Lab Networks in Engineering Education: A Proposed Structure for Organizing Information

Experimentation plays an essential role in engineering education, allowing to balance theoretical proofs and emphasis on physical intuition. Laboratories can fulfil several goals at once, but they also involve high costs, mostly due to equipment, space, and human resources for operating and maintaining them. Remote-access labs have been proposed as a feasible alternative: developed since the early 2000s by an ever-increasing research community, they are real or virtual labs accessible at distance through a computer network. Recently, alternative bibliometric taxonomies and classifications of current networked remote-access labs have been proposed. Yet, none of these works proposes a comprehensive structure to collect and organize the information, especially from a technical perspective, aiming at the definition of the state of the art and future outlooks of provided solutions. In the present work, we fill this gap extending previous works by enlarging their set of criteria towards a general multi-layer model for networked remote-access labs. We performed a systematic review of relevant literature to retrieve useful information to design the structure and then validated it by using a mini-Delphi method.


Introduction
Experimentation and laboratories play a main role in technical education, since they allow to achieve pedagogical objectives, such as learning to manipulate the physical environment and understanding its constraints, by applying theory to practice [1]. According to Antsaklis et al. [2], lab experimentation fulfils the following goals: • Comparing theoretical results with real world results, thus validating the theory and justifying analytic concepts. • Introducing real world factors, which are usually disregarded in analytical solutions, e.g. saturation, noise, uncertainty. • Widening students' focus on secondary design issues, e.g. hardware selection and implementation, and economic considerations. • Providing students with the opportunity to experience professional techniques and practice, e.g. maintaining engineering notebooks, writing reports, brainstorming, team building and problem solving.
Especially in Science, Technology, Engineering and Mathematics (STEM) education, laboratories are used to combine formal mathematical learning with practical experience. Indeed, they make it possible to develop simulative models (based on theoretical principles) and to compare them with real equipment and devices: this balance between analytical solutions and experimental evidences is considered a major challenge in control education [3]. Although it is highly recommended to provide a learning path that includes learning through an 'actual experience', hands-on labs, which are the traditional solution, involve high costs relating to equipment, space, and maintenance staff [4]. For this reason, the interest of the research community in 'non-traditional labs' has been growing steadily since 2008 [5]. Although in Section 2 we provide a brief discussion on the vocabulary used in the literature on labs and experimentations, we anticipate that, in this work, the term 'non-traditional labs' is used to underline the difference from the traditional hand-on ones [6].
Many authors have analysed the evolution of non-traditional labs to provide insights of their continuously-improving learning outcomes, which are nowadays often proved to be comparable to those of traditional labs [6] - [8]. Other authors even claim that non-traditional labs can broaden the learning outcomes, since they provide insight into experimentation aspects otherwise indiscernible (e.g. heat transfer, electromagnetism), and they allow to interact with equipment parts otherwise covered or protected, e.g. robot mechanisms [9], [10]. Gravier et al. [11] identified additional advantages of nontraditional labs, such as: Laboratory safety, Availability, Accessibility, Experimentation observability.
Potkonjak et al. [12] focused on the possibility of conducting the experiments by several roles and involvement levels (i.e. multiple access, flexibility and change in the system configuration), and on equipment's preservation. However, the literature also provides some disadvantages of non-traditional labs, as virtualisation requires a highcomputing capacity and may negatively impact on the users' attitude towards experiments. Relatively to the last issue, virtual experiments could be performed with little care, as in a video-game experience, and this may create a lack of a seriousness to approach the experiments [12]. These limitations often suggest providing also an 'actual experience', combined with the non-traditional one.
Since the early 2000s, the number of studies on non-traditional labs has grown amazingly [5]. Zappatore et al. [13] state that, in this research area, two common threads exist: (i) The rising pedagogical interest in lab-based education. (ii) Technology advancements in lab connection and networking.
These criteria will be discussed more in-depth in Section 3. Here, we just want to stress the differences between the approach by Dormido Bencomo [19] and the one by Potkonjak et al. [12]. As it is shown in Table 1, where we try to match the four criteria proposed in the two studies, the approach of Dormido Bencomo [19] seems to be more focused on the 'location' and on the 'features' of the software module behind the virtual system, whereas the one by Potkonjak et al. [12] is more oriented on its 'performance on the user side'. Perhaps this equivalence is rather stiff (e.g. where the 'nature of the simulation kernel' is matched with 'visualisation'), nonetheless it should be of help for a preliminary understanding of the main criteria that are used to class non-traditional labs, and of the trend of related researches, since twelve years passed between the two works.
Although the above-mentioned approaches are very useful, we believe that both Dormido Bencomo [19] and Potkonjak et al. [12] addressed the problem from a qualitative point of view, without a real technical focus, especially concerning lab networking development of both remote and virtual labs. This issue can be quite important for those who may want to pursue the goal, reported in several studies, of framing a network of networks-of-labs, which may become a need, more than an alternative, for non-traditional labs [21]. In the present study, we extend the previously cited works providing a set of assessment criteria based on a broad multi-layer model for a network of labs, and we propose a comprehensive structure to collect and organize general and, most of all, technical information on LNIs. Thus, the aim of this work is to provide the scientific community with a tool that can be used to depict from a, mostly, technical point of view the state of the art of LNIs. It is opinion of the authors that such a structure can provide great value to organizations that shall build a new LNI and, more in general, it can improve the knowledge of the scientific community on technologies and resources available within different LNIs. Table 1. Difference among the models for virtual systems (VS) from Dormido Bencomo [19] and Potkonjak et al. [12] Authors Dormido Bencomo [19] Potkonjak et al. [ User interface Dormido Bencomo [19] Potkonjak et al. [12] The remainder of the paper is organized as follows. In Section 2 we propose an overview of exiting literature on remote-access labs and LNIs. In Section 3 we focus on a recent attempt of classifying information on LNIs and we illustrate its merit and limitations. In Section 4 we propose a structure to organize relevant information on LNIs, and we provide an example on how this information must be filled in. Finally, we provide discussion and conclusions in Section 5.

A Brief Analysis of Existing Literature
The debate on non-traditional labs was first introduced by the National Science Foundation and the Control Systems Society at a workshop on 'New Directions in Control Engineering Education', held at the University of Illinois at Urbana-Champaign twenty years ago [3]. After the study of Antsaklis et al. [2], the research community has produced a lot of literature on this topic. Just to have an idea, Heradio et al. [5] counted more than 1,000 works published in the three-years period spanning from 2012 to 2014. In the early 2000s, the scientific debate focused, mainly, on the following topics: (i) The necessity to classify and characterize the laboratories (ii) The definition of a common vocabulary (iii) The further developments needed to ensure the use of non-traditional laboratories in STEM education [22].
Regarding the first issue, related to the need of a classification of the labs, the first attempt to define a proper taxonomy belongs to Dormido Bencomo [19], who proposed a two-dimensional classification that uses the 'access to the resource' and the 'nature of the resource' as classification criteria. Specifically, he defined four cases by combining local or remote access to the resource, with its real or simulated nature: • Local access-real resource • Local access-simulated resource • Remote access-real resource • Remote access-simulated resource A further attempt is the one by Zutin et al. [23], who discussed 'the creation of a common framework to describe laboratories according to the semantic web technology'. They identified four cases by considering the utilization of equipment and devices rather than the access to the resources. Pfeiffer and Uckelmann [24] further extended the possible scenarios by introducing the concept of network (see Figure 1), which has become a trendy topic in recent literature. In the remainder of the paper we will stick to the classification of labs presented by Zutin et al. [23].
Regarding the second point, related to the lexicon setting, a common vocabulary of non-traditional laboratories is still missing. Although progress has been made and the term 'online lab' is generally accepted as the standard [25], [26], other terms such as 'virtual and remote labs' [3], [27], and 'non-traditional labs' [6] are frequently used in relevant literature, to identify both remote and virtual labs. Since we believe that the 'access method' is more appropriate than the 'type of performed experimentation' to distinguishing between different labs, as in the classification provided by Zutin et al. [23], we will use the term remote-access laboratories to identify any remote-experimentation environments, either virtual or real. Regarding the third topic, recent attempts have contributed significantly to the selection of worthwhile literature, and to its mapping, with the aim of identifying specific research topics [5], [13]. Beyond proposing a normalized dataset of bibliographic references in lab-based education, Zappatore et al. [13] proposed a structured data processing framework for producing it. Their work is very useful, not only for the comprehensive dataset that it has made available, but also because it provides the research community with an approach to define which are the relevant features in lab-based education. Heradio et al. [5] have instead performed a science mapping of the lab-based education research community, by means of co-word and co-citation analysis. Through this approach, they provided the readers with useful information concerning the linkages between research themes and their evolution in a twenty-two-year timespan, trying to address the future outlooks for lab-based education and research. In particular, the authors indicated the following emerging needs: (i) An efficient combination of virtual and remote labs (ii) The development of increasingly collaborative learning environments (iii) The assessment of the pedagogical effectiveness of remote-access learning concepts.  [23] and extended by Pfeiffer and Uckelmann [24] A further interesting contribution to the research on remote-access labs comes from the experience of research projects on laboratories and LNIs, such as the aforementioned works by Harward et al. [18], Richter et al. [16], and Halimi et al. [17], relating to the iLab Shared Architecture project (http://ilab.mit.edu/wiki), the Library of Labs -LiLa project (http://library-of-labs.org), and the Go-Lab project (https://www.golabz.eu/), respectively. In this context, i.e. projects that both promote and are the object of the research, we can also cite: • The WebLab-Deusto project (https://weblab.deusto.es/), which has provided an interesting contribution to research on remote-access labs, with an holistic approach to the development and implementation of labs and LNIs [21], [28].  [29]. • The University Network of Interactive Labs (UNILabs, https://unilabs.dia.uned.es/), a network of several Spanish and South-American universities that share their laboratory resources for education purposes [30], in a web 2.0 environment, i.e. the "dynamic web", whose goal is to allow users to read, write and collaborate on the topics they are addressing [31]. • The OpenScience Laboratory (https://learn5.open.ac.uk/course/view.php?id=2), of the Open University in the UK, which has been very used in experiment-based research activities [32], [33]. in Sweden together with National Instruments in USA and Axiom EduTech in Sweden, developed hands-on, virtual, and real-remote laboratories in Electrical and Electronics Engineering. The aim is to promote new teaching and learning methodologies for engineering education [7] shared by a network of academic partners that nowadays also includes eleven institutions from Spanish, Portuguese and Latin American areas [34].
• The Platform Integration of Laboratories based on the Architecture of visiR (PILAR, http://www.ieec.uned.es/pilar-project/index.html?lng=en), funded by the EU program Erasmus+, with the aim to interconnect all VISIR systems in a superstructure, providing a well-defined set of open practices, available through the Internet as a set of services [35].
Heradio, de la Torre and Dormido [3] stated that, although the research projects on laboratories and LNIs have provided an outstanding contribution to the research community, there are few reviews that analyse and classify them. Starting from these premises, they provided a survey of nine research projects and of the related forty remoteaccess labs active in 2016; all labs were fully characterized in terms of access type and experimentation availability. Conversely, Potkonjak et al. [12] followed a different approach to perform a qualitative analysis, with a particular focus on robotic virtual labs, of an interesting set of remote-access labs that were selected among twenty research projects funded by national and international frameworks (e.g. LiLa and Go-Lab), learning platforms provided by single Universities (e.g. Virtual laboratory of Process Control) [36], and commercial products (e.g. Robologix by Logic Design Inc -LDI Canada).
By analysing all the twenty-nine projects of these two reviews, excluding those ones focusing on software development for robotics' virtual labs (as can be deduced from the work of and Potkonjak et al. [12]), it is possible to identify a sub-set of fifteen suitable research projects, eight of which, more than the 50% of the sample, relate to LNIs or network of networks. This result highlights the key role played by public-funded research projects, in supporting the development of remote-access laboratories. Considering the gap in analysing and classifying the project experiences, as pointed out by [3], we argue that it is useful to perform further analysis to introduce suitable indicators to describe the context and the technical aspects of the LNIs.

3
The Model Proposed by Potkonjak et al. [12] In their classification, Potkonjak et al. [12] make use of four evaluation criteria, that we report below in a slightly modified version, which is more in line with our own classification (original text from [12] is reported in quotes).
• C1 -Realistic User Interface: 'The user interfaces, for each piece of equipment, must be identical to the corresponding real devices.' Meeting this requirement is very important, especially for the labs used for operators' training. It mainly concerns programming, and it is relatively simple to be achieved. Possible criticalities could arise in case of device copyright to replicate. • C2 -Realistic Systems Replication: 'The behavior of the virtual system (e.g. state and control variables) must be equivalent to the system behavior in the physical paradigm.' This requirement concerns the kinematics and the dynamics of the virtualised equipment. To accurately replicate the behaviour of the physical system, a mathematical model is generally needed. Its formulation depends on the teaching field of the system and it can be made using either general purpose or dedicated simulative models. • C3 -Realistic Graphical Representation: 'Visualization must be provided that makes students feel like they are looking at a real authentic thing.' Meeting this criterion is fundamental to experimentation labs reproducing complex systems (e.g. manufacturing cells, logistic system, etc.) whose elements move synchronously in space, can interact, collide and interfere with each other. In case of static and/or 2D systems, achieving this requirement is less important. • C4 -Support for Communication and Collaboration: 'A 3D laboratory space must be created which allows for communication and collaboration among students and with the lab supervisor (or expert in the field).' Probably, this is the most innovative criterion introduced by the authors. To meet this requirement, developers should create a virtual world to support not only the planning and the implementation of the experimentations, but also the overall learning environment, to which the experimentation activities belong to. In addition to the experts and/or the developers, also basic user (e.g. students) should be allowed to improve the learning process, by operating in a Virtual Interactive 3D Environments (VIDE). Students, as well as supervisors, may also choose an avatar in a VIDE [37], in perfect analogy with a serious game.
Potkonjak et al. applied these criteria to map the labs of the twenty initiatives onto requirements' meeting, and they found that only the Virtual laboratory at Stevens Institute of Technology [25], [38] fully complies to all four criteria. It is also worth noting that, since only nine, out of twenty, were robotics' labs, the authors made it clear that the four criteria can be used also for labs that do not relate to the robotics' field, and that their aim was to propose a general model suitable for all web-based learning services. In details, the model entails: (i) A lower level of learning for system operator (ii) A higher level of education towards the design and development of applications (iii) The use of platforms for inquiry learning, towards openness and interoperability through traditional LMS and new 3D virtual world platforms ( [12]).
The purpose of designing a model in such a way is twofold, as it allows: (i) Dealing with new technologies and services, which have not yet been really considered in education, e.g. equipment for immersive environments and virtual worlds for experiments design and implementation. (ii) Highlighting the weaker aspects of the existing virtual and remote laboratories, i.e. criteria C2 and C4 related to dynamic modelling and virtual spaces reproduction and, more in general, a higher level of flexibility to create new objects to be included in the experiment.
These two aspects are particularly relevant if the experiment relies on virtual and remote labs developed in a mixed-reality environment.
As noted in the introduction of the present study, albeit suitable, the model by Potkonjak et al. has some limitations: (i) The lack of a structured general contextualization of the labs and lab-networks, i.e. labs' scope is disregarded (ii) The lack of a common reference-framework addressing the selection of the criteria, i.e. the classification of labs and lab-networks is more qualitative than quantitative (iii) The lack of a set of criteria for framing a network of networks of labs.
Owing to these issues, the goal of the present study is to expand the four criteria of the model by Potkonjak et al., using a suitable layered reference-framework for remoteaccess labs. The proposed structure is detailed in Section 4.

The Proposed Framework to Collect and Organize Information on LNIs
In this section we provide the structure to collect material on LNIs and to organize them into a display-board. Also, to prove the applicability of our structure, and to provide the reader with a practical example of how many fields can be filled in when information is collected on a specific LNI, we introduce a descriptive case study composed of three LNIs, which is reported in Appendix A. These examples show that, although most of the LNIs will not fill every field of the database, the database itself is complete, in the opinion of the authors, as it allows to gather all information that is relevant to describe the specific solution.
Concerning the structure, we have first defined three dimensions to cluster the suitable information. They are: • 'General Information' -This dimension is related to the organizational aspects of the project. Indeed, it organizes data relative to the stakeholders of the lab (network), to the project's duration and activities. It also reports all the sources where related information can be found. • 'Context Information' -This dimension is mostly related to the didactic and thematic aspects of the labs, as it deals with general labs' characteristics and availability. It also details the type of experiments (that can be performed) and the kind of users that can log on the lab.
• 'Technical Information' -This dimension specifically addressed the technical details of the implemented solutions. We note that, in this dimension, the technical point of view might also be directed on 'non-technical' aspects of the LNI, such as didactic (e.g. learning analytics and learner record store) or organisational aspects (functionalities of the client layer and access levels).
Concerning the last point, we are aware that, in case of a network of labs, collecting technical details for them all and synthesizing them in a single data collection framework is a hard and time-consuming task. Still, since this information can be very valuable, we propose to proceed as follows. When answering to each question, the analyst should consider the best performing lab with respect to the specific question. Should the number of labs be too big to be handled, labs could be clustered in groups with similar characteristics and technologies. Thus, the best performing lab of each cluster could be more easily taken as a reference. This process will then generate a pseudo-lab which synthesizes the best characteristics of the whole LNI. This approach is aligned with our aim to describe the state-of-the-art in terms of devices, techniques and services [39]. We are aware that this approach cannot entail information about learning objectives for which the labs were designed, as those learning objectives can differ among the labs and cannot be reported at our pseudo-lab level. Nevertheless, our intent is more focused on the technical aspects and on the lab architecture, rather than on teaching and learning objectives.
Next, to detail the three dimensions, we defined a set of Key Attributes (KAs), whose nature and number depend on the dimension itself. The KAs are information that categorize the laboratory, and they are described through a series of items, structured in a multi-level hierarchical way, that provide low-level information (relative to the implemented solution), either of quantitative or qualitative nature. Both KAs and items belonging to the three dimensions were selected through an accurate analysis of relevant literature on remote-access labs and were validated by the cross-functional team of the DigiLab4U project, comprehending seventeen subject matter experts from five German and Italian higher education institutions (for the list of institutions, see www.digi-lab4u.com). Specifically, the validation process was performed by means of a mini-Delphi strategy as the one suggested in the work by Neaga and Henshow [40]. The full list of KAs and their relative items is shown in Table 2, where the structure is applied to a descriptive case study, and in Table 3, where the exact meaning of each item is reported.
The first dimension 'General Information' (G) is detailed through ten KAs related either to the organizational aspects of the project or to the sources of information related to the project. Note that, being a descriptive dimension, G is directly described through KAs, and it does not need lower-level items.
The second dimension 'Context Information' (C) comprises seven KAs related to the network (if the project involves a network), to the type of experimentation and to the access to the labs. Concerning lab typology, the detailed information is gathered by means of four first-level items, namely: 'Virtual', 'Remote', 'Hybrid', and 'Gaming'. Note that the first three items correspond to the 'remote-experiment' part of the lab classification proposed by Zutin et al. [23], but there is not a specific item concerning the 'Networked' scenarios suggested by Pfeiffer and Uckelmann [24]. This is because we believe that the network type and its architecture are the core of a remote-access labs; hence we dedicated to the network description a specific KA, within the 'Context Information' dimension. Also, we opted to consider 'Gaming' as a specific item, because it both relates to virtual worlds supporting specific laboratories [12], as well as to Game-Based Learning and to Serious Games. Since gaming-based learning scenarios often make use of computer-based gaming, it is our opinion that they are a real mix of real/virtual and local/remote solutions [41]. Furthermore, in previous classifications, serious games have never been considered. However, the use of this learning-game tool is growing considerably, thus we decided to consider this item separately, because of its novelty and specificity.
Finally, the third-dimension deals with 'Technical Information' (T) and it addresses the technical solutions implemented by the lab-network. The KAs in this dimension are six, and they are described through thirty-eight items. Four KAs, out of the six of this dimension, directly refer to the four-layer reference-framework (for remote-access labs) introduced by Zapata Rivera and Larrondo Petrie [26], of which, a readapted version is shown in Figure 2.
1. The upper layer is the 'Client Layer' that provides the user an interface to re-quest\book a practical session and to interact with remote lab experiments by communicating with the learning environment, or directly with the remote lab server. This layer coincides with the KA 'Client', which comprises six first-level items about the user and the access to the lab-network. The second-level items describe the possible roles of the users. 2. The second layer is the 'Learning Environment Layer', which exists when labs are embedded in an LMS. If so, performed activities can be evaluated as part of an exam or can be used to define a full pedagogic scenario and to assess learning analytics. The corresponding KA is the 'Learning Platform' and it collects three first-level items on the LMS (e.g. Moodle, edX and so on). The second-level items describe its functionalities. 3. The third layer is the 'RemoteLab Server Layer', which gives access to a remote laboratory through three resources: (i) the Access Manager, for validating the identity and authorization rights; (ii) the Scheduler, for the management of the user appointments and validation of the specific experiment, and (iii) the Resource Module, which implements the searching service over the indexed list of resources, using the metadata associated with each resource. The corresponding KA is the 'Remote Lab Server' and it consists of nine first-level items concerning the access to the experimentation, e.g. the Access Manager and the Lab Scheduler, and the experimentation typology, which consists of four second-level items properly related to the storage of the result, e.g. 'batch', 'sensor', 'interactive experiment', and 'repository'. 4. The fourth and last layer is the 'Physical Laboratory Layer', which implements a centralized Administrator Module responsible for the communications with the Re-moteLab Server Layer and for the management of experimentation and devices. It also performs administrative activities (e.g. update new resources, report changes of schedules, etc.) and activates alerts. The corresponding KA is the 'Physical / Virtual lab' whose first-level items allows describing the software and the devices for the experimentation; the latter is described by means of four second-level items. In particular, we specify the difference between the second-level items 'smart sensors' and 'smart devices', to make clear that smart devices are able, self-sufficiently, to process data and to make them available online.

Fig. 2.
The four-layer reference-framework readapted from Zapata Rivera and Larrondo Petrie [26] used to identify the items of the dimension 'Technical Information' It is worth noting the very technical characterization of the proposed structure, as, just the 28% of the items are of descriptive and qualitative nature, whereas the remaining 72% relate to technical features. Furthermore, the structure is deliberately very detailed, and maybe even characterized by an abundance of items. In our opinion, in fact, it should be considered as a checklist to guide the analyst in considering all the relevant issues when he or she classifies the existing remote-access lab, without missing important features. Finding a value for each item of the display-board would be the optimum but, to properly describe a laboratory, there is no need to fill in all its fields, as the amount of information depends on the typology of the lab and on the performed experimentation. In addition, we opted for a very detailed structure because, in this way, it can be used as a guideline also for designers developing novel and innovative remoteaccess labs. In summary, a very detailed structure has a twofold aim: (i) Supporting the provider of new remote-access labs in designing it (ii) Creating a common standard, in order to make the provided solution more visible to the research community.
The structure that we propose, provided with the explanation of each field is provided in Table 2. In the heading of the tables, D are the dimensions and KA are the key attributes.
For the sake of clarity, a part of the 'Technical Information' dimension is also described below. The first two KAs provide general information on the architecture of the lab: they are 'Number of layers of the lab architecture', and 'Full list of layer names'. The other four KAs correspond to the layers of the reference-framework for remoteaccess labs provided by Zapata Rivera and Larrondo Petrie [26]: 'Client', 'Learning Platform', 'Remote Lab Server' and 'Physical / Virtual (experiment)'. The items of the latter KA are: 1. The single-level information about the design software, used to develop the virtual labs. 2. The double-level information about the physical equipment and devices installed in the lab, namely: a) Need of human actions (i.e., activities carried on by lab's staff) to perform the experiment. b) List of devices and equipment used to operate and/or to automate the lab. c) List of the devices used to transmit information, and classification of their nature (i.e. smart devices or smart sensors).
As a further example, the hierarchical structure of the 'Physical / Virtual' KA is shown in the cellular chart of Figure 3. As it can be seen, this KA belongs to the third dimension and it is described by means of two first-level items, highlighted in dark orange. Also, in case of Physical labs, additional information is collected by means of four second-level items, namely: 'Fully automated', 'Sensors, actuator, and controller', 'Smart Devices', and 'Smart Sensors'. 1 Used among others, for retrieving the information filling the database 2 Although useful, this information is difficult to find. This field can be considered as optional As we mentioned before, Table 3 (in Appendix A) reports an application of the structure to three different LNIs, also referred to as projects or 'ecosystems', namely Go-Lab (Global Online Science Labs), LiLa (Library of Labs) and WebLab-Deusto. Go-Lab (https://www.golabz.eu/) was a project co-funded by the European Commission (7 th Framework Programme, Grant agreement No 317601) which has developed the Go-Lab Ecosystem, a complex of services to share and create labs (i.e. the Go-Lab Sharing and Authoring Platform), which targets science teachers from primary and secondary schools and aims to help them enriching their teaching practices with innovative teaching approaches and supportive technical tools. LiLa (http://library-of-labs.org), co-funded by the eContentplus programme of the European Commission, is an initiative of eight universities and three enterprises, for the access and exchange of virtual and remote laboratories, including services like a scheduling and tutoring system, as well as connection to library resources and an environment for online collaboration. Web-Lab-Deusto (https://weblab.deusto.es/), started on early 2000s by testing different settings of a Computer Programmable Logic Devices through the internet, is an initiative of the University of Deusto aiming to increase experiential learning by means of remote and virtual laboratories, freely accessible through the Internet via an Open Source license software. Physical / Virtual (experiment) -belonging to the dimension 'Technical Information', and its two-level items We selected these projects, which significantly differs from each other, because they are well-established learning environments that propose a great number of different laboratories. The information needed to fill Table 3 were collected as follows: • We retrieved information on existing labs and previous initiatives, by reading the relevant literature on the topic. This literature was obtained by querying databases of peer-reviewed literature (e.g. Scopus), and freely accessible web search engine (i.e. Google Scholar). By analysing this literature, we identified suitable labs and networks • We visited the website of the specific lab/network, as well as the website of the related project initiatives. • We retrieved information by reading relevant literature on the specific subject.
• In case of missing data, we directly contacted the lab provider or the authors of relevant papers. • We asked the lab provider to update or amend the information related to its lab, if possible.
Concerning the fourth point of the bullet list, we explicitly noted above that we did not fill each display-board in all its fields. This is because our aim was to propose a data collecting framework and not to perform the analysis of the existing lab-networks, which will be the topic of future works. So, the results of Table 3 must be considered only as a demonstration example of the methodology, which is why the display board was not filled with data which were not readily available, or which would have required a direct use of the laboratories. Still, we note that the missing data average of the three examples is 30%, a fact that confirms how, although somehow redundant, data included in the structure are quite easy to be found.

Discussion and Conclusions
Nowadays, providing laboratory experiences for students in STEM education is extremely important, both in terms of teaching and training reasons. Experimentations, in fact, allow explaining analytical concepts in practice. Moreover, they allow educating pupils in operational methodologies that will otherwise be learned just once they enter the labour market. Since traditional hands-on labs involve high costs (relating to equipment, space, and maintenance staff), and they require full-time personnel devoted to limited tasks, the interest of the scientific community in remote-access labs is continuously growing. Although researches have been originated for cost-savings reasons, a definite answer concerning the actual savings of remote-access labs is still missing, as they generate high costs in the development and in the maintenance phase (e.g. highcomputing devices, software development, reliable equipment and so on).
However, what emerges from the literature is that remote-access labs provide learning outcomes that are comparable to traditional hands-on labs, and they may provide further benefits widely validated by the research community. Despite researchers have produced a huge amount of literature on the topic, there is a gap between the literature proposed to solve specific problems for an individual remote-access lab or network, and the literature proposed to analyse the labs-network from an integrated and comprehensive point of view. This is noteworthy, because the information that can be obtained from past experiences can be useful to address new projects and design new remoteaccess labs and networks. With the aim of supporting both researchers analysing remote-access labs and practitioners who want to develop a new solution, in the present study we have provided a general pattern to retrieve the suitable information on labs and networks, by defining how to collect information and organize data in a suitable structure. Specifically, we have defined which information are required to fulfil the definition of the state-of-the-art in the remote-access lab-based education, and to support practitioners in the development of their own solution. We have distinguished between three information dimensions: the 'General Information' and the 'Context Information', which define the lab/network 'Identity Card', and the 'Technical Information', which sounds like the laboratory 'Specification sheet'. In particular, we believe that 'Technical Information' constitutes the real novelty of the paper, since previous studies have never provided technical information on the proposed solutions in a structured and comprehensive way. The proposed data collecting structure has also been reported in a practical display-board, where all relevant information of three different LNIs, namely Go-Lab, LiLa, and WebLab-Deusto, were filled in. To conclude, we note that such structure could be used to collect and organise information on existing LNIs (as well as past ones), with the aim of depicting the state-of-the-art on this topic. Also, the very same structure could be useful as a guideline for new projects and application of LNIs, as it promotes a technical approach to a wide set of LNIs, without disregarding organisational and didactical aspects. Indeed, the authors of these paper are already working on those topics for future research. 6 his research results. He has published over 60 research papers in qualified international journals, that were cited more than 1.500 times. His h-index is 16. Some of his works have been awarded international prizes by the scientific community. His research activity produced over 60 papers indexed in Scopus (plus 30 non-Scopus indexed ones). The main topics of the research concern (i) development and testing of artificial intelligence algorithms for the optimization of logistic/production processes; (ii) development and validation of systems for the automation of logistic / production processes through the use of Auto Id technologies (such as RFID); (iii) process optimization through lean management principles (iv) project management and (v) supply chain management. Dieter