Design the modern supply chain: The SmarTwin Project Michele Di Capua1*,†, Emanuel Di Nardo1,†, Angelo Ciaramella2,†, Gennaro Iannuzzo2,†, Aniello De Prisco3,†, Daniele Ruggeri Laderchi3,†, Pietro Catalano4,† , Pietro D’Ambrosio4,†, Salvatore Moscariello4,† 1US srl, via Porzio, Centro Direzionale di Napoli Isola G2, Naples, 80143, Italy. 2DIST – Dipartimento di Scienze e Tecnologie – Università di Napoli “Parthenope”, Naples, 80143, Italy. 3Magsistem spa, Zona Industriale, 81030 Gricignano di Aversa (CE). Italy. 4LinearIT spa, Via Giovanni Severano, 28, 00161 Rome, Italy. Abstract The SmarTwin project aims to define and explore an innovative service model with two main strategic objectives: on the one hand, to anticipate needs arising from both the market (growing demand for quality and consumer awareness) and the business world (cost reduction, environmental sustainability, optimization); on the other hand, to identify and begin to occupy a strategic convergence space related to future and emerging trends in a number of key enabling technologies such as Artificial Intelligence, IoT, Digital Twin and others. The basic idea is to enable supply chains (and especially those dealing with perishable goods) to achieve new levels of efficiency in terms of overall quality and service cost reduction, certification and tracking of each activity carried out in the production process, minimization of human health risks and reduction of the ecological footprint of products. The supporting software platform of the project has been designed on AI (Artificial Intelligence) components and will extend a number of key enabling technologies such as IoT, Blockchain, AI and Digital Twin to be able to represent and manage the new levels of complexity expected to be required to effectively address the market needs identified. Keywords Supply Chain, Digital Twin, Machine Learning, Blockchain, Internet of Things.1 1. Introduction and one financial. The methodological aspect is aimed at providing normative and operational support to The SmarTwin project aims to investigate an supply chain actors for the implementation of the new innovative service model for cost optimization, risk hypothesized service model. reduction, micro-traceability of a product's The technological aspect is aimed at automatically processing steps, certification of the ecological detecting events, typical activities, quality levels (of footprint and, finally, financial support for complex products) and operating conditions (of the various supply chains [2], with a focus on production chains contexts in which activities take place) through the dealing with perishable products (i.e. vegetables). use of IoT technologies, Artificial Intelligence The project has several innovative aspects. The (machine learning [3] and computer vision [4]) and first is represented by the approach to the problem, predictive analytics systems (e.g., with artificial which is developed on three parallel and integrated neural networks and big data analytics systems). This logical levels: one methodological, one technological information properly collected and analyzed will Ital-IA 2024: 4th National Conference on Artificial Intelligence, 0000-0001-8904-180X (M. Di Capua); 0000-0002-6589-9323 organized by CINI, May 29-30, 2024, Naples, Italy (E. Di Nardo); 0000-0001-5592-7995 (A. Ciaramella); 0009- ∗ Corresponding author. 0003-5962-8302 (G. Iannuzzo); † These authors contributed equally. © 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0). m.dicapua@usmail.it (M. Di Capua); e.dinardo@untec.it (E. Di Nardo); angelo.ciaramella@uniparthenope.it (A. Ciaramella); gennaro.iannuzzo@uniparthenope.it (G. Iannuzzo); CEUR ceur-ws.org Workshop ISSN 1613-0073 Proceedings specific specification), and reduction of environmental impact (e.g. minimization of energy consumption). These objectives are essentially pursued by realizing a technological platform that combines the dynamic functionalities required to manage the configuration and constant evolution of the proposed service model and those of descriptive and predictive analysis on the Digital Twin of the supply chain ecosystem (see figure 1), with a series of specific technological components, based on AI, for the recognition of events, activities occurring in the Figure 1. The Digital Twin schema. operational context, potential critical scenarios and for the constant verification of consistency between enable the feeding of a much more extensive product what is detected in the real process with what is certification and tracking system than those currently provided for at the contractual and regulatory level. used. The financial aspect, on the other hand, represents an absolute novelty for this type of solution and will allow for versatile and secure 3. Data integration management of economic relations between the As already mentioned, the main idea of the project various supply chain partners. In particular, within is to approach the problem of managing a complex the system, in addition to supporting with specific supply chain ecosystem in a holistic way, constantly functions the process of evaluation and analysis of monitoring every aspect, every phase of work, every each step of the economic stage, a virtual currency will relationship between partners and every event that be introduced (to be connected to the Digital Twin of may be useful to ensure the quality specifications the supply chain [1] model) to manage payments, expected by the customer and to maximize the anticipate liquidity within the circuit and to securitize benefits for the operators in terms of optimizing costs, credit. Overall, the presence of the financial time and the objectives set. Obviously, the model must "dimension" in the governance of supply chain be adaptable to any supply chain context and be able processes allows us to enhance the new level of "trust" to follow the natural evolutions that occur in these that we aim to achieve with the application of the contexts to manage seasonality, workload proposed service model and that not only involves the discontinuities and any contingencies that may arise. economic operators in the supply chain but also It will therefore be necessary to define all the extends to end consumers. information relating to the actors, the processes, the quality objectives to be guaranteed (including 2. The Digital Twin metaphor sustainability objectives, specific production specifications, etc.) and, of course, the commercial The proposed service model is characterized by its relationships established between the various actors. innovative features and, in particular, by its focus on Furthermore, for each "measurable" element to be defining a digital twin of the entire supply chain detected in the process, it is assumed that an IoT ecosystem to achieve a holistic view and a new level sensor network or different types of cameras (e.g. of detail and pervasiveness of process monitoring, thermal), whose data streams are constantly analyzed tracking, quality control and certification activities to by the AI components of the system, can be plugged in address and resolve the typical complexity of contexts and used. All the configuration information required such as productive supply chains. to build the model should be defined, captured and The digital model aims to represent the various managed by the software platform in such a way that organizational, economic and regulatory dynamics of its acquisition is easy and immediate and does not a modern supply chain, analyzing the huge amount of create functional access or operational cost barriers to data collected by the network of integrated sensors adoption. Therefore, it is necessary to provide and devices (e.g. cameras), in order to suggest functions for defining the configuration logic and the optimization actions aimed at cost reduction (e.g. rules behavior of the system that are, as far as avoidance of waste), prevention of risks to people (e.g. possible, driven by automatic processes and are timely control of the correct arrangement of goods in dynamically adaptable and reusable in other phases of the warehouse), constant monitoring of product the supply chain. On the other hand, as far as the IoT quality (e.g. verification of consistency of storage and sensor network and image stream acquisition sources transport conditions with what is required by the is used to detect the status of activities and operational contexts both referring to working environments (e.g., warehouses for the storage of goods) and to the mobile network of multimodal transportation (e.g., sensors on board refrigerated trucks for the transport of perishable goods). These data, appropriately normalized and aggregated, can already represent an element of interest for the purposes of tracking and monitoring Figure 2: The general architecture of the designed the operating conditions of the supply chain software platform. ecosystem, but at the same time they also constitute a knowledge base that can be used for the creation of are concerned [6], the system must be able to census predictive models suitable for recognizing the devices, identify their location and usefulness for characteristic behaviors or events, that can be the purpose of assessing the operating conditions and associated with the statements of supply chain the associated rules for normalizing, aggregating and operators or for evaluating elements of optimization analyzing the collected data. and risk prevention. From this point of view the IoT The goal of the system in this case is to be able to platform, integrated into the SmarTwin system, verify continuous compliance with the optimal provides for collecting data, aggregating them on a operating conditions and, at the same time, to certify time basis, doing systematic checking of predefined the consistency of what is detected in the real rules and triggering any alerts, making them environment with what is stipulated in the searchable by the application functions that agreements and work processes defined in the service implement the supply chain Digital Twin and making model. them available to the various AI-based analytics tools Finally, the monitoring of all events in the supply and optimization components working in the chain (micro-traceability) and the certification of background. processes and activities carried out, together with the digitalization of contractual agreements through 4.2. Video analysis and machine learning smart contracts, will also enable the designed platform to automatically settle economic aspects This subsystem is in charge of the camera network between parties, both by generating payment flows census activities and the management of video and by offering operators the possibility of streams, from the various devices installed in the securitizing their acquired (and certified) receivables supply chain environment (in places where typical for use within the system as an alternative form of activities take place such as warehouses, loading and liquidity. unloading pallets, etc.), and for each of them it is concerned with submitting the streams to the analysis and image processing components to identify events, 4. The software platform typical activities or situations of potential interest. The software platform, that implements the The final components (see figure 3) that will be in service model, represents the basic technological charge of image stream analysis (thus based on infrastructure into which are integrated both the machine learning and computer vision), will be dynamic system configuration functionalities designed as Artificial Neural Networks (ANN) with (necessary to digitize the model), the services offered different tasks, each of which is concerned with by the specific components (such as the IoT platform, detecting particular types of events (e.g. the presence Blockchain and AI components), and the operational of people in the area) or capturing particular functionalities and those that make the Digital Twin of information (e.g. the license plate of a vehicle that is in the overall supply chain ecosystem usable to various a loading/unloading area). Obviously due to the types of users. The different components of the specificity of the technology used, the context and the platform will be detailed below and are shown in purpose of the analysis, it is not always possible to figure 2. successfully use pre-trained or fully reusable components in different similar contexts. We will 4.1. The IoT layer therefore proceed on the one hand with the identification of a set of typical events or actions that It presides over all activities of continuous can be generalized and, on the other hand, with the acquisition of data from the IoT sensor network that artificial neural networks. The main objective here is to predict potential problems and recurring data patterns, to improve supply chain performance, from the data collected by the IoT platform envisaged by the project. From a design perspective, it's necessary to develop versatile and reusable machine learning modules that can be "plugged" into the platform to provide information about potential problems that may arise along the supply chain being analyzed. The models will learn the normal dynamics of information flows that characterize one or more processes and will be able to recognize "anomalous" patterns of data to prevent risks and consequent process-related damage or waste, with a view to making the supply chain more effective and sustainable. The theoretical approach adopted for this purpose is the time-series analysis, which focuses on understanding data that vary over time. It is particularly relevant in IoT, as much data produced by connected devices is Figure 3: Component Diagram (UML) of the software organized into time series, such as sensor data and platform of the project. measurements at regular intervals. In the more recent landscape, transformer identification of technological solutions that can allow models, which were initially introduced for natural the integration into the system of specific components language processing problems, are also showing good trained ad hoc for a particular task. For example, it results in this area. Transformers can handle data may be necessary to monitor the state of preservation sequences of varying lengths and can be used for time of a particular type of product in a warehouse, and to series prediction and analysis tasks. Finally, it is useful achieve this type of monitoring it may be necessary to to mention also autoencoders, which are deep employ two cameras of different technology (e.g. a learning models used for dimensionality reduction thermal camera and an RGB camera) and to train a and feature representation of data, and which can be specific neural network to process these video used in time series analysis to extract relevant streams and detect the state of preservation at a features and reduce dimensionality of data, which is a specific frequency. Obviously, this component will potential issue to be addressed in the application have to be purpose-built and can only be used in the scenarios of the project. Our analysis is confirming same environments, conditions and for the same type that these deep learning techniques can be used for of product for which it was originally designed. In the various time series analysis tasks related to a supply context of the final prototype, we will try to imagine chain, such as short or long-term forecasting, anomaly some case scenarios where such conditions might detection, trend and pattern analysis. Relative to the occur, to try to integrate purpose-built AI components transformer models identified in the first part of the into the system for a specific function. state-of-the-art analysis of the project, some As a whole, the subsystem that deals with the preliminary tests were conducted on literature acquisition of image streams collects images at a set datasets to verify the effective ability of these models frequency from the various devices and submits them to also be used in the context of time series to the network of specialized AI components for forecasting. Preliminary results confirmed that recognition of typical events and activities, it collects transformer models can be used with good all outputs towards the Digital Twin, in order to make effectiveness also with these types of data. data available (API) to the monitoring and alerting In the specific area of IoT sensor data analysis, an component of the service platform. autoencoder-based neural network was tested to identify vibration anomalies from sensor readings 4.2.1. Predictive tasks based on ML installed on a series of bearings. The aim of the preliminary test was to be able to predict future In order to integrate predictive task capabilities bearing failures before they occur. into the project, recent machine learning techniques have been explored, with a focus on the use of deep 4.2.2. Quality inspection and computer vision In the next stages of the project, possible automated approaches will be analyzed, to support quality control processes within warehouses and the quality verification of perishable products, through the adoption of computer vision techniques. The analysis referred to should be non-invasive, i.e. based on the placement of fixed cameras in places where operators normally carry out visual inspections. Figure 4: Pre/Post explainability modelling approach Digital images from cameras and videos can be used schema [10]. to train computer vision models dedicated to the inspection and analysis of products/goods. The help better understand why an AI model generates computational vision tasks that will be implemented certain decisions by describing how it works. next will cover both general and specific tasks in the According to recent research made by Forrester, area of supply chain process control [5]. For example, XAI represents a phenomenon capable of generating machine learning models will be trained to detect numerous tangible benefits for those who regularly specific machinery (e.g. forklifts) moving in logistics adopt it within their process management: aisles, rather than the presence of people in storage areas. It will also be possible to define and integrate • a reduction in model monitoring efforts more specific and advanced neural network models ranging from 35 percent to 50 percent; that instead perform Human Activity Recognition • an increase of up to eight times the number (HAR) tasks, i.e. the recognition of specific human of models in production; activities (e.g. picking) according to specific logistics • an overall accuracy of the model itself that use cases. can be estimated at 15 percent to 30 percent; • an increase in the profit range that can even 4.3. Explainable AI (XAI) triple. Within the designed platform of the project there Multimodal data, which characterize control are several components that adopt Artificial functions, will be treated with innovative computer Intelligence (AI) or Machine Learning (ML) vision techniques, like image classification, visual techniques to perform tasks traditionally reserved for object tracking, and reliable and interpretable AI the human operator and in particular: for the techniques (ante-hoc and post-hoc methods). In automatic classification and analysis of information addition, methodologies to improve resilience will be acquired from the supply chain activities, the used (see figure 4). For the design of decision support performance of automatic checks, to make predictions models adopted for risk assessment [8] from about operating conditions and finally to suggest information flows that characterize one or more possible optimizations. In this case, the technological supply chain processes, XAI methodologies will be issues to be addressed are essentially: the recognition used to improve multimodal data aggregation and of typical events and activities (HAR - Human Activity decision logic (e.g., Fuzzy Logic). Data mining Recognition) using image processing techniques and methodologies will be explored for preprocessing methodologies based on deep architectures, the multimodal data (e.g., intrinsic dimension) along with creation of predictive models from the large amount signal processing and computer vision techniques. of data collected from the IoT sensor network (using Data for learning and testing the designed models will machine learning techniques), and also the be acquired from information flows, images and implementation of a decision support system (DSS), videos obtained during the operational phases of the which is necessary to be able to apply the expected supply chain. In the early design stages of the automatic control rules. SmarTwin system, synthetically created realistic data In this regard, to mitigate the distrust that often will be used and data augmentation, concept drift and arises in adopting AI components as mere functional Active Learning methodologies will be adopted to black boxes, whose actual behavior in terms of make the models more robust. Among the possible decisions made is not always clear and justifiable in methodologies adopted for explainability purposes, detail [7], development approaches, based on the Neuro-Symbolic (NeSy) AI has been explored in this emerging branch of eXplainable, AI will be adopted. early design stage of the project. The set of tools and techniques used in XAI aims to 4.3.1 Neuro-Symbolic Artificial Intelligence 5. Conclusions The goal is to create a system that can learn from This paper presents the SmarTwin project, which large amounts of data, much like neural networks, but aims to integrate the latest machine learning and also has the high explainability and provable blockchain technologies into production supply correctness of symbolic systems. One of the key chains in order to improve the quality of their differences between symbolic and neural systems is component processes, as well as their associated how they represent knowledge. Symbolic systems use economic and environmental impacts. The project is explicit representations that humans can understand. still at an early stage and will be completed in 2026, They structure knowledge in a logical way, often using but the preliminary results of the research activities logic-based languages. Neural systems, on the other outlined here confirm the validity of the initial idea. hand, use distributed representations that are hard to interpret. They learn representations from data in an Acknowledgements implicit way, which makes them highly adaptable but also opaque. It’s seen as a promising approach to The project was funded by the Italian Ministry of overcoming the limitations of purely symbolic or Economic Development (MISE): it started in 2023 and purely neural methods and to advance towards more it will end in 2026. Project reference no. intelligent and human-like AI systems. However, NeSy F/310218/01-02-03-04-05/X56. AI also faces several challenges. These include the difficulty of integrating symbolic and neural methods, the need for large amounts of data to train neural References networks, and the lack of interpretability of neural [1] H. van der Valk, et al. “Supply Chains in the Era of representations. Digital Twins – A Review”, Procedia Computer Science, Volume 204, (2022). 4.4. Blockchain integration [2] Lambert, D. M., et al. "Supply chain management: implementation issues and research opportunities." This subsystem deals with the recording on The international journal of logistics management 9.2 Blockchain of any information, event, transaction or (1998): 1-20. activity that needs to be tracked or is involved in the [3] H. Wenzel et al. "A literature review on machine certification processes under the service model. This learning in supply chain management." Proceedings of will make it possible to make such records "public," the Hamburg International Conference of Logistics transparent and unchangeable. Of course, some of the (HICL), Vol. 27. Berlin, 2019. information recorded on the Blockchain, although [4] Sun, Yuan, et al. "Improved Commodity Supply Chain Performance through AI and Computer Vision public, may also not be made visible to all users (in Techniques." IEEE Access (2024). which case it would be encrypted) to implement [5] S.S. Abosuliman et al. "Computer vision assisted different access policies and differentiate levels of human computer interaction for logistics management visibility based on products and supply chain rules. using deep learning." Computers & Electrical Each record will include, where possible, elements to Engineering 96 (2021): 107555. certify the information entered also by including [6] Ben-Daya M. et al., "Internet of things and supply chain possible references to further records made in other management: a literature review." International related Blockchains. Regarding the latter aspect, journal of production research 57.15-16 (2019) considering the needs and operations [9] that will be [7] G. Mugurusi et al. "Towards explainable artificial envisaged on the Blockchain, we will evaluate in the intelligence (xai) in supply chain management: a typology and research agenda. "IFIP Conference. next future, as part of the specific research activities Springer International Publishing, 2021. that will be carried out, which implementations and [8] Nimmy, Sonia Farhana, et al. "Explainability in supply how many Blockchains to use and how to link them chain operational risk management: A systematic together. Recall, that in addition to the needs for literature review." Knowledge-Based Systems 235 tracking and certification of events and activities, we (2022): 107587. plan to manage by means of Smart Contract all [9] Chang, Shuchih E., and Yichian Chen. "When contracting between supply chain actors and all blockchain meets supply chain: A systematic literature economic transactions. A further use that will be made review on current development and potential of Blockchain and Smart Contract will be that applications." Ieee Access 8 (2020): 62478-62494. necessary for the virtualization of accrued credit and [10] Pandian, S. (2022). “Explainable Artificial Intelligence (XAI) for AI & ML Engineers”. [online] Analytics for the eventual securitization and transfer of the Vidhya. same to a secondary market.