The use of AI and data-driven technologies and infrastructures for innovation and development of advanced research and industrial applications requires a strong degree of integration across a broad range of tools, disciplines and competences. In spite of a huge disruptive potential, the role of AI for research and development in the context of industrial applications is often hampered by the lack of consolidated and shared practices for transforming domain-specific processes for generating knowledge into added value. These issues are particularly striking for small-medium enterprises (SMEs), which must adopt clear and efective policies for implementing successful technology transfer paths for innovation. The activities of the DAIMON Lab of the CNR-ISMN focus on the design, development, implementation and application of integrated modelling, data-driven and AI methods and infrastructures for innovation in hi-tech applications. Our approach is based on the development of horizontal platforms, which can be applied to a broad range of vertical use-cases. Namely, we target the realisation of high-throughput workflows, related to specific domains and use cases, which are able to collect and process simulations and/or physical data and information. The implementation of an interoperable integration framework is a prerequisite for further application of AI tools for predictivity and automation. With a strong focus on the development of key enabling technologies (KETs), such as advanced materials, the approach pursued is extended to a broad range of application fields and scenarios of interest in industry, including electronic and ICT, advanced and sustainable manufacturing, energy, mobility.
relies on a trial-and-error approach to innovation and improvement, which can be time-consuming, expensive, and inefective. Machine learning and AI have the potential to revolutionize problem-solving and decisionmaking in research and, in particular, for innovation in Industry 4.0. Accordingly, recent years have witnessed the increasing role of machine learning and AI in shaping the future of innovation in high added-value applications, in manufacturing and in industry. The lack of systematic improvement in several research and application fields is a major challenge that can be addressed by the integration of knowledge and data and the automation of the innovation process. The impact of data-driven integration technologies can potentially afect multiple value chains and fields of interest in industry applications and related domains, including but not limited to advanced materials and manufacturing, electronics, mobility, environment, and more.
The realisation of data-driven workflows is critical for
generating integrated knowledge that can support AI
and data-driven methods for prediction and automation.
By implementing data-driven workflows, R&D processes
and activities can be optimised, identifying areas for
improvement and innovation, and promoting collaboration
and knowledge sharing across diferent departments and
areas of expertise. However, the implementation of
eficient data-integration frameworks in several application
domains is still hampered by a manifold of theoretical
and technical issues. Indeed, the uptake of digital
technologies for innovation requires to face the challenges
related to the integration of data-driven techniques with
consolidated processes in specific application and
industrial domains. The main dificulties are related to the need
innovation still lack shared technical procedures,
knowledge and standardisation in the adoption of digital and
data-driven technologies. The role of fully-digital
approaches, from modelling and simulation to AI, must
therefore be consolidated for an eficient link to
specific value chains. Several recent research eforts tried
to address the challenges of data-driven integration for
domain-specific applications. For example, Barbella et
al. [
One of the major challenges for the adoption of
digital technologies for research and innovation is the
realisation and implementation of automated data-centric
workflows. This challenge is further complicated by the
number of potential sources and the increasing volume
of available data, generated by research activities or by
specific R&D high-throughput analysis tools.
Automating the process of data generation and providing native
support for data integration and elaboration has
therefore become a crucial focus area for both academic and
industrial research. In particular, physical modelling and
simulation tools have proved their potential in a large
number of application fields [
One of the most critical steps for integration is therefore related to the processes involved in the acquisition of data. An eficient integration strategy, interfacing with sources generating data, is essential for the development of a data-driven platform for decision making. Moreover, integration is required also at the knowledge level, which is also defined by the application domains, thus addressing the issue of interoperability at diferent levels.
In analogy with previous eforts [
Namely, AI other data-driven tools (expert systems, decision support systems, etc.) can operate on data integration frameworks as services, providing answers to technological queries in terms of predictions, analysis and process automation (see Fig. 2).
One of the key ingredients for the multi-level inte- Figure 2: Data Integration Framework for Predictive Analytics gration of data and knowledge relies on the adoption of and Process Automation. The diagram illustrates the seamless eficient methodological frameworks and technologies integration of semantic queries, external data sources, and adfor knowledge representation and syntactic and seman- vanced workflows to generate AI-powered predictive services tic interoperability [10], which can be considered as a and cloud-native solutions, enabling eficient and efective key enabler for integration. In particular, ontologies can analysis and automation of complex technological queries. support and enable semantic interoperability between diferent systems and applications within or across application domains, by providing a shared understanding of ular, an increase of the overall throughput of integration concepts and relationships and supporting the formal en- frameworks can be associated to the improvements in coding of semantic meanings [11]. The formalisation of GPU computing technologies and to the development of the intrinsic structure and patterns of a domain of knowl- large-scale data centers. As we will discuss in the next edge as a mean for creating encoding of properties and section, the integration approach based on multiple interthe relationships between diferent concepts constitutes connected abstraction levels and corresponding software an invaluable tool for gathering new knowledge from stacks, and the strong connection with the knowledge historical data, improving current processes and data and competences of specific application domains are key collection procedures [12]. For these reasons, semantic to develop solutions addressing research and innovation platforms can therefore be a key element of unification challenges. in the research and development space, when backed by the cooperation among teams of researchers, and make existing research lines more interoperable and shareable 3. Applications to AI for industry: [13]. advanced materials, Another key enabling technology for the implementation of integrated data-driven frameworks is the application manufacturing, electronics, of high-performance computing (HPC). The implemen- energy, smart mobility tation of integration strategies on high-performance infrastructures can support innovation by providing the The general approach to the development of data-driven required power to sustain large volumes of data in high- and automated workflows and frameworks has recently throughput acquisition processes and analysis. In particbeen applied by our research team to a manifold of difer- systems, which are thus suitable for an image-like repent activities. The common trait of this paradigm consists resentation of structural features and patterns, as done in the integration between a horizontal layer of tools, recently in similar work [24]. In particular, we used commethods and technologies to vertical use-cases. As stated putational data [23, 25] to train a deep learning model above, the development of several application fields de- by encoding structural information on graphene sampends crucially on innovation in a manifold of critical ples into a standard image format. The application of technologies, involving phenomena at diferent levels convolutional neural networks (CNNs) to the encoded and scales, from the very basic constituent of the phys- information exhibits remarkable accuracy in predicting ical world to processes and products impacting on spe- the physical properties of graphene samples (average cific socio-economic sectors. Our work aims at integrat- error below 4%) with a gain of several orders of magniing these diferent levels through digital and data-driven tude in computational time with respect to calculations strategies, interconnecting competences and knowledge based on physical models. Remarkably, the use of image (see Fig. 3). One of the most strategic KETs for indus- encodings also outperforms standard machine learning try and manufacturing is that of advanced materials. methods in materials science for the representation of In particular, the development of new nanostructured structural features. The realisation of an eficient datafunctional materials can enable a wide range of applica- driven workflow for the prediction of critical materials tions in fields including electronics and optoelectronics, properties, for example for electronics applications as in energy, health [14, 15, 16]. To address complex struc- the case of graphene materials, can enable the eficient ture/property relationships in materials through data- design of advanced materials and boost the development centric approaches, our research group is active in R&D of applications and products. These results also highlight projects on the application of multiscale materials and the potential of a cross-disciplinary combination of comprocess modelling. Our approach consists in the develop- petences, as computer vision, image analysis and object ment of computational automated data-driven and high- detection and computational materials science, which throughput workflows for gaining a better understanding are commonly focused on diferent application fields. on the materials properties in the context of technology The integration of data and knowledge must be supported applications. Physical models provide a valuable predic- by robust approaches to standardisation and interopertive platform for the design of new materials and pro- ablity. To this end, our recent research eforts were tarcesses, correlating the results with available experimental geted to the application of semantic technologies, with data across a broad range of scales. Recent work demon- a focus on advanced materials, as a KET for research strated the potential of this approach in the development and industrial innovation. Indeed, the field of advanced of multi-scale functional novel materials for applications materials applications still requires significant eforts for example in optoelectronics (displays, lighting) and for standardisation, integration and interoperability. In for new-generation solar cells [15, 17, 18, 19]. Graphene- this respect, we recently carried out the development based materials, such as nanographenes and graphene of MAMBO - Materials and Molecules Basic Ontology oxide (GO) [20, 21, 22], constitute another particularly [26], as a fundamental step for the application of semanrelevant class of nanostructured low-dimensional mate- tic technologies in the field of advanced materials. In rials with a huge potential for the development of new addition to providing a domain ontology for materials applications in technology. One of the challenges for applications, the development of MAMBO helped us to the uptake of graphene research in industrial applica- assess the general requirements for the integration of tions is, however, related to the translation of lab-scale knowledge in fields where data and information are scatinnovation into technological solutions. The develop- tered and standardisation of workflows is still lacking. ment of high-throughput approaches for automating re- The integration of research tools and methods, from search on graphene-based materials is therefore a key high-throughput simulation methods, HPC, semantic step for innovation in this field. Basing on computational and software technologies, aims at providing a paradigm tools for building meaningful structure/property data for implementing multi-scale modelling frameworks for about GO samples [23], data-driven approaches can be advanced materials. This approach is currently being applied to generate integrated datasets, linking suitable pursued in the framework of collaborative national and representations of the structure of GO materials to target international R&D projects (for example, the BIO-SUSHY chemico-physical properties. The GrapheNet project, project, for developing sustainable surface protection by carried out by our lab, aims at exploiting AI technolo- glass-like hybrid and biomaterials coatings [27]). The gies in the field of graphene research. The main idea integration approach allows to connect high-throughput behind GrapheNet consists in applying AI and computer simulations to data-driven technologies, enables multivision frameworks commonly used in the analysis of im- scale links for the description of materials properties and ages to graphene-based materials. This approach stems provides a basis for predictive and generative platforms from the quasi-2D morphology of graphene and related for the design and development of functional materials and applications. smart cities, making it capable of communicating with The multi-scale approach finds application in the devel- the infrastructure and its environment. In the field of opment of functional devices where the properties of new industrial automation frameworks, current projects and advanced materials can be exploited. In this context, the activities aim at creating a link between physical models DAIMON Lab operates in tight connection with experi- of complex manufacturing components and processes mental research groups by providing digital platforms for and data-driven technologies. For example, low-code the design and predictive modelling of advanced devices. decision support design platforms (DSDP) can enable The development of devices for electronics and optolelec- the design, develop, deploy, and use of AI systems on tronics can enable a broad range of applications in several top of physical models of manufacturing systems. A advanced fields, such as bioelectronics, renewable energy DSDP therefore implements multi-level digital twins of sources and next-generation solar cells. Our eforts aim goods, equipment, parts, and processes, connected to AI at developing data-driven automated workflows for the services to improve their production and use. Work is generation of computational data on physical models in progress to develop application-specific digital twins of full-scale devices, also incorporating parameters an based on data-integration platforms. Similarly to the models at a lower scale. In this context, recent activi- other application fields discussed above, the design and ties include the development of automated workflows for implementation of digital frameworks for industry authe simulation of electronic devices based on functional tomation requires a strong degree of integration between molecular materials [28] and the analysis of data and data sources, physical and data-driven modelling workknowledge related to next-generation perovskite-based lfows and data and knowledge representation. In addition solar cells [29]. to the support in the development of industrial processes, A relevant application field related to industrial innova- this approach can enable automation in the R&D process, tion is that of mobility. Our integrated approach to tackle thus accelerating industrial innovation. innovation challenges in the context of smart mobility Most of the work described above is carried out within applications involves a multi-scale and multi-level per- collaborative national and international networks and spective. Namely, we address the challenges of smart initiatives. One of the most crucial aspects of the intemobility in current and future scenarios by considering gration eforts described above consists in the activation the interlinked data and information from the design of of successful paths for collaborative research and trainvehicle components to trafic in complex environments. ing. In this context, the DAIMON Lab is active in several Accordingly, this approach also requires a multisciplinary multidisciplinary initiatives, aimed at implementing this integration of data and technologies, in strong analogy integration approach. Recent initiatives include particiwith other use cases. Recently, we developed a proof pation in the national PhD programme in AI (phd-ai.it), of concept called SUMOhtms, an automation and stan- participation in H2020 and Horizon Europe projects and dardisation framework for the generation of scenarios networks, targeting a broad range of clusters and parfor the simulation of urban trafic based on the SUMO ticipation in Next Generation EU - PNRR activities and simulation package [30]. By automating workflows, we projects. A particularly relevant aspect concerns collabimproved the overall throughput of urban trafic simu- oration with industrial partners, in the framework of lations, thus broadening their scope and their potential national and international projects, public-private partintegration with data-driven and AI frameworks. Auto- nerships and other initiatives. The realisation of the R&D mated workflows simulate mobility scenarios by using objectives described above in a tight collaboration with the open-source geospatial data from OpenStreetMap industrial research teams and end users has proven a (OSM) as an input. The entire automated process is im- key enabler for the definition of successful technology plemented and managed by using Docker containers, transfer paths. enabling standardisation and cross-platform interoperability, leading to an overall increase of eficiency of the simulations. Additionally, we also took part in activities 4. Conclusions involving industrial partners in the field of trafic microsimulations, such as the MoMoTec (Modern Mobility The automation of digital tools for R&D and the integraTechnological Ecosystem) project, where microsimula- tion with AI and data-driven technologies and infrastructions were incorporated into the procedure for solving tures is crucial for driving innovation and development of the Capacitated Vehicle Routing Problem. Current re- advanced applications in research and industry. However, search in this context is focused on the realisation of a the lack of consolidated and shared practices, particularly digital twin of a self-driving car and its integration into in small-medium enterprises (SMEs), hampers the potenan urban trafic context: the aim is to train Reinforce- tial of AI for research and development. The approach of ment Learning algorithms for the active safety of the the DAIMON Lab tries to face some of these innovation vehicle, and integrate the trained agent in the context of challenges by developing horizontal platforms that can be applied to a broad range of vertical use-cases through the H2 technologies, the MoMoTec project and the PNRR implementation of interoperable integration frameworks. Ecosister project for support. Professor Mauro Gaspari Based on multidisciplinary eforts, the integration step (University of Bologna, Italy) and the aHead research is a prerequisite for the eficient application of AI tools team of Spindox SpA are also gratefully acknowledged for predictivity and automation. This approach enables for frutiful discussion and ongoing collaboration. the potential translation of processes of specific value chains into data-driven workflows for the digitalisation and modelling of complex systems and processes. The References use of data-driven and automation workflows is critical for generating integrated knowledge that can support AI and data-driven methods for prediction and automation, and for promoting collaboration and knowledge sharing across diferent areas of expertise. In turn, this multilevel technological integration can enable the application of AI tools for a broad range of operational processes, including the realisation of predictive platforms, design of processes and products, automation, optimisation, analysis of what-if scenarios, etc.
initiative for the realisation of the AI@ISMN infrastructure and the Italian National PhD program in Artificial Intelligence. We also thank the BIO-SUSHY project of the Horizon Europe programme, the MiTE project for phology of functional molecular aggregates on sub- ergy Materials 9 (2019) 1901019. doi:https://doi. strates, Advanced Theory and Simulations 2 (2019). org/10.1002/aenm.201901019. doi:10.1002/ADTS.201900156. [20] W. Gao, The chemistry of graphene oxide, [9] J. Fish, G. J. Wagner, S. Keten, Mesoscopic Graphene oxide: reduction recipes, spectroscopy, and multiscale modelling in materials, Na- and applications (2015) 61–95. ture Materials 20 (2021) 774–786. doi:10.1038/ [21] M. Baldoni, F. Mercuri, Evidence of benzenoid s41563-020-00913-0. domains in nanographenes, Physical Chemistry [10] T. Berners-Lee, J. Hendler, O. Lassila, The seman- Chemical Physics 17 (2015) 2088–2093. tic web, Scientific American 284 (2001) 34–43. [22] A. K. Geim, Graphene: status and prospects, science URL: http://www.sciam.com/article.cfm?articleID= 324 (2009) 1530–1534.
00048144-10D2-1C70-84A9809EC588EF21. [23] A. Barnard, M. Soumehsaraei, Benyamin, B. Sun, [11] T. R. Gruber, A translation approach to portable L. Lai, Neutral graphene oxide data set. v1. csiro. ontology specifications, Knowledge Acquisition 5 data collection, 2019. URL: https://doi.org/10.25919/ (1993) 199–220. doi:10.1006/KNAC.1993.1008. 5e30b44a7c948. doi:10.25919/5e30b44a7c948. [12] B. Bayerlein, T. Hanke, T. Muth, J. Riedel, [24] G. B. Goh, C. Siegel, A. Vishnu, N. O. Hodas, M. Schilling, C. Schweizer, B. Skrotzki, A. Todor, N. Baker, Chemception: A deep neural network B. M. Torres, J. F. Unger, C. Völker, J. Olbricht, A with minimal chemistry knowledge matches the perspective on digital knowledge representation in performance of expert-developed qsar/qspr models, materials science and engineering, Advanced En- 2017. arXiv:1706.06689. gineering Materials 24 (2022). doi:10.1002/ADEM. [25] K. Mills, I. Tamblyn, Big graphene dataset, 202101176. 2019. URL: https://doi.org/10.4224/c8sc04578j.data. [13] M. T. Horsch, S. Chiacchiera, Y. Bami, G. J. Schmitz, doi:10.4224/c8sc04578j.data, 1 compressed G. Mogni, G. Goldbeck, E. Ghedini, Reliable and in- .tar.gz file (3.77 GB) containing 60,744 .h5 testing teroperable computational molecular engineering: ifles and 501,473 .h5 training files. 2. semantic interoperability based on the european [26] F. L. Piane, M. Baldoni, M. Gaspari, F. Mercuri, Inmaterials and modelling ontology, arXiv (2020) troducing mambo: Materials and molecules basic 1–25. URL: https://arxiv.org/abs/2001.04175. ontology, arXiv preprint arXiv:2111.02482 (2021). [14] D. Selli, M. Baldoni, A. Sgamellotti, F. Mercuri, [27] BIO-SUSHY, 2023. URL: https://www.bio-sushy.eu/, Redox-switchable devices based on functionalized accessed: 2023. graphene nanoribbons, Nanoscale 4 (2012) 1350– [28] D. O. Rocha Cadena, Simulación y fabricación de 1354. celdas solares de perovskita híbrida con arquitec[15] A. Lorenzoni, M. Muccini, F. Mercuri, Correla- tura invertida (pin), Master’s thesis, Tesis (MC)– tion between gate-dielectric morphology at the Centro de Investigación y de Estudios Avanzados nanoscale and charge transport properties in or- del IPN . . . , 2023. ganic field-efect transistors, RSC advances 5 (2015) [29] T. J. Jacobsson, A. Hultqvist, A. García-Fernández, 11797–11805. A. Anand, A. Al-Ashouri, A. Hagfeldt, A. Crovetto, [16] M. Seri, F. Mercuri, G. Ruani, Y. Feng, M. Li, Z.-X. A. Abate, A. G. Ricciardulli, A. Vijayan, et al., An Xu, M. Muccini, Toward real setting applications of open-access database and analysis tool for perorganic and perovskite solar cells: A comparative ovskite solar cells based on the fair data principles, review, Energy Technology 9 (2021) 2000901. Nature Energy 7 (2022) 107–115. [17] A. Lorenzoni, A. M. Conte, A. Pecchia, F. Mercuri, [30] P. A. Lopez, M. Behrisch, L. Bieker-Walz, J. ErdNanoscale morphology and electronic coupling at mann, Y.-P. Flötteröd, R. Hilbrich, L. Lücken, J. Rumthe interface between indium tin oxide and organic mel, P. Wagner, E. Wießner, Microscopic trafic molecular materials, Nanoscale 10 (2018) 9376– simulation using sumo, in: 2018 21st international 9385. doi:10.1039/c8nr02341g. conference on intelligent transportation systems [18] A. Lorenzoni, M. Muccini, F. Mercuri, A computa- (ITSC), IEEE, 2018, pp. 2575–2582. tional predictive approach for controlling the morphology of functional molecular aggregates on substrates, Advanced Theory and Simulations 2 (2019) 1900156. [19] Y. Feng, Q. Hu, E. Rezaee, M. Li, Z.-X. Xu, A. Lorenzoni, F. Mercuri, M. Muccini, High-performance and stable perovskite solar cells based on dopantfree arylamine-substituted copper(ii) phthalocyanine hole-transporting materials, Advanced En