PrimaVera: Synergising Predictive Maintenance

Round 1

Reviewer 1 Report

• In 2.1., Data acquisition; line 57; there is a statement dividing sensing methods, such as next level, more advanced and most advanced. A reference is missing, and the authors confusing CM methods with predictive maintenance or smart predictive maintenance (Smart PDM). Additional perspectives regarding CBM/CM and PdM are necessary to be addressed, such as; randomness of the picked up signals/data especially when talking about vibration; measuring frequency/policy because it is different for different deterioration processes (e.g. fatigue or wear); PdM accuracy with respect to data acquisition; and PdM applicability and cost-effectiveness with respect to industry demands.
• In page3, 2.2. Data processing and diagnosis; line 104: Current monitoring … use of physical models. It seems to be that the authors focus on only one type of CM technique, i.e. those using streaming data, while it is not always necessary having this type of CM especially when monitoring the condition of mechanical components/equipment. The latter takes, in general, long time from damage initiation to potential failure and real failure, but the authors unfortunately generalized their judgement.
• In 2.3.; Line 134, and in 2.5, line 184; the first challenge is the gap between component and system …. .: It contradicts the authors´ suggestion in page 8, line 298, i.e. using also critical components. The failure of a significant/critical component (i.e. the components whose failures are either expensive or dangerous) such as bearing or shaft leads directly to machine/system failure. Machine availability is usually based on the availability of these components.
• In 2.2. Maintenance and logistics optimization: maintenance management cover a wide range of activities that should be planned and conducted based on the information provided by CM in order to avoid failures in a cost-effective way. Maintenance is an economic tool and not only technical tool. Therefore, maintenance action should be done at as accurate as possible timing to reduce unnecessary economic losses. Securing the functionality of every significant component in a machine is actually securing the availability of the machine. I believe these issues and much more are very important to be studied and discussed in this section.
• In 2.5, lines 215-217; which unresolved decisions that are needed to be addressed before data-driven maintenance can be actually implemented? It is important to describe clearly, which unresolved decisions, why and how.
• In 3. Obstacles to overcome; we see incomplete introduction of the obstacles, no clear motivations, necessity of overcoming them supported by references; and the solutions suggested of how to overcome the obstacles are in general terms which cannot be acknowledged without a more detailed discussion with respect to, for example why, how and which benefit can be gained.
• In 4. New generic process model; the model is an extension of the model introduced by Jardine et al. (2006). Many models have been developed and introduced in literature since then, which can be much more advanced than that. It is necessary to highlight what other models are introduced in literature and why you selected just this model (motivations for -and argumentation- the selection are very necessary). The model developed using Jardine’s experience from, among others, his model/software EXAKT. In addition, the authors needed to motivate and discuss why the cycle is closed and additional two steps were added to the original model. Also, why and how you will incorporate all the information into a strategic asset management. The generic model of the PdM suggested in the paper misses important parts that are; visualizing/showing its interoperability through developing application examples highlighting data input, output and processes; reasoning and discussion of the potential results and model applicability.
• In page 8, lines 306—309, the authors demand using realistic assumption (which is necessary!), but they make unrealistic assumption when excluding effects of changing operational conditions on measurements and isolate these from the effects caused by failures. It is well-known that changing operational condition influence appreciably deterioration rate, which in turn influences the condition of the monitored components/equipment and consequently prediction accuracy.
• The methodology introduced, i.e. innovative solutions are developed together with stakeholders. It means we cannot realize any potential results/solutions before the end of the project. This makes the manuscript very weak because it is not easy to any reader to realize properly; how the challenges will be treated, what solutions can be expected, applicability and soundness of the solution, etc.

Author Response

Dear reviewer,

First of all we would like to thank you for your time reviewing this article and the comments made. They have been valuable in the process of improving our article. We have addressed the issues raised, please find our remarks, indicated by a '>', below.

Kind regards,

Bram Ton

------------------------------------------------------------

In 2.1., Data acquisition; line 57; there is a statement dividing sensing methods, such as next level, more advanced and most advanced. A reference is missing, and the authors confusing CM methods with predictive maintenance or smart predictive maintenance (Smart PDM). Additional perspectives regarding CBM/CM and PdM are necessary to be addressed, such as; randomness of the picked up signals/data especially when talking about vibration; measuring frequency/policy because it is different for different deterioration processes (e.g. fatigue or wear); PdM accuracy with respect to data acquisition; and PdM applicability and cost-effectiveness with respect to industry demands.
> Within academic literature the term 'Smart Predictive Maintenance' has not (yet) taken on (6 results on scopus.com). Furthermore, the distinction between 'Predictive maintenance' and 'Smart Predictive Maintenance' has not been established yet. Hence, in this paper we will only use the term 'Predictive Maintenance'.
> Randomness of signals, measuring frequency/policy and PdM accuracy are taken into account by an optimal sensing strategy decision support tool. A paragraph has been added to 2.1 to reflect this.
> The cost-effectiveness of PdM is addressed in the introduction of the article. " Estimates of the impact of predictive maintenance vary widely, but in general the return of investment is deemed to be favourable [3]."

In page3, 2.2. Data processing and diagnosis; line 104: Current monitoring … use of physical models. It seems to be that the authors focus on only one type of CM technique, i.e. those using streaming data, while it is not always necessary having this type of CM especially when monitoring the condition of mechanical components/equipment. The latter takes, in general, long time from damage initiation to potential failure and real failure, but the authors unfortunately generalized their judgement.
> We agree that this part has been formulated in a too-limited way. In fact, within the PrimaVera project also CM techniques other than those based on (streaming) data will be developed, like structural health monitoring techniques based on vibrations and other dynamic responses. The latter techniques are much more physics-based. This has been modified in the manuscript in section 2.2

In 2.3.; Line 134, and in 2.5, line 184; the first challenge is the gap between component and system …. .: It contradicts the authors´ suggestion in page 8, line 298, i.e. using also critical components. The failure of a significant/critical component (i.e. the components whose failures are either expensive or dangerous) such as bearing or shaft leads directly to machine/system failure. Machine availability is usually based on the availability of these components.
> The reviewer is right that typically a small number of (critical) parts determines the availability of the complete system. That is exactly the point we also want to make here. However, the challenge for a large complex system is to find those critical components, in which effort should be put to predict failures. The text in section 2.3 has been modified to clarify this.

In 2.4. Maintenance and logistics optimization: maintenance management cover a wide range of activities that should be planned and conducted based on the information provided by CM in order to avoid failures in a cost-effective way. Maintenance is an economic tool and not only technical tool. Therefore, maintenance action should be done at as accurate as possible timing to reduce unnecessary economic losses. Securing the functionality of every significant component in a machine is actually securing the availability of the machine. I believe these issues and much more are very important to be studied and discussed in this section.
> We fully agree that (1) maintenance optimisation requires an economic decision for which technical aspects are an input only and (2) that unavailability of a critical component causes unavailability of the complete system. We have drastically modified the second paragraph of Section 2.4 to emphasise these points.

In 2.5, lines 215-217; which unresolved decisions that are needed to be addressed before data-driven maintenance can be actually implemented? It is important to describe clearly, which unresolved decisions, why and how.
> At hindsight, the term "unresolved decisions" was not clear. It has now been rephrased to "organisational decisions", these decisions are now mentioned in the text (line 238 -- 239).

In 3. Obstacles to overcome; we see incomplete introduction of the obstacles, no clear motivations, necessity of overcoming them supported by references; and the solutions suggested of how to overcome the obstacles are in general terms which cannot be acknowledged without a more detailed discussion with respect to, for example why, how and which benefit can be gained.
> The reviewer has a valid point, motivation and necessity of the obstacles to overcome have been added to the article. The obstacles identified are novel insights established by the PrimaVera project and therefore are scarcely mentioned in existing literature, so far only Wagner et al. (2019) mentions some items.

In 4. New generic process model; the model is an extension of the model introduced by Jardine et al. (2006). Many models have been developed and introduced in literature since then, which can be much more advanced than that. It is necessary to highlight what other models are introduced in literature and why you selected just this model (motivations for -and argumentation- the selection are very necessary). The model developed using Jardine’s experience from, among others, his model/software EXAKT. In addition, the authors needed to motivate and discuss why the cycle is closed and additional two steps were added to the original model. Also, why and how you will incorporate all the information into a strategic asset management. The generic model of the PdM suggested in the paper misses important parts that are; visualizing/showing its interoperability through developing application examples highlighting data input, output and processes; reasoning and discussion of the potential results and model applicability.
> The reviewer is right, more advanced methods are indeed present in literature. These methods are now referenced within the article. Motivation for closing the cycle is added together with the justification for the organisational and human factors stage. Furthermore the evaluation of the model its usability and applicability is addressed by explicitly mentioning that this is done by applying the model to the demonstrators.

In page 8, lines 306—309, the authors demand using realistic assumption (which is necessary!), but they make unrealistic assumption when excluding effects of changing operational conditions on measurements and isolate these from the effects caused by failures. It is well-known that changing operational condition influence appreciably deterioration rate, which in turn influences the condition of the monitored components/equipment and consequently prediction accuracy.
> We agree on this point, and realise that the formulation is not clear. We do mean to incorporate effects of changes in operating conditions, the point here is to properly separate these effects form changes in measurements due to degradation or failures. The text has been modified.

The methodology introduced, i.e. innovative solutions are developed together with stakeholders. It means we cannot realize any potential results/solutions before the end of the project. This makes the manuscript very weak because it is not easy to any reader to realize properly; how the challenges will be treated, what solutions can be expected, applicability and soundness of the solution, etc.
> The reviewer is right, no specific solutions are provided yet. Obtaining these solutions is the goal of the project and will be disseminated in future publications. The article does mention research directions to achieve these solutions. The elements of section 4.1.1 - 4.1.5 have been updated to provide a more detailed description of the work being done in the project.

Reviewer 2 Report

The paper is well written and it is rather interesting although the authors describe obstacles to tackle in order to utilize predictive maintenance at its full potential and emphasize many problems without writing how to overcome them effectively. The paper is a good overview of the topic of predictive maintenance but does not provide any specific solutions to the problems described.

The paper is written in a rather general way, for example, "novel sensing techniques and sources of data gathering are explored to improve the reliability and the quality of data" (lines 300-301). Such sentences are very vague and a research paper should be more specific especially when it describes the authors' own contributions. Likewise, the following sentences: "The PrimaVera project will propose a framework to construct these causal inference graphs and extend them to perform in an industrial setting" (lines 316-318) or "These hybrid prognostic methods will result in considerably more accurate and efficient methods" (lines 328-329).

Additionally, there are a few statements which require justification, such as:
"Current predictive maintenance solutions often focus on a single step in the predictive maintenance chain, with poor alignment to the rest of the workflow" (line 256) or "Current applications of predictive maintenance involve a large number of manual steps." (line 262 ) please show some examples, moreover, I suggest at least adding the word "usually" here and I wonder why "manual steps".
And first of all the final statement "The results of the PrimaVera project will synergise the individual building blocks of predictive maintenance and will pave the way towards utilising predictive maintenance at its full potential." is only a promise which should be followed in order for it to be verified.

Some minor flaws should be also corrected, such as

line 46 "Section 5 is details" should be "Section 5 details"
line 50 "Predictive maintenance entails six steps; (...)" - I suggest using a colon instead of a semicolon, and these 6 steps are not clearly identified. That is why I suggest in line 51 "asset management and organisational factors, and human factors" (like in next subsections) or at least "asset management, and human factors" instead of "asset management and human factors".
line 282 "data acquisition,data processing" a space is needed after the comma
line 287 "based on prognostics maintenance and associated logistics are optimised" - I can not see where is the subject in this sentence.
line 292 "The subsequent sections will focus (...) and describe" sections can not focus or describe (only the humans can) that is why it should be also corrected

Author Response

Dear reviewer,

First of all we would like to thank you for your time reviewing this article and the comments made. They have been valuable in the process of improving our article. We have addressed the issues raised, please find our remarks, indicated by a '>', below.

Kind regards,

Bram Ton

------------------------------------------------------------

The paper is well written and it is rather interesting although the authors describe obstacles to tackle in order to utilize predictive maintenance at its full potential and emphasize many problems without writing how to overcome them effectively.
> The elements of section 4.1.1 - 4.1.5 have been updated to provide a more detailed description of the work being done in the project.

The paper is a good overview of the topic of predictive maintenance but does not provide any specific solutions to the problems described.
> Reviewer 1 also raised this issue. We copy the explanation given at that comment of reviewer 1:
"> The reviewer is right, no specific solutions are provided yet. Obtaining these solutions is the goal of the project and will be disseminated in future publications. The article does mention research directions to achieve these solutions. The elements of section 4.1.1 - 4.1.5 have been updated to provide a more detailed description of the work being done in the project."

The paper is written in a rather general way, for example, "novel sensing techniques and sources of data gathering are explored to improve the reliability and the quality of data" (lines 300-301). Such sentences are very vague and a research paper should be more specific especially when it describes the authors' own contributions.
> Sections 4.1.x have been modified in general to make them more specific.

Likewise, the following sentences: "The PrimaVera project will propose a framework to construct these causal inference graphs and extend them to perform in an industrial setting" (lines 316-318) or "These hybrid prognostic methods will result in considerably more accurate and efficient methods" (lines 328-329).
> The reviewer has a valid point. Section 4.1.3 has been altered considerably to make it more specific.

Additionally, there are a few statements which require justification, such as:
"Current predictive maintenance solutions often focus on a single step in the predictive maintenance chain, with poor alignment to the rest of the workflow" (line 256) or "Current applications of predictive maintenance involve a large number of manual steps." (line 262 ) please show some examples, moreover, I suggest at least adding the word "usually" here and I wonder why "manual steps".
> These statements are based on expert interviews, hence references to literature have been omitted.
Usually has been added and "manual steps" has been replaced by "non-automated procedures".

And first of all the final statement "The results of the PrimaVera project will synergise the individual building blocks of predictive maintenance and will pave the way towards utilising predictive maintenance at its full potential." is only a promise which should be followed in order for it to be verified.
> The statement has been extended to make it verifiable. "..., achieving higher availability of assets at lower cost."

Some minor flaws should be also corrected, such as
> Thank you for mentioning them, albeit minor, they still contribute to the quality of the paper.

line 46 "Section 5 is details" should be "Section 5 details"
> fixed
line 50 "Predictive maintenance entails six steps; (...)" - I suggest using a colon instead of a semicolon, and these 6 steps are not clearly identified. That is why I suggest in line 51 "asset management and organisational factors, and human factors" (like in next subsections) or at least "asset management, and human factors" instead of "asset management and human factors".
line 282 "data acquisition,data processing" a space is needed after the comma
> fixed
line 287 "based on prognostics maintenance and associated logistics are optimised" - I can not see where is the subject in this sentence.
> fixed
line 292 "The subsequent sections will focus (...) and describe" sections can not focus or describe (only the humans can) that is why it should be also corrected
> In our opinion, these type of phrases are common practice within academic publications. Therefore they are left as they are.

Reviewer 3 Report

The scope of the analysis and the planned project work are impressive and the span of the consortium is extensive. I would argue that this is the strength of the paper that needs to be brought out much more strongly. The paper itself is a consortium-developed prospectus on how predictive maintenance can be broken down into six distinct steps which combine to form a process model. The process model is in turn used to organize a research and development plan. Cross-cutting obstacles are identified. What I call a DevOps methodology (in the paper it is called Action Research) is applied to three industrial application areas in such a way that combinations of steps are investigated in rapid cycles of trying objectives-based approaches. There are no results yet from the applications work.

As mentioned, the strength of the paper stems from how the public-private partnership of industry and academic experts came together to agree on the framework and the priorities. It would be most interesting to have some discussion about how agreements were reached on extending the condition-based model and why the additional steps were added. Typically a step comprises tightly coupled elements that are not so easily separated conceptually and/or operationally, while the relations between and among steps are more separable and the separation is used to advantage. For example, data acquisition is often separated from data processing because connecting and transmitting data from various kinds of sources are very different from applying data models and normalization processes to prepare data for say diagnostic models. However, it is interesting in the paper that data process and diagnosis are closely linked even though the paper distinguishes monitoring from diagnosis and also notes the data are typically lacking for diagnosis. Similarly prognostics could be linked with monitoring in that some changes in state could be incipient signals used to predict when maintenance is needed, i.e. before an event becomes a diagnostic situation. I mention these examples not to dispute the analysis but to recommend that “how” these were determined are the findings I would find as valuable as the breakdown and priorities themselves.

The section on ‘obstacles to overcome’ is also interesting as a public-private consortium agreement. Of the four that apparently rose in priority, orchestration is particularly interesting in that its prioritization is an argument that the relationships between steps cannot be ignored. Automation, data uncertainty and human factors are referenced far more often than orchestration. Of note, orchestration and automation are not mentioned again and are therefore left hanging as concepts. It is important to tie these into the framework and/or the methodology.

The comments above are about the first five sections. I am recommending interspersing discussion about how and why the consortium reached agreements on what formed the overall framework and its elements. The section that is least developed is the ‘methodology’ section. Each one of the industry projects is significant and valuable but potentially very large in its own right. All three together is a massive undertaking. How the potential scale is being managed or constrained needs to be addressed given the scale and complexity of each step considered can also be very large. While I understand and agree on the value of a diverse set of applications to prevent lock in on problem-specific approaches, it is not clear how more generalized aspects of the approaches developed for one application are tested on another application. This of course makes the project even larger and more complex. Nevertheless, a better understanding of plans for managing scale and cross linking results is also needed to bring the paper to a satisfactory close.

Author Response

Dear reviewer,

First of all we would like to thank you for your time reviewing this article and the comments made. They have been valuable in the process of improving our article. We have addressed the issues raised, please find our remarks, indicated by a '>', below.

Kind regards,

Bram Ton

------------------------------------------------------------

The scope of the analysis and the planned project work are impressive and the span of the consortium is extensive. I would argue that this is the strength of the paper that needs to be brought out much more strongly. The paper itself is a consortium-developed prospectus on how predictive maintenance can be broken down into six distinct steps which combine to form a process model. The process model is in turn used to organize a research and development plan. Cross-cutting obstacles are identified. What I call a DevOps methodology (in the paper it is called Action Research) is applied to three industrial application areas in such a way that combinations of steps are investigated in rapid cycles of trying objectives-based approaches. There are no results yet from the applications work.
> Thank you for pointing out the DevOps methodology, we have noticed this is a valuable method for implementing the demonstrators (Section 5.1.1). The lack of results has also been raised by reviewer 1 and 2, the response that we formulated is copied here: "> The reviewer is right, no specific solutions are provided yet. Obtaining these solutions is the goal of the project and will be disseminated in future publications. The article does mention research directions to achieve these solutions. The elements of section 4.1.1 - 4.1.5 have been updated to provide a more detailed description of the work being done in the project."

As mentioned, the strength of the paper stems from how the public-private partnership of industry and academic experts came together to agree on the framework and the priorities. It would be most interesting to have some discussion about how agreements were reached on extending the condition-based model and why the additional steps were added.
> Reviewer 1 rightfully raised the question why we didn't mention the more advanced process models available in literature. Based on this remark we altered the beginning of Section 4 significantly to incorporate the newer process models. These newer process models already have the additional steps included.

Typically a step comprises tightly coupled elements that are not so easily separated conceptually and/or operationally, while the relations between and among steps are more separable and the separation is used to advantage. For example, data acquisition is often separated from data processing because connecting and transmitting data from various kinds of sources are very different from applying data models and normalization processes to prepare data for say diagnostic models. However, it is interesting in the paper that data process and diagnosis are closely linked even though the paper distinguishes monitoring from diagnosis and also notes the data are typically lacking for diagnosis. Similarly prognostics could be linked with monitoring in that some changes in state could be incipient signals used to predict when maintenance is needed, i.e. before an event becomes a diagnostic situation. I mention these examples not to dispute the analysis but to recommend that “how” these were determined are the findings I would find as valuable as the breakdown and priorities themselves.
> We agree that the boundaries between some of the elements and steps are not that solid. The choices that we made in structuring them in our process model are primarily based on experience of the involved researchers in executing predictive maintenance research projects with industry, rather than on thorough theoretical analyses. And that indeed revealed that processing the acquired data in many cases already brought us close to diagnosing the system. Also the mentioned link between monitoring and prognostics is a relation we recognise from practice. Since this experience has only been implicitly used to develop the proposed model, we chose to not elaborate on the specific clustering in this work.

The section on ‘obstacles to overcome’ is also interesting as a public-private consortium agreement. Of the four that apparently rose in priority, orchestration is particularly interesting in that its prioritization is an argument that the relationships between steps cannot be ignored. Automation, data uncertainty and human factors are referenced far more often than orchestration. Of note, orchestration and automation are not mentioned again and are therefore left hanging as concepts. It is important to tie these into the framework and/or the methodology.
> Orchestration is now addressed in section 4.1.4 "Maintenance and logistics optimisation". Automation is addressed in section 2.3 "Prognostics". Moreover, we fully agree that the orchestration is an essential obstacle, so this has been linked to the introduction of 4.1

The comments above are about the first five sections. I am recommending interspersing discussion about how and why the consortium reached agreements on what formed the overall framework and its elements. The section that is least developed is the ‘methodology’ section. Each one of the industry projects is significant and valuable but potentially very large in its own right. All three together is a massive undertaking. How the potential scale is being managed or constrained needs to be addressed given the scale and complexity of each step considered can also be very large. While I understand and agree on the value of a diverse set of applications to prevent lock in on problem-specific approaches, it is not clear how more generalized aspects of the approaches developed for one application are tested on another application. This of course makes the project even larger and more complex. Nevertheless, a better understanding of plans for managing scale and cross linking results is also needed to bring the paper to a satisfactory close.
> A new section (Section 5.1.1) has been added to the article detailing the implementation of the demonstrators and addresses the scaling issue..

Round 2

Reviewer 1 Report

• In 2.1; Several additional aspects are necessary to be addressed, such as; The randomness of the picked up signals/data especially when talking about vibration; measuring frequency/policy because it is different for different deterioration processes (e.g. fatigue or wear); PdM accuracy with respect to data acquisition; and PdM applicability and cost-effectiveness with respect to industry demands.
• In 2.3; the first challenge is the gap between component and system …. .: It contradicts the authors´ suggestion, i.e. using also critical components. The failure of a significant/critical component (i.e. the components whose failures are either expensive or dangerous) such as bearing or shaft leads directly to machine/system failure. Machine availability is usually based on the availability of these components.
• In 2.4. Maintenance and logistics optimization: maintenance management cover a wide range of activities that should be planned and conducted based on the information provided by CM in order to avoid failures in a cost-effective way. Maintenance is an economic tool and not only technical tool. Therefore, maintenance action should be done at as accurate as possible timing to reduce unnecessary economic losses. Securing the functionality of every significant component in a machine is securing the availability of the machine. I believe these issues, and much more are very important to be addressed and discussed in this section.
• In 2.5; How the author(s) knows there are a number of organisational issues needed to be identified and addressed? It should be clarified.
• In 3. Obstacles to overcome; It is good adding new references, but these references are technical reports reflecting interviews and are not scientific papers providing results based on scientific researches (as it is expressed). It is always necessary to give examples and argumentations for the claims provided by the technical reports’ to motivate addressing the obstacles and the necessity of overcoming them, because you introduce very hard statements in the description of the obstacles without scientific proof.
• In 4. It is good adding new references, but the model is still an extension of the model introduced by Jardine et al. (2006) and improved/modified in previous publications. Many other models have been developed and introduced in literature, which can be much more advanced than that. It is necessary to highlight some of the models introduced in literature and argue for why you selected just this model, i.e. motivations and argumentation for the selection are very necessary.
• The generic model of the PdM suggested in the paper misses important parts that are; visualizing/showing its interoperability in real application(s) and highlighting data input, output, processes and results; reasoning and discussion of the  results and model applicability. What is introduce is planned of how the project will be conducted, which is not enough.
• The methodology introduced (innovative solutions are developed together with stakeholders). It means we cannot realize any results/solutions before the end of the project. This makes the manuscript very weak because it is not easy for any reader to realize properly; how the treatment of the challenges is done in reality, how realistic were the assumptions, whether the suggested solutions will work, what additional challenges are faced, applicability and soundness of the solutions, etc
• In 4 and 5 the authors introduce a roadmap showing how Prima Vera project will possibly handle the challenges introduced in 2 and 3.
• Therefore, the conclusions are mainly a summary of what has been introduced. The conclusions usually based on results, data gathered methodology, analysis and analysis tools, research methodology, etc. 

Author Response

Dear reviewer,

First of all we would like to thank you for your time reviewing this article and the comments made. Our impression is that in your opinion, one of the major shortcomings of this article is the lack of results. We would like to point out that some of the suggested topics for this special issue "Overcoming the Obstacles to Predictive Maintenance" are:
* Implementation methodologies
* Architectures
* Process modelling and reasoning for predictive maintenance of complex assets

In general topics with this nature tend not to have tangible or concrete results, which is also the case with this article. Nonetheless, it is still very valuable to publish the envisioned research approach in a peer-reviewed journal to ascertain the validity of the approach. The independent feedback on the article has been very valuable to us to fine-tune the approach. Moreover, structuring the challenges and obstacles and relating these to existing insights in scientific literature, as we did in this paper, is valuable for other researchers in the field, even though the specific solutions to these challenges cannot be reported yet.

We have addressed the issues raised, please find our remarks, indicated by a '>', below.

Kind regards,

Bram Ton

------------------------------------------------------------

In 2.1; Several additional aspects are necessary to be addressed, such as; The randomness of the picked up signals/data especially when talking about vibration; measuring frequency/policy because it is different for different deterioration processes (e.g. fatigue or wear); PdM accuracy with respect to data acquisition; and PdM applicability and cost-effectiveness with respect to industry demands.
> Goal of condition monitoring, e.g. fatigue or wear is now listed as an input factor of the optimal sensing strategy decision support tool. The sole purpose of mentioning vibration in this section is as an example. We do not see it necessary to further elaborate on the role of randomness in this context.
> Uncertainty, which entails randomness of data, is now addressed in Section 4.1.2. Methods are suggested to quantify uncertainty.
> The cost-effectiveness and applicability of PdM are addressed in the introduction of the article. "Estimates of the impact of predictive maintenance vary widely, but in general the return of investment is deemed to be favourable [3]. Despite the favourable return on investment, implementation of predictive maintenance in practice is still limited in many industries [4,5]"

In 2.3; the first challenge is the gap between component and system …. .: It contradicts the authors´ suggestion, i.e. using also critical components. The failure of a significant/critical component (i.e. the components whose failures are either expensive or dangerous) such as bearing or shaft leads directly to machine/system failure. Machine availability is usually based on the availability of these components.
> As we already mentioned in our previous revision / response, we agree with the reviewer that typically a small number of (critical) parts determines the availability of the complete system. That is exactly the point we want to make here. However, the challenge for a large complex system is to find those critical components, in which effort should be put to predict failures. This is not a trivial activity, and therefore still one of the challenges in this field. The text in section 2.3 has been modified to clarify this by adding " In the former case, the selection of these critical components, especially for large and complex systems, is not trivial and requires attention. ".


In 2.4. Maintenance and logistics optimization: maintenance management cover a wide range of activities that should be planned and conducted based on the information provided by CM in order to avoid failures in a cost-effective way. Maintenance is an economic tool and not only technical tool. Therefore, maintenance action should be done at as accurate as possible timing to reduce unnecessary economic losses. Securing the functionality of every significant component in a machine is securing the availability of the machine. I believe these issues, and much more are very important to be addressed and discussed in this section.
> We clarify the previous response in more detail. In our opinion the issue of timing is addressed by: "this means that there is an economically optimal moment to perform maintenance that incorporates these costs and the probability of failure or RUL estimate." The challenge of securing significant components is addressed by: "This optimisation is further complicated because assets contain many (critical) components, and grouping maintenance leads to fewer disruptions for the customer and lower logistics costs."


In 2.5; How the author(s) knows there are a number of organisational issues needed to be identified and addressed? It should be clarified.
> New references have been added to support the need for addressing organisational issues. This is reflected in lines 223-239


In 3. Obstacles to overcome; It is good adding new references, but these references are technical reports reflecting interviews and are not scientific papers providing results based on scientific researches (as it is expressed). It is always necessary to give examples and argumentations for the claims provided by the technical reports’ to motivate addressing the obstacles and the necessity of overcoming them, because you introduce very hard statements in the description of the obstacles without scientific proof.
> Main source of identifying the obstacles are our own expert interviews with members from academia and industry. Nonetheless, new sources have been identified to support the statements.


In 4. It is good adding new references, but the model is still an extension of the model introduced by Jardine et al. (2006) and improved/modified in previous publications. Many other models have been developed and introduced in literature, which can be much more advanced than that. It is necessary to highlight some of the models introduced in literature and argue for why you selected just this model, i.e. motivations and argumentation for the selection are very necessary.
> Motivation and argumentation for the proposed model have been added to the introduction of Section 4.

The generic model of the PdM suggested in the paper misses important parts that are; visualizing/showing its interoperability in real application(s) and highlighting data input, output, processes and results; reasoning and discussion of the results and model applicability. What is introduce is planned of how the project will be conducted, which is not enough.
> Case-studies and large scale field demonstrators will be used to evaluate the applicability of the proposed PdM model, this is mentioned in Section 5.1. The project is still in a premature state, hence it is not possible to provide detailed descriptions of input, out, processes and results.

The methodology introduced (innovative solutions are developed together with stakeholders). It means we cannot realize any results/solutions before the end of the project. This makes the manuscript very weak because it is not easy for any reader to realize properly; how the treatment of the challenges is done in reality, how realistic were the assumptions, whether the suggested solutions will work, what additional challenges are faced, applicability and soundness of the solutions, etc
> The reviewer is right, there are no results yet. Still it is very valuable to publish the envisioned research approach in a peer-reviewed journal to ascertain the validity of the approach. Of course, actual results regarding the treatment of challenges in reality, how realistic were the assumptions, whether the suggested solutions will work, what additional challenges are faced, applicability and soundness of the solutions, etc will be published during the duration of the project.

In 4 and 5 the authors introduce a roadmap showing how PrimaVera project will possibly handle the challenges introduced in 2 and 3.
Therefore, the conclusions are mainly a summary of what has been introduced. The conclusions usually based on results, data gathered methodology, analysis and analysis tools, research methodology, etc.
> This issue has been addressed in the accompanying letter at the top of the rebuttal.

Round 3

Reviewer 1 Report

1. Thanks for your reply
2. I agree with what you mentioned in your reply´´structuring the challenges and obstacles and relating these to existing insights in scientific literature, as we did in this paper, is valuable for other researchers in the field, even though the specific solutions to these challenges cannot be reported yet`` is interesting and valuable. In this case, you may need to focus on just this part and provide a conceptual model describing how PdM will work and its potential results associated with reasonable discussions, argumentations and motivations.
3. The methodology makes the manuscript very weak because it is not easy for any reader to realize properly; how the treatment of the challenges is done in reality, how realistic were the assumptions, whether the suggested solutions will work, what additional challenges are faced, applicability and soundness of the solutions, etc. This methodology maybe suitable to report with the results when the project is conducted.
4. Please remember that the generic model of the PdM suggested in the paper misses the most important parts that are; visualizing/showing its interoperability in real (or at least assumed) application(s) and highlighting data input, output, processes, application possible/real challenges, solutions( or suggested solutions) and (potential or real) results, result implications; reasoning and discussion of the results and model applicability. What is introduce is planned of how the project will be conducted.
   I understand that the project is still in a premature state and case studies and demonstrations will be done later to evaluate the applicability of the proposed PdM model. But this will not prevent you from developing a conceptual PdM model; describing how to meet challenges, obstacles, and specify (real or potential) results/output, implications, why, all should be associated with reasonable motivation and argumentations. Describing a hypothesis as a platform for conducting the solution and developing PdM suitable to the context you are dealing with, for example (Development of a conceptual model for PdM – implementation challenges – suggested solutions to overcome challenges (references, motivation and argumentations) – potential results - application example(s)- real or anticipated implications). The project (Prima Vera) will be conducted to verify your hypothesis and planning.
5. How the authors will conduct the project is interesting for the funding organisation(s). The text in the Sections of Chapter 4 are OK, but they should be accommodated to a scientific paper and not for advertising Prima Vera. In a scientific paper the author does not need to promise doing activities in the coming project, but they needed to describe, motivate and argue for the solutions suggested as well as showing application example(s) to clarify how it is described theoretically.
6. In 4. Many other models have been developed and introduced in literature, which can be much more advanced than the one you emended. It is necessary to highlight some of the models introduced in literature and argue for why you selected just this model, i.e. motivations and argumentation for the selection are very necessary, which I could not see!
7. In 4.1.1., line 377; you mentioned that several methods were evaluated, but there is no result reported and nothing about why you just focused on these four methods to get more failure data, and what are the implications of these methods.
8. Unfortunately, the text in 5.1.1 is almost relevant for a proposal.

Author Response

Dear reviewer,

First of all, thank you for your useful feedback, it has significantly improved the manuscript.

The last status of the manuscript was tagged as 'minor revision', though in our opinion the comments do not reflect a minor revision. Accommodating the comments would mean a major overhaul of the manuscript, therefore we have asked the editor to look into this disagreement and to make the final decision regarding publication. Points 6-8 have been addressed, please find them below.

Kind regards,

Bram Ton

-------------------------------------------------

2 I agree with what you mentioned in your reply´´structuring the challenges and obstacles and relating these to existing insights in scientific literature, as we did in this paper, is valuable for other researchers in the field, even though the specific solutions to these challenges cannot be reported yet`` is interesting and valuable. In this case, you may need to focus on just this part and provide a conceptual model describing how PdM will work and its potential results associated with reasonable discussions, argumentations and motivations.

3 The methodology makes the manuscript very weak because it is not easy for any reader to realize properly; how the treatment of the challenges is done in reality, how realistic were the assumptions, whether the suggested solutions will work, what additional challenges are faced, applicability and soundness of the solutions, etc. This methodology maybe suitable to report with the results when the project is conducted.

4 Please remember that the generic model of the PdM suggested in the paper misses the most important parts that are; visualizing/showing its interoperability in real (or at least assumed) application(s) and highlighting data input, output, processes, application possible/real challenges, solutions( or suggested solutions) and (potential or real) results, result implications; reasoning and discussion of the results and model applicability. What is introduce is planned of how the project will be conducted.

5 I understand that the project is still in a premature state and case studies and demonstrations will be done later to evaluate the applicability of the proposed PdM model. But this will not prevent you from developing a conceptual PdM model; describing how to meet challenges, obstacles, and specify (real or potential) results/output, implications, why, all should be associated with reasonable motivation and argumentations. Describing a hypothesis as a platform for conducting the solution and developing PdM suitable to the context you are dealing with, for example (Development of a conceptual model for PdM – implementation challenges – suggested solutions to overcome challenges (references, motivation and argumentations) – potential results - application example(s)- real or anticipated implications). The project (Prima Vera) will be conducted to verify your hypothesis and planning.
How the authors will conduct the project is interesting for the funding organisation(s). The text in the Sections of Chapter 4 are OK, but they should be accommodated to a scientific paper and not for advertising Prima Vera. In a scientific paper the author does not need to promise doing activities in the coming project, but they needed to describe, motivate and argue for the solutions suggested as well as showing application example(s) to clarify how it is described theoretically.

6 In 4. Many other models have been developed and introduced in literature, which can be much more advanced than the one you emended. It is necessary to highlight some of the models introduced in literature and argue for why you selected just this model, i.e. motivations and argumentation for the selection are very necessary, which I could not see!
> Lines 329-338 provide motivations for the model. Motivations are provided for it's generalisability, the break down of the data processing step into diagnostics and prognostics, and the addition of human factors.

7 In 4.1.1., line 377; you mentioned that several methods were evaluated, but there is no result reported and nothing about why you just focused on these four methods to get more failure data, and what are the implications of these methods.
> The tense of the sentence has been altered to indicate the work lies in the future. "more failure data are evaluated" -> "more failure data will be evaluated"

8 Unfortunately, the text in 5.1.1 is almost relevant for a proposal.
> This section has been added based on suggestions from the other reviewers after the first round.