Review Comment:
The paper has significantly improved with respect to the original submission. The authors have sufficiently answered the main comments of the reviewers and refined the paper accordingly.
In general, the paper is well-written and technically sound. It represents an interesting contribution about the use of semantic technologies to increase the FAIRness and machine-actionability of M&S capabilities. However, the novel contribution of the work is mainly technical. I would suggest the authors to better emphasize methodological aspects in order to better emphasize the scientific contribution of the work.
One of the main questions addressed by the authors about "which resource to expose" (sec. 4.2.2) has not been sufficiently addressed in my opinion. Namely, the authors should better explain the methodology that would guide a user to describe and expose an M&S resource and thus define an instance of a model. In this regard, it is not clear the role of the ontology and the detail of representation supported the considered model. The examples provided by the authors are too general and simple to fully capture the expressivity level of the proposed approach. Especially for readers that are not familiar with FMU/FMI, it would describe the main concepts and properties that would be used to characterize M&S instances (i.e., the TBox) and propose a methodology about how concretely build the resulting knowledge graphs.
Below, there are some comments the authors may find useful to improve the quality of the work and better emphasize some central aspects of the contribution.
----
I found the emphasis on "intelligent software agent" somehow contradicting the answers of the reviewers about PPA. This assumes a kind of autonomy in the composition FMI-modeled entities that does not seem a "requirement" of the envisaged integration. In this regard, their motivation of not discussing HTN or more in general automated planning is not totally satisfactory.
The PPA is listed as one of the conbtributions of the work and the functionalities supported by PPA are really similar to those supported by and automated planner. The authors should therefore discuss a bit the distinguishing features of PPA and motivate its use with respect to existing planning frameworks that already support the verification of the desired "causality chain".
----
A description of the fmi2rdf package would be useful to better evaluate the contribution.
Although the mapping seems rather sintactic the authors should better describe how the representation is automatically generated. In the current form, it is not possible to understand the kind of inference necessary and the concepts/properties that would be created to properly describe a FMU into the ontological model. Considering that not all readers are experts of the specific formalism, a description of the procedure would facilitate understanding of the managed information.
----
How do they support reification of triples? Do they use patterns e.g., "Singleton" or other approaches? The discussion of section 4.2.2. does not sufficiently describe the model and the methodology used to represent information and support FAIRness. It is not clear if the approach concretely supports provenance and which set of information they concretely consider. Do they integrate (or plan to integrate) PROV-O ontology or other similar models? Could they provide additional information about the model used/defined to represent information?
Section 4.2.2. does not sufficiently answer to the authors' question "which triples should the resource representation contain?" I would suggest the authors to further specify the model, and the methdology proposed to reprsent resources. It is not clear for example the kind of information that should be specified to "publish" a resource.
----
Concerning PPA, could the authors provide some more detailed and concrete examples? The current description is rather general. It is not clear for example what goals and states actually are and how are they represented.
Which are the conditions to achieve a goal correctly? Do services have dependencies within a "procedure"? Can a service need the outcome or data produced by other services as input in order to work correctly? Can they specify conditions on the results so that validate the input/output of the data exchanged between services?
Can it be possible to specify constraints about the expected "computation times" of the invoked resources? Can the PPA also evaluate the possibility of parallelizing the computation and thus the invocation of the different services? Information about duration and temporal constraints could for example expressed and evaluated using temporal planning formalisms.
Does the RESTdesc support all this information?
----
Examples are rather general? Which constructs do exist to specify types of models and related properties? The listing 3 for example is very general and can be applied to any instance of any model. I guess applications would need more detailed information about models and computation outcomes in order to compose the correct instances. More detailed description of the level of details supported by the proposed representation is necessary.
---
Concerning the example at the beginning of sec. 5.2. Do the ontology contains a specific class for each possible type of model? e.g., a class to explicitly represent PV systems? Otherwise, are there a set of properties concerning e.g., the types of data and of a model a user can take into account to "discover" model instances that are useful to her/his own objectives?
| 
