalence - and concerning the probability distribution? Are the authors referring to equivalence in the dimensionality of the feature spaces between domains? It seems, from the earlier notation, that the statement implies [Xs, p(Xs)] = [Xt, p(Xt)]. This would imply the two domains are the same, so there is no need for transfer.

There are few references in the above sections - which should clearly reference established definitions

The 'encoded space' is discussed but never clearly defined, usually represented with 'phi(X)' or similar

p16/17 appear to conflict in using similar notation for the labels themselves and the label spaces, also with respect to page 4

pg 6: 'rewriting' - typo?

pg 12: broken reference to AdaBoost

WTG acronym appears undefined
Citation: https://doi.org/10.5194/wes-2022-93-RC2
- AC2:
  'Reply on RC2', Innes Black, 30 Jan 2023
  Dear Reviewer,
  Thank you for your detailed review and constructive feedback on our paper. We appreciate your efforts in evaluating our work and are grateful for the opportunity to improve it.
  Regarding your first concern, we agree that our referencing could be improved and we will make sure to acknowledge the overlap with earlier work in PBSHM more clearly in the introductory section 2. We will also add more references to the definitions of the underlying theory of domain adaptation throughout the paper to make our work more transparent.
  Regarding the other concerns. We will revise the text to improve the clarity and conciseness of our explanations.
  Specifically, we will:
  Clearly differentiate between feature data and feature space in our notation.
  
  Clearly define our set notation, including any conventions we are following.
  
  Provide more explanation for the statement Ds = X = (Xi,s, Yi,s)^N.
  
  Clearly reference established definitions for the concepts of equivalence between domains and the encoded space.
  
  Revise the notation for the labels and label spaces to avoid confusion.
  
  Correct the typo in "rewriting."
  
  Add a definition for the WTG acronym.
  
  Thank you again for bringing these issues to our attention. We hope that our revisions will meet your expectations and lead to a successful publication.
  Sincerely,
  Innes Black
  
  Citation: https://doi.org/10.5194/wes-2022-93-AC2
RC3:
'Comment on wes-2022-93', Anonymous Referee #3, 05 Jan 2023
The paper presents a new domain adaptation approach for the PBSHM to estimate the fatigue damage equivalent moments for the jacket support structure of an offshore wind farm under the homogeneous population assumption.

Major shortcomings:

The paper should be revised both linguistically and structurally. Currently it is very difficult to follow the idea of the paper.

The tables and figure numbers in the text do not match the figures/tables (e.g. line 199: Figure 3 is probably Figure 4).

The citations are not always clear, because years are missing (e.g. line 21) and poorly integrated into the text flow. In addition, literature sources are missing, such as in the definitions section.

A state of the art is missing, as well as a classification of the proposed methodology in it.

The figure are mostly difficult to read and should be revised.

Unfortunately, for the reasons stated above, I recommend a rejection of the paper for publication. For a possible resubmission, a few more comments:

section 2.1: Where are the wind turbines under consideration located in the wind farm? Where exactly are the strain gage measurement locations. In section 4.2 describe that the transition piece must be the same for a population. Therefore, it should be mentioned here that the height of the transition piece is the same for the wind turbine.

section 4.2: If I understand correctly, the boxplots in Figure 2 belong to the statistical quantities in Table 2. The minima and maxima definitely differ from each other. I also can't imagine that the standard deviation and the mean are exactly the same. Either correct, or explain more precisely where the difference between the two is.

section 6.4: Figure 5 and 6 Look exactly the same. What is the reason for this? The high DEM are often underestimated. Could it be due to the training datasets, since rare events cannot be considered? What is the impact of general model errors on lifetime prediction?
Citation: https://doi.org/10.5194/wes-2022-93-RC3
- AC3:
  'Reply on RC3', Innes Black, 30 Jan 2023
  Dear Reviewer,
  Thank you for taking the time to review our paper. We appreciate your constructive feedback and take your comments into consideration. We apologize for the issues in the text flow and the inconsistencies in the citations and figures/tables.
  We have addressed the following points in our revision:
  The language and structure of the paper have been improved to make the idea easier to follow.
  
  The tables and figures have been corrected to match the text.
  
  The missing years in the citations have been added and the literature sources have been included in the definitions section.
  
  A state of the art and classification of the proposed methodology have been added.
  
  Regarding your specific comments:
  The location of the wind turbines and the strain gage measurement locations have been clarified in section 2.1.
  
  The difference between the minima and maxima in Figure 2 and the standard deviation and mean have been explained more precisely in section 4.2.
  
  The reason for the similarity between Figure 5 and 6 has been explained and the impact of the training datasets and general model errors on lifetime prediction have been discussed in section 6.4.
  
  We hope that our revisions have addressed your concerns and we would be grateful for the opportunity to resubmit the paper for further consideration. Thank you again for your valuable feedback.
  Sincerely,
  Innes Black
  
  Citation: https://doi.org/10.5194/wes-2022-93-AC3
AC4: 'Comment on wes-2022-93', Innes Black, 10 Feb 2023

Dear Referees,
I am writing to express my gratitude for the time and effort you have taken to review our manuscript. Your comments and suggestions have been extremely valuable, and I have worked hard to incorporate them into the revised version of the paper.
I have addressed the technical issues that you pointed out and made several revisions to improve the overall clarity and quality of the paper. I hope that these revisions have addressed your concerns and that you will find the paper to be suitable for publication in the journal.
Kind regards,
Innes

Citation: https://doi.org/10.5194/wes-2022-93-AC4

Status: closed

RC1:
'Comment on wes-2022-93', Anonymous Referee #1, 04 Nov 2022
This paper proposes a domain adaptation model for population-based structural health monitoring in an offshore wind farm. The idea is to predict the fatigue damage equivalent moments in the jacket support structure of a wind turbine based on the trained machine learning model on another wind turbine in the same wind farm. The proposed idea is novel. However, there are several issues in the paper that should be addressed:

The text is quite confusing for the reader to capture the idea of the proposed method because there are several unclear sentences and a large number of grammatical errors and typos throughout the paper.

The paper may not be well organized. For example, why is the data description included in subsection 2.1? The organization of the article is needed.

There is no literature review in the introduction section to state the novelty and advantage of this work over the previous research studies, e.g., https://doi.org/10.1016/j.renene.2021.01.143, https://doi.org/10.1016/j.renene.2020.10.121, https://doi.org/10.1016/j.ymssp.2021.108519, https://doi.org/10.3390/en15020558, https://doi.org/10.1016/j.renene.2022.07.117.

The motivation for the study should be further emphasized. In particular, the contribution should be highlighted in the introduction and conclusion sections.

The reference style is not consistent throughout the text, as well as the numbering of figures and tables (table 3.1 in line 71, figure 3 in line 195, and figure 4 in line 199, for example)

The description on the train and test data should be provided. For example, what percentage of source and target domains data is used for training and testing the models?

In your domain adaptation model, how do you consider the wind turbine position in the wind farm? Because based on different positions, different severity of loads may be imposed on wind turbines; hence, the trained machine learning model on a wind turbine may not give promising results in another wind turbine.
Citation: https://doi.org/10.5194/wes-2022-93-RC1
- AC1:
  'Reply on RC1', Innes Black, 30 Jan 2023
  Dear Referee,
  Thank you for taking the time to review our paper and providing your valuable feedback. We appreciate your comments and suggestions, which will help us to improve the quality of our work.
  Regarding your points:
  We will revise the text to make it more clear and concise, and also correct all the grammatical errors and typos.
  
  We will revise the paper's organization to make it more clear and concise.
  
  We will include a literature review in the introduction section.
  
  We will emphasize the motivation for the study and highlight the contribution in both the introduction section.
  
  We will revise the reference style to make it consistent throughout the text and the numbering of figures and tables.
  
  We will provide a description of the train and test data, including the percentage of source and target domains data used for training and testing the models.
  
  We have tried to create a general model, one where we do not break down the model into different directions. But it has been noted that one could make a less general model and split the data into wind directions to potentially increase the performance.
  
  The data has the wind direction, mean wind speed, and wind speed standard deviation. We conducted a case study where we compared the damage equivelant moments using the wind direction to split them and compare this against the turbulence intensity. We saw small deviations in the Fréchet number and the Persons correlation.The small deviations may provide improvements but this study was to create a general classifier.
  
  We are grateful for your constructive feedback and will take your comments into consideration in our revised submission.
  Best regards,
  Innes Black
  
  Citation: https://doi.org/10.5194/wes-2022-93-AC1
RC2:
'Comment on wes-2022-93', Anonymous Referee #2, 30 Dec 2022
The paper presents a novel application of domain adaptation - and a critical focus for the progression of SHM (sharing data/information between systems). However, the reviewer cannot recommend publication on the grounds of two concerns. The first requires more thorough reviewing/referencing, and the second concerns the interpretability of the method/notation.

While the authors reference earlier work in PBSHM, there appears to be an overlap that should be more clearly acknowledged. Especially in the introductory section 2. In the reviewer's opinion, the text doesn't justifiably reference the original framework papers - discussions around figure 1, for example. More referencing is required throughout, especially to definitions of the underlying theory of domain adaptation.

The second (more major concern) is that the paper is rather confusing in terms of notation, to the point that the methodology becomes unclear

Both feature data and the feature space appear to be defined as 'X'

The set notation throughout is abnormal - missing brackets? (Maybe a rendering issue.) If the authors are following some other convention it should be clearly defined

page 4: the summations are misleading to the reviewer, and the statement, Ds = X = Xi,s, Yi,s)^N needs more explanation

page 5: there appears to be a notation equating 'domains' for homogeneous transfer, but isn't the transformation (into the encoded space) required for the assumption of (approximate) equivalence - and concerning the probability distribution? Are the authors referring to equivalence in the dimensionality of the feature spaces between domains? It seems, from the earlier notation, that the statement implies [Xs, p(Xs)] = [Xt, p(Xt)]. This would imply the two domains are the same, so there is no need for transfer.

There are few references in the above sections - which should clearly reference established definitions

The 'encoded space' is discussed but never clearly defined, usually represented with 'phi(X)' or similar

p16/17 appear to conflict in using similar notation for the labels themselves and the label spaces, also with respect to page 4

pg 6: 'rewriting' - typo?

pg 12: broken reference to AdaBoost

WTG acronym appears undefined
Citation: https://doi.org/10.5194/wes-2022-93-RC2
- AC2:
  'Reply on RC2', Innes Black, 30 Jan 2023
  Dear Reviewer,
  Thank you for your detailed review and constructive feedback on our paper. We appreciate your efforts in evaluating our work and are grateful for the opportunity to improve it.
  Regarding your first concern, we agree that our referencing could be improved and we will make sure to acknowledge the overlap with earlier work in PBSHM more clearly in the introductory section 2. We will also add more references to the definitions of the underlying theory of domain adaptation throughout the paper to make our work more transparent.
  Regarding the other concerns. We will revise the text to improve the clarity and conciseness of our explanations.
  Specifically, we will:
  Clearly differentiate between feature data and feature space in our notation.
  
  Clearly define our set notation, including any conventions we are following.
  
  Provide more explanation for the statement Ds = X = (Xi,s, Yi,s)^N.
  
  Clearly reference established definitions for the concepts of equivalence between domains and the encoded space.
  
  Revise the notation for the labels and label spaces to avoid confusion.
  
  Correct the typo in "rewriting."
  
  Add a definition for the WTG acronym.
  
  Thank you again for bringing these issues to our attention. We hope that our revisions will meet your expectations and lead to a successful publication.
  Sincerely,
  Innes Black
  
  Citation: https://doi.org/10.5194/wes-2022-93-AC2
RC3:
'Comment on wes-2022-93', Anonymous Referee #3, 05 Jan 2023
The paper presents a new domain adaptation approach for the PBSHM to estimate the fatigue damage equivalent moments for the jacket support structure of an offshore wind farm under the homogeneous population assumption.

Major shortcomings:

The paper should be revised both linguistically and structurally. Currently it is very difficult to follow the idea of the paper.

The tables and figure numbers in the text do not match the figures/tables (e.g. line 199: Figure 3 is probably Figure 4).

The citations are not always clear, because years are missing (e.g. line 21) and poorly integrated into the text flow. In addition, literature sources are missing, such as in the definitions section.

A state of the art is missing, as well as a classification of the proposed methodology in it.

The figure are mostly difficult to read and should be revised.

Unfortunately, for the reasons stated above, I recommend a rejection of the paper for publication. For a possible resubmission, a few more comments:

section 2.1: Where are the wind turbines under consideration located in the wind farm? Where exactly are the strain gage measurement locations. In section 4.2 describe that the transition piece must be the same for a population. Therefore, it should be mentioned here that the height of the transition piece is the same for the wind turbine.

section 4.2: If I understand correctly, the boxplots in Figure 2 belong to the statistical quantities in Table 2. The minima and maxima definitely differ from each other. I also can't imagine that the standard deviation and the mean are exactly the same. Either correct, or explain more precisely where the difference between the two is.

section 6.4: Figure 5 and 6 Look exactly the same. What is the reason for this? The high DEM are often underestimated. Could it be due to the training datasets, since rare events cannot be considered? What is the impact of general model errors on lifetime prediction?
Citation: https://doi.org/10.5194/wes-2022-93-RC3
- AC3:
  'Reply on RC3', Innes Black, 30 Jan 2023
  Dear Reviewer,
  Thank you for taking the time to review our paper. We appreciate your constructive feedback and take your comments into consideration. We apologize for the issues in the text flow and the inconsistencies in the citations and figures/tables.
  We have addressed the following points in our revision:
  The language and structure of the paper have been improved to make the idea easier to follow.
  
  The tables and figures have been corrected to match the text.
  
  The missing years in the citations have been added and the literature sources have been included in the definitions section.
  
  A state of the art and classification of the proposed methodology have been added.
  
  Regarding your specific comments:
  The location of the wind turbines and the strain gage measurement locations have been clarified in section 2.1.
  
  The difference between the minima and maxima in Figure 2 and the standard deviation and mean have been explained more precisely in section 4.2.
  
  The reason for the similarity between Figure 5 and 6 has been explained and the impact of the training datasets and general model errors on lifetime prediction have been discussed in section 6.4.
  
  We hope that our revisions have addressed your concerns and we would be grateful for the opportunity to resubmit the paper for further consideration. Thank you again for your valuable feedback.
  Sincerely,
  Innes Black
  
  Citation: https://doi.org/10.5194/wes-2022-93-AC3
AC4: 'Comment on wes-2022-93', Innes Black, 10 Feb 2023

Dear Referees,
I am writing to express my gratitude for the time and effort you have taken to review our manuscript. Your comments and suggestions have been extremely valuable, and I have worked hard to incorporate them into the revised version of the paper.
I have addressed the technical issues that you pointed out and made several revisions to improve the overall clarity and quality of the paper. I hope that these revisions have addressed your concerns and that you will find the paper to be suitable for publication in the journal.
Kind regards,
Innes

Citation: https://doi.org/10.5194/wes-2022-93-AC4

Innes Murdo Black, Moritz Werther Häckell, and Athanasios Kolios

Viewed

Total article views: 707 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
478	200	29	707	19	15

HTML: 478
PDF: 200
XML: 29
Total: 707
BibTeX: 19
EndNote: 15

Views and downloads (calculated since 10 Oct 2022)

Month	HTML	PDF	XML	Total
Oct 2022	70	26	3	99
Nov 2022	52	10	4	66
Dec 2022	36	11	2	49
Jan 2023	73	21	9	103
Feb 2023	55	17	2	74
Mar 2023	20	2	0	22
Apr 2023	15	18	0	33
May 2023	6	4	1	11
Jun 2023	5	1	0	6
Jul 2023	14	15	1	30
Aug 2023	11	7	1	19
Sep 2023	27	12	1	40
Oct 2023	6	4	0	10
Nov 2023	1	4	0	5
Dec 2023	14	12	3	29
Jan 2024	19	2	0	21
Feb 2024	11	6	0	17
Mar 2024	26	19	0	45
Apr 2024	17	9	2	28

Cumulative views and downloads (calculated since 10 Oct 2022)

Month	HTML	PDF	XML	Total
Oct 2022	70	26	3	99
Nov 2022	52	10	4	66
Dec 2022	36	11	2	49
Jan 2023	73	21	9	103
Feb 2023	55	17	2	74
Mar 2023	20	2	0	22
Apr 2023	15	18	0	33
May 2023	6	4	1	11
Jun 2023	5	1	0	6
Jul 2023	14	15	1	30
Aug 2023	11	7	1	19
Sep 2023	27	12	1	40
Oct 2023	6	4	0	10
Nov 2023	1	4	0	5
Dec 2023	14	12	3	29
Jan 2024	19	2	0	21
Feb 2024	11	6	0	17
Mar 2024	26	19	0	45
Apr 2024	17	9	2	28

Viewed (geographical distribution)

Total article views: 690 (including HTML, PDF, and XML) Thereof 690 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 25 Apr 2024

Short summary

Population based structural health monitoring is a low-cost monitoring campaign. The cost reduction from this type of digital enabled asset management tool is manifested by sharing information, in this case a wind farm foundation, within the population. By sharing the information in the wind farm this reduces the amount of sensors and physical model updating, reducing the cost of the monitoring campaign.


Total:	0
HTML:	0
PDF:	0
XML:	0