Distributed cyber-physical system supported by collaboratively updated digital twins

Abstract

The present invention discloses a distributed cyber-physical system (CPS) (1) comprising a plurality of CPS cells (2) provided with sensors, first processing means, first data transmission and reception means, and first storage means. The system further comprises an intermediate communication module (4), communicatively connected to the CPS cells (2) and suited with second processing means and second data transmission and reception means; and a global storage module (3), based on blockchain technologies, communicatively connected to the intermediate communication module (4) and suited for storing data in the CPS cells (2). This configuration allows implementing a method for collaboratively improving digital twin models (6) of part or all of the entire CPS system (1), or of particular processes or services that the system executes or provides, without interfering with said system's operation.

Description

[0001] DESCRIPTION

[0002] Distributed cyber-physical system supported by collaboratively updated digital twins

[0003] FIELD OF INVENTION

[0004] The present invention falls within the field of systems composed of physical elements intertwined and interconnected with software elements via a communication network, known as cyber-physical systems (CPS), as well as their monitoring and control using digital twin models. More specifically, the present invention discloses a cyber-physical system supported by digital twin models that are collaboratively improved or updated in real time by the various CPS cells comprising the CPS system.

[0005] BACKGROUND OF THE INVENTION

[0006] The combined and parallel advancement of various technologies such as ubiquitous computing, sensors, embedded systems, and machine learning has enabled the development of systems comprised of diverse devices, each equipped with sensors, that connect and communicate with each other via the Internet or other communication networks. These systems, which also include data processing tools and software, are often categorized as the Internet of Things (IoT) and offer significant advantages in monitoring, operation, and process control in all types of applications: industrial, medical, transportation, etc.

[0007] Today, IoT systems, combined with Artificial Intelligence (AI) or Machine Learning (ML) techniques, enable the analysis of large amounts of data collected by the sensors of different devices, offering a high level of control and insight into the operating conditions of these systems. This enables the creation of models called Digital Twins (DTs), which are virtual or software copies of physical entities and / or processes. These digital copies are created and configured so that, when they receive the same stimuli or input data as the physical object or process of which they are twins, they provide the same responses or output data as the equivalent physical object or process. The implementation of DTs is especially advantageous in cyber-physical systems (CPS).In these systems, the physical and software elements are deeply intertwined, each operating at different scales, exhibiting multiple behaviors, and interacting with the rest of the system in a context-dependent (and therefore variable) manner. For this reason, the use of DGs as part of a SCF's operation allows it to perform operations such as predicting potential failures of varying severity or predicting certain behaviors before they occur, or offering responses (or service) to certain requests from external entities or users with the precision of emulated entities or processes from the physical world. Considering the complexity of these types of systems, this offers the possibility of avoiding, for example, the suspension of the entire system and its subsequent restart, which, if it occurs, could entail significant losses.

[0008] However, precisely because of the high complexity of SCFs, generating and updating associated DGs is an extremely arduous task requiring large data storage spaces and significant processing power. Furthermore, certain applications of these DG-equipped systems, such as medical or defense applications, impose strict security requirements regarding the processing of data involved in the system and in the DGs, as well as additional requirements related to resource efficiency and the capacity or response speed of SCF processes.

[0009] Blockchain technologies make it possible to meet some of these requirements, as they provide a secure (verifiable) and reliable distributed data storage environment. In this sense, data is distributed across the various elements that comprise the SCF, and data processing and transmission protocols are established, which must be validated by one or more elements of the system itself in order to be executed. Furthermore, blockchain-type data structures can act as a database containing an irrefutable history of information shared by the SCF.

[0010] Despite the advantages of using blockchain in SCFs that utilize DGs in their operations, its current implementation has resulted in rigid systems in which the development and improvement of the DG model is separated from the operation of the system itself. Furthermore, in most of these cases, it is necessary to halt the operation of the system or several subsystems within it to update the DG model (e.g., by training it with new data). Once updated, it is uploaded to the SCF, which is subsequently restarted. This procedure is not optimal due to the losses caused by the temporary cessation of system operations and the costs associated with stopping and resuming its activity.

[0011] On the other hand, blockchain technologies have typically been used as a mechanism for storing and managing data and transactions, but never as solutions focused on providing a collaborative space for developing and improving AI / ML models, including GD models.

[0012] Therefore, the state of the art lacks solutions that provide a SCF system comprising a global operating space accessible to all elements of the SCF, enabling the secure, low-cost, and real-time collaborative development and improvement of DG models.

[0013] BRIEF DESCRIPTION OF THE INVENTION

[0014] In order to address the limitations of the state of the art described above, the present invention aims to provide a novel cyber-physical system, SCF, composed of a plurality of SCF cells connected to a global storage and / or operation space through an intermediate communication module.Each of the SCF cells is provided with sensors and processing means comprising software configured such that each cell can: generate one or more digital twin models of the entire SCF system or a part of said SCF system; transmit and / or receive one or more versions of a digital twin model and / or data to / from the global space; train a digital twin model from data collected by its sensors, data received from the global space and / or data relating to the state, processes or services of the SCF system received from a repository external to the SCF system connected to said SCF system; and execute a digital twin model from data collected by its sensors, data received from the global space and / or data relating to the state, processes or services of the SCF system received from a repository external to the SCF system connected to said SCF system.Furthermore, all of this can be done while the other SCF cells within the system continue to operate, allowing the creation, updating (training), and execution of digital twin models to be performed in real time without having to pause the operation of the entire system.

[0015] On the other hand, the global system is adapted to host a blockchain-type data structure (corresponding to a global storage space), and the intermediate communication module comprises software configured to manage the transmissions and receptions of digital twin models and / or data carried out in each SCF cell as smart contract transactions executable by the global module. In this way, data processing is performed securely, and a history of non-redundant versions of the digital twin model(s) can be stored.

[0016] The intermediate communication module is also configured to collect data from one or more SCF cells and create one or more training data sets from them, thereby collaboratively creating data sets from across the SCF system to optimally train the digital twin models.

[0017] More specifically, a first object of the invention relates to a distributed cyber-physical system, SCF, comprising:

[0018] - a plurality of SCF cells, each of said CPS cells comprising, in turn:

[0019] • one or more sensors adapted to obtain operational or performance data from the SCF cell;

[0020] • first means of data processing;

[0021] • a first means of transmitting and receiving data;

[0022] • first data storage media;

[0023] - an intermediate communication module, communicatively connected to the plurality of SCF cells and adapted with second data processing means; and with second means for transmitting and receiving said data; and

[0024] - a global data storage module, communicatively connected to the intermediate communication module, and adapted to store the data of the SCF cells.

[0025] In one embodiment of the invention, the intermediate communication module may be included in the global storage module. Additionally, the global storage module may be formed in whole or in part by the set of first data storage means included in the SCF cells. Advantageously, the first processing means and the first data transmission and reception means of each SCF cell are configured with one or more artificial intelligence algorithms, adapted to:

[0026] • generate one or more digital twin models of at least a part of the SCF system;

[0027] • transmit information from a digital twin model to the global storage module;

[0028] • receive information from a digital twin model stored in the global storage module;

[0029] • execute a digital twin model from the operational or performance data of one or more SCF cells of the system obtaining predictive data;

[0030] • train, in whole or in part, a digital twin model using operational or performance data from one or more SCF cells and / or predictive data as training data;

[0031] • transmitting operational or performance data from one or more SCF cells and / or predictive data generated by executing a digital twin model to the global storage module and / or to a repository external to the SCF system connected to said SCF system; and

[0032] • download operational or performance data from one or more SCF cells, predictive data generated by running a digital twin model stored in the global storage module and / or data relating to the status, processes or services of the SCF system stored in a repository external to the SCF system connected to said SCF system.

[0033] In this way, the described system allows for the collaborative improvement or updating of digital twin models of an SCF system (and / or subsystem). This improvement or update is carried out thanks to the contributions of each SCF cell individually. Therefore, the SCF cells of the described system can work autonomously or individually, performing the processes or operations associated with each of them as part of the SCF system, while also contributing to the improvement or updating of digital twin models of the entire system or a subsystem thereof. This is possible because the global storage module functions as a collaborative and secure (verifiable) space through which the SCF cells can communicate via the intermediate communication module.In another embodiment of the invention, the global storage module and the first data storage means are configured to host a blockchain-type data structure, storing consecutive, non-redundant versions of one or more digital twin models. This allows, on the one hand, to secure data transmissions as transactions within the blockchain structure and, on the other, to store an ordered history of non-redundant versions of the digital twin models, which is advantageous when it is desired to predict the system's behavior under certain past conditions.

[0034] In another embodiment of the invention the second processing means comprise software means configured to:

[0035] - manage the transmissions and receptions of digital twin models and / or data performed in each SCF cell as smart contract transactions executable by the global storage module; and

[0036] - collect operational or performance data from one or more SCF cells and / or predictive data and use them to create one or more training data sets.

[0037] In another embodiment of the invention, the algorithm(s) comprised by each cell comprise a neural network. This neural network may correspond to different typologies, including, but not limited to, feedforward, perceptron, multi-layer perceptron, or deep (such as convolutional or recurrent). The use of neural networks (and specifically deep neural networks) enables the execution of digital twin models in real time, since, once trained, their execution as predictive (or classification) models is performed almost instantaneously. This is advantageous because, although the time consumed in training the models may be extended, once fully or partially trained, said models have very low response times, meeting the typical demands of complex systems such as SCFs.

[0038] In another embodiment of the invention, the connection between the global storage module and the intermediate communication module and / or the communication between the intermediate communication module and each cell is remote, belonging, for example, to the same telecommunications network. This telecommunications network can be wired, wireless, or a combination of these. Furthermore, each cell can be connected to the other cells through a telecommunications network, facilitating the exchange of information and data between the different elements of the system, even under the verification protocols implemented using blockchain technology.

[0039] A second object of the invention relates to a method for real-time updating of a digital twin model of a cyber-physical system (SCF), according to any of the described embodiments. The different embodiments of the method correspond to the generation, updating or improvement through training and / or execution of one or more digital twin models of the entire SCF system in a collaborative manner through the SCF cells. Each of these cells can generate, update or execute a model individually while the rest of the SCF cells comprised in the system maintain their operation. In this way, the models are updated and improved in the different cells without the SCF system having to pause its processes at any time.The information (data and models) received or generated in each SCF cell is shared with the rest through the global storage module that functions as a collaborative space for this task, being verifiable by all SCF cells.

[0040] The method of the invention comprises performing the following steps in any technically possible order: a) receiving, in a first SCF cell (2):

[0041] - operating or performance data of one or more of the other SCF cells (2) via the intermediate communication module (4) and / or the global storage module (3)

[0042] - predictive data of a physical entity, process and / or service associated with one or more of the other SCF cells (2) through the intermediate communication module (4) and / or the global storage module (3); and / or

[0043] - data relating to the status, processes or services of the SCF system stored in a repository external to the SCF system (1) connected to said SCF system (1); while at least one of said other SCF cells (2) is in operation in the system (1); b) generating, in the first SCF cell, a digital twin model of at least a part of the SCF system; c) transmitting information about the digital twin model generated in step b) to the global storage module from the first SCF cell through the intermediate communication module. The information about the digital twin model corresponds to parameters of said model from which the model can be reconstructed without needing to be generated again. For example, if the model were a neural network, this information or parameters would correspond, among others, to the weights of the neural network.In deep neural networks (very large and complex networks that allow the construction of highly complex and precise GDs), storing the entire network model (i.e., the complete network) would be very expensive and would significantly overload the blockchain that serves as the global storage space. Transferring only information, and not the entire model, to the global storage space greatly reduces memory overhead and speeds up the processes of receiving and transmitting the model by SCF cells, since relatively little information is transferred between the global space and the SCF cell(s) in question.

[0044] In another embodiment of the method of the invention, this comprises the additional performance of the following steps in any technically possible order: d) training totally or partially, in the first SCF cell, the digital twin model generated in step b) using as training data:

[0045] - the operating or performance data received in step a);

[0046] - the operating or performance data generated in the first SCF cell (2) itself, from the sensors included therein; and / or

[0047] - data stored in a repository external to the SCF system (1) connected to said SCF system (1) obtained in step a); and e) transmitting information about the digital twin model trained in step d) to the global storage module from the first SCF cell through the intermediate communication module.

[0048] In another embodiment of the method of the invention, this comprises the additional performance of the following steps in any technically possible order: f) executing the digital twin model trained in whole or in part in step d) using as input data operating or performance data received in step a), and / or operating or performance data generated in the first cell itself, obtaining first predictive data; and g) transmitting, from the first SCF cell, the first predictive data obtained in step f) to the global storage module through the intermediate communication module and / or to a repository external to the SCF system connected to said SCF system. The predictive data may be transmitted to a repository other than the global storage space (blockchain) connected to the SCF system when said data sets are large in size so that they could overload the memory capacity of the blockchain.Thus, faster and more efficient access to such predictive data is achieved if, for example, they are stored in an external database (which may be document-oriented or, in a second derivative, a relational database shared by the SCF cells, whose access must be synchronized).

[0049] In another embodiment of the method of the invention, it comprises the further performance of the following steps in any technically possible order: h) receiving, in a second SCF cell (2), operating or performance data of one or more of the other SCF cells (2), predictive data of a physical entity, process and / or service associated with one or more of the other SCF cells (2) and / or data relating to the state, processes or services of the SCF system stored in a repository external to the SCF system (1), and an untrained or partially trained digital twin model (6) through the intermediate communication module (4) and / or the global storage module (3), while at least one of the other SCF cells (2) is in operation in the system (1);(i) fully or partially training, in the second SCF cell, the digital twin model received in step h) using as training data the operating or performance data and / or the predictive data received in step h), and / or operating or performance data generated in the second cell itself, from the sensors included therein; (j) transmitting information about the digital twin model trained in step i) to the global storage module from the second SCF cell through the intermediate communication module;

[0050] In another embodiment of the method of the invention, this comprises the additional performance of the following steps in any technically possible order: k) executing the digital twin model received in step h) using as input data operating or performance data received in step h), and / or operating or performance data generated in the second cell itself, obtaining second predictive data; and l) transmitting, from the second SCF cell, the second predictive data obtained in step k) to the global storage module through the intermediate communication module and / or to a repository external to the SCF system connected to said SCF system.In another embodiment of the method of the invention, this comprises the additional performance of the following steps in any technically possible order: m) executing the digital twin model trained in step i) using as input data operating or performance data received in step h), and / or operating or performance data generated in the second cell itself, obtaining third predictive data; and n) transmitting, from the second SCF cell, the third predictive data obtained in step m) to the global storage module through the intermediate communication module and / or to a repository external to the SCF system connected to said SCF system.

[0051] Finally, in another embodiment of the method of the invention, this comprises the iterative execution of steps a) - n) in different CPS cells, thus updating:

[0052] - the generated and / or trained digital twin model(s) transmitted to the global storage module;

[0053] - the operating or performance data of one or more SCF cells stored in the global storage module and / or in an external repository connected to the SCF system; and

[0054] - the predictive data set stored in the global storage module and / or in an external repository connected to the SCF system.

[0055] For the purposes of this invention, the term "SCF System" shall be understood as a networked, intelligent system with onboard sensors, processors, and actuators, designed to monitor and interact with the physical environment (including human users) and offering guaranteed, real-time performance in safety-critical applications. "SCF System" shall also be understood not only as the physical entity corresponding to said system, but also as the operations, processes, and services associated with it. Likewise, "part of an SCF system" or "SCF subsystem" shall be understood as a subset of the physical entities, processes, operations, and / or services that comprise said SCF system.

[0056] For the purposes of this invention, the term “digital twin model” shall be understood as a computational model corresponding to a virtual copy or software of physical entities and / or processes. These digital copies are created and configured such that, upon receiving the same input stimuli or data as the physical object or process of which they are twins, they provide the same responses or output data as the equivalent physical object or process and can therefore be used in various applications within a SCF such as monitoring, operation and offering of certain services to external users. These applications include fault prediction, structural health prediction, cyber-attack prediction, prediction of physical and functional health conditions, etc.

[0057] Likewise, within the scope of interpretation of the invention, the term "real-time update" shall be understood as that update (which may be extended to creation, improvement or execution) of a digital twin model that occurs in a time shorter than the pre-established execution deadlines for the operations of the SCF system. Traditionally, the implementation of these systems has been approached using real-time system planning techniques and programming mechanisms that assume that the software used has extraordinarily low execution times. This classic assumption is tolerable if applied exclusively to the lowest software levels (or close to the hardware level) and to real-time operating systems, since the operating times of the primitives (or more basic operations offered by the low-level software or by the operating system) are on the order of a few microseconds.However, currently, using only a real-time operating system and drivers is unrealistic to deliver the functionality expected in a SCF. Currently, many of these systems have components that integrate communication middleware, web interfaces, backends, real-time web, AI / ML libraries, containerization and packaging software frameworks, and even blockchain technology. As a result, a higher level of abstraction is needed to address temporal behavior and real-world knowledge about the entire software stack that has grown in complexity. The present invention, through the use of the proposed system and method, addresses these limitations of the state of the art by taking this temporal requirement into account.

[0058] DESCRIPTION OF THE DRAWINGS

[0059] The above and other features and advantages will be more fully understood from the detailed description of the invention, as well as from the preferred embodiments referred to in the attached drawings, in which:

[0060] Figure 1 schematically shows an SCF system according to an embodiment of the present invention. Figure 2 schematically shows the actions or steps that may be performed during the implementation of a digital twin model update method in an SCF system according to an embodiment of the present invention.

[0061] Figure 3 schematically shows the structure and components of the middleware designed to perform a digital twin model update method in an SCF system according to an embodiment of the present invention.

[0062] Figure 4 shows the various elements involved in the design, deployment and execution of a smart contract according to an embodiment of the present invention.

[0063] NUMERICAL REFERENCES OF THE DRAWINGS

[0064] DETAILED DESCRIPTION OF THE INVENTION

[0065] A detailed description of the invention is set forth below based on Figures 1-4 herein. This description is provided for illustrative, but not limiting, purposes of the claimed invention. Figure 1 schematically represents a distributed SCF cyber-physical system (1) according to a preferred embodiment of the invention. Said system (1) comprises a plurality of SCF cells (2), each specialized in a task or operation within the system. Figure 1 shows, as non-limiting examples, four SCF cells (2) oriented to different purposes among which are the management of mechanical actuators, the control of assembly systems, monitoring or participation in communication networks, etc.

[0066] Each SCF cell (2) is equipped with sensors for collecting data related to the environment and operation of the SCF cell (2), such as operating data of the devices included in said SCF cell (2). Additionally, each SCF cell (2) has local processing and storage means (i.e., included or associated with the cell itself).

[0067] The system (1) also comprises a global data storage module (3) and an intermediate data communication module (4), both connected to each other and the intermediate communication module (4) also being connected to each of the SCF cells (2). The intermediate communication module (4) comprises processing means that allow data to be manipulated and transformed in such a way that secure transfers (in the sense of verifiable by the different elements of the SCF system) of information can be carried out between the SCF cells (2) and the global storage module (3).

[0068] In the system configuration (1) of Figure 1 , therefore, each SCF cell (2) can operate individually, carrying out its corresponding processes. However, the SCF cells (2) can also operate jointly thanks to the global storage module (3) acting as a collaborative space. In this way, the data collected by the sensors of the SCF cells (2) and the data processed by the processing means comprised in said SCF cells (2) can be transferred from one SCF cell (2) to another in a verifiable manner through the global storage space (3). Additionally, the SCF cells (2) can be connected to each other directly, for example, through a communications network (5) as shown in Figure 1. This network (5) may also be connected to the Internet. In this sense, 5G and 6G networks represent an advantageous scenario for the implementation of this type of communications.The possibility of operating both individually and collaboratively by the SCF cells (2) allows to develop, update and improve specific models, and in particular digital twin models (6) of the complete SCF system (1), without stopping the operation of said system (1). For this purpose, the processing means comprised in each SCF cell (2) are provided with software configured with one or more artificial intelligence algorithms adapted to:.

[0069] • generating one or more digital twin models (6) of at least a part of the SCF system (1);

[0070] • transmit a digital twin model (6) to the global storage module (3);

[0071] • receive a digital twin model (6) stored in the global storage module (3);

[0072] • execute a digital twin model (6) from the operating or performance data of one or more SCF cells (2) of the system (1) obtaining predictive data;

[0073] • train, in whole or in part, a digital twin model (6) using operational or performance data from one or more SCF cells (2) and / or predictive data as training data;

[0074] • transmitting operational or performance data of one or more SCF cells (2) and / or predictive data generated by the execution of a digital twin model (6) to the global storage module and / or to a repository external to the SCF system connected to said SCF system (3); and

[0075] • download operational or performance data from one or more SCF cells (2), predictive data generated by the execution of a digital twin model (6) stored in the global storage module and / or data relating to the state, processes or services of the SCF system stored in a repository external to the SCF system connected to said SCF system (3).

[0076] In this way, an SCF cell (2) can be responsible for generating, executing and / or improving / training a digital twin model (6) from its own data (collected by its sensors) and / or data from other SCF cells (2) without needing to stop the operation of the rest of the SCF cells (2). These processes being carried out in different SCF cells (2) and at different times, and thanks to the communication enabled by the global storage module (3), the digital twin models (6) are updated and improved in a collaborative manner without significantly interfering with the operation of the complete SCF system (1). The digital twin models (6) generated, improved and / or executed by each SCF cell (2) can be associated with the complete SCF system (1) or with a subsystem (part) thereof and can be intended for various applications, specifically monitoring, operation and third-party services.For example, an SCF system (1) implemented in a heavy vehicle manufacturing plant where there are distributed production cells would comprise digital twin models (6) corresponding to the three categories described:.

[0077] - Monitoring: for example, to predict the failure of an engine turbine or an error in an assembly machine.

[0078] - Operation: for example, to transport a series of large items to certain destinations via the safest routes and with curves of less than a certain angle.

[0079] - Third-party services: for example, to provide real-time production information and a launch forecast for a particular unit.

[0080] On the other hand, an SCF system (1) implemented in a hospital could equally comprise digital twin models (6) corresponding to the three categories described:

[0081] - Monitoring: for example, to predict cyberattacks against infrastructure.

[0082] - Operation: for example, for the instrumental operation performed in a surgical intervention.

[0083] - Third-party services: for example, for predicting the risk of type 2 diabetes in patients who submit their data remotely.

[0084] Preferably, the artificial intelligence algorithms used by the SCF cells (2) will correspond to neural networks that can be deep. Depending on the application or intrinsic properties of the system (1) in question, such as the type of sensors included in the SCF cells (2), a specific type of neural network will be used, optimal for the modeling of said SCF system (1). For example, if the data collected by the sensors of the SCF cells (2) correspond mainly to images, it is convenient to use convolutional neural networks, since these are optimized for the analysis of said images.

[0085] On the other hand, the global storage system (3) is configured, in a preferred embodiment of the invention, to host a blockchain-type data structure. This allows the storage of data and models to be decentralized, which can be collected mainly in the local storage media with which each SCF cell (2) is provided. Thus, when an SCF cell (2) requests the reception of data stored in another SCF cell (2), the global storage module (3) acts as a mediator of said transaction, enabling the certification of the transmission of information in a secure manner by the SCF cells (2). This is implemented thanks to the configuration of the intermediate communication module (4), intended to manage the transmissions and receptions of information carried out in each SCF cell (2) as transactions of a smart contract (7) executable by the global storage module (3).Alternatively, data and information may be stored in whole or in part in the global storage module (3).

[0086] Another relevant advantage of implementing blockchain technologies in the global storage module (3) is that it allows for the verifiable recording of an ordered history of non-redundant versions of information, in this case referring to different versions of the digital twin model(s) (6) of the SCF system (1) and to different sets of data collected by the sensors of the SCF cells (2) or generated in the execution of the digital twin models (6). Figure 1 shows examples of an untrained model (6), two versions of a partially trained model (6) and three versions of a fully trained model (6), all stored in the global storage module (3).

[0087] Figure 2 schematically shows the steps to follow in a method for updating digital twin models (6) of an SCF system (1) as described in previous paragraphs. Said method involves performing the following three basic operations: data transfer between the SCF cells (2) and the global storage module (3) through the intermediate communication module (4); management (creation, training and / or execution) of one or more digital twin models (6) by the SCF cells (2); and transfer of digital twin models (6) between the SCF cells (2) and the global storage module (3) through the intermediate communication module (4).

[0088] Thus, in a first stage of the method, the SCF cells (2) are in operation operating under the requirements of the complete SCF system (1) and collecting performance and / or operation data through the sensors they comprise. In a second stage, while the rest of the SCF cells (2) continue their operation normally, one or more SCF cells (2) generate a digital twin model (6) of the complete SCF system (1) or of an SCF subsystem and transfer it to the global storage module (3). Before performing the transfer, optionally, the SCF cells (2) can: receive data stored in the global storage module (3); fully or partially train the generated digital twin model (6) with the received data or with performance or operation data collected by the sensors of the SCF cell (2) itself; and transmit the trained digital twin model (6) to the global storage module (3).

[0089] Additionally, before transmitting the trained digital twin model (6), the SCF cells (2) can also execute said trained model (6), obtaining predictive data of the behavior of the process replicated by the GD in the real SCF system (1), which can also be transmitted to the global storage module (3).

[0090] Once one or more digital twin models (6), trained or not, have been stored (i.e., transmitted from an SCF cell (2)) in the global storage module (3), a third step of the method comprises collaboratively improving said models (6) in a manner similar to that described. In this case, the SCF cells (2) may receive a digital twin model (6) from the global storage module (3), train it fully or partially with data stored in the global storage module (3), with the data collected by the sensors of the SCF cell (2) and / or with data relating to the state, processes or services of the SCF system received from an external repository connected to the SCF system and transmit the new version of the trained model (6) to the global storage module (3).Again, optionally, before transmitting the new version of the trained digital twin model (6), the SCF cells (2) may also execute said version of the model (6), obtaining new predictive data of the behavior of the SCF system (1) that may also be transmitted to the global storage module (3).

[0091] By repeating the different stages of the method by one or more SCF cells (2) in an iterative manner, the collaborative creation and improvement of digital twin models (6) of the entire SCF system (1) is achieved without the need to significantly interfere with its operation.

[0092] In order to carry out the different stages of the method, a specific middleware (8) has been designed that can be used by the SCF cells (2) of the SCF system (1) and acts as an intermediary between said SCF cells (2) and the global storage system (3) based on blockchain. To this end, the design of said middleware (8) has been based on the use of a specialized language (such as Solidity), while maintaining its adaptability to the needs of the development process of the middleware (8), since it can be efficiently interpreted in other programming languages ​​such as JavaScript, Java, etc. To this end, the design of the middleware (8) is based on a unified modeling language (UML) that contains the classes that can be used and reproduced by other languages ​​and scenarios.

[0093] The middleware (8) designed to manage the processes of the SCF system (1) of the invention is provided with four modules, each of said modules being associated with one of the four main processes involved: data management (input data observation module (9)), execution of the digital twin models (6) (model execution module (10)); training of the digital twin models (6) (model training module (11)) and interaction with the blockchain (trusted storage module (12)). The characteristics of the different modules (9, 10, 11, 12) are set out below.

[0094] Input data observation module (9): is responsible for collecting the data captured by the SCF cell sensors (2) and the predictive data generated in the execution of the digital twin models (6). From these, the input data observation module (9) prepares data sets that serve as training data for other models (6) or versions of digital twin models (6). It also optionally collects a set of training parameters that refer to the training options of the neural networks, such as the speed of the process or the activation function of the neural network in question.

[0095] Model execution module (10): is the entity that manages the execution of a version of a fully or partially trained digital twin model (6). To do this, the model execution module (10) has implemented an execution function that acts on the version of the digital twin model (6) to be executed. This function receives as input data the data collected by the sensors of the SCF cell (2) and provides, as output data, predictive data on the behavior of the SCF system (1). Model training module (11): is the one that allows any SCF cell (2) to train a digital twin model (6) using as training data a data set prepared or obtained by the Input data observation module (9).When a digital twin model (6) is to be trained in an SCF cell (2), said SCF cell (2) sends a request to the model training module (11) indicating the digital twin model (6) to be trained, the specific version of said model, the training data and the training parameters. These last parameters define how this training will be carried out, for example, they determine aspects such as the speed of the training and / or the precision of the result obtained, among others.

[0096] Reliable storage module (12): is responsible for managing the interaction of the SCF cells (2) with the blockchain implemented in the global storage module (3) to receive and transmit data and digital twin models (6), benefiting from their inherent properties and traceability. Furthermore, the reliable storage module (12) allows retrieving versions of the digital twin model(s) (6) from the blockchain to execute them within any SCF cell (2). For all this, the main task of the reliable storage module (12) is the elaboration of smart contracts (7) for execution in the global storage space (3). An erroneous smart contract (7) can cause an information transaction to not be executed or not validated and, therefore, not linked to the blockchain.Therefore, the reliable storage module (12) ensures that the smart contract (7) is correctly developed and that it conforms to the structure of the digital twin models (6) based on neural networks, in order to generate valid transactions that can be subsequently validated by any SCF cell (2).

[0097] In this sense, the present invention also contemplates the design of specific smart contracts (7) for implementation in the SCF system (1) described. Said smart contracts (7) are designed to be used in any standard programming language, thus facilitating the integration of the SCF system (1) of the invention into any field of application or pre-existing system.

[0098] In the SCF system (1) of the invention, any trained or untrained model (6), i.e. any version of a digital twin model (6), can be retrieved from the blockchain by issuing a transaction on the managing smart contract (7) where the corresponding information has been stored. Figure 4 shows how smart contracts (7) are used in the blockchain as a synchronization point to build and update digital twin models (6) in this distributed context. The smart contract (7) corresponding to such transactions must contain the data fields and functions necessary to represent the model (6) truthfully. When a request to execute the smart contract (7) is received, the infrastructure executes the requested function. Such function may be a simple read operation to retrieve stored information or it may be a function that requires more transactions.Smart contracts (7) are redesigned in the medium term each time an action is performed on a version of a digital twin model (6), that is, each time said model (6) is executed for the prediction or classification of elements of the SCF system (1) or each time the model (6) is trained.

[0099] Finally, the viability and functionality of the SCF system (1) and the digital twin model update method (6) of the invention have been tested in a real blockchain network by carrying out a set of exhaustive experiments that extract the temporal behavior of the different phases involved in the interaction with the blockchain network.

[0100] The digital twin models (6) produced and trained by the SCF cells (2) comprised in the SCF system (1), based on neural networks and used for prediction purposes, manage more than twenty input measurements producing prediction values ​​of the behavior of the SCF system (1). Different measurements and data shared with thousands of training samples are used for training.

[0101] The results obtained in this context demonstrate that the implementation of the method described above in an SCF system (1) according to a preferred embodiment of the invention is effective and efficient when updating and executing digital twin models (6) in said system without requiring the pause of its operations. Said implementation shows values ​​of 48 ms and 21.7 ms corresponding to the preprocessing and postprocessing of the representation of the model (6), respectively. Access to the blockchain to receive a model (6) by an SCF cell (2) is around 390 ms on average, while transmitting and storing a model (6) requires 1.358 s on average with an additional average time of 6.74 s associated with the consensus time of the blockchain network.On the other hand, the proposed method and system turn out to be efficient when managing simultaneous requests for interaction of several SCF cells (2) with the blockchain, providing stable execution times for all types of operations in the global storage space (3), ie, in the collaborative space for training and improving digital twin models (6) in the blockchain network.

Claims

CLAIMS 1.- Cyber-physical system (1), SCF, distributed that includes: - a plurality of SCF cells (2), where each of said SCF cells (2) comprises, in turn: • one or more sensors adapted to obtain operating or performance data from the SCF cell (2); • first means of data processing; • a first means of transmitting and receiving data; • first data storage media; - an intermediate communication module (4), communicatively connected to the plurality of SCF cells (2) and adapted with second data processing means; and with second means for transmitting and receiving said data; - a global data storage module (3), communicatively connected to the intermediate communication module (4), and adapted to store the data of the SCF cells (2); and said system (1) being characterized in that: the first processing means and the first means of transmitting and receiving data from each SCF cell (2) are configured with one or more artificial intelligence algorithms, adapted to: • generating a digital twin model (6) of at least a part of the SCF system (1); • transmit information from a digital twin model (6) to the global storage module (3); • receive information from a digital twin model (6) stored in the global storage module (3); • execute a digital twin model (6) from the operating or performance data of one or more SCF cells (2) of the system (1) obtaining predictive data; • train, in whole or in part, a digital twin model (6) using operational or performance data from one or more SCF cells (2) and / or predictive data as training data; • transmitting operational or performance data of one or more SCF cells (2) and / or predictive data generated by the execution of a digital twin model (6) to the global storage module and / or to a repository external to the SCF system (1) connected to said SCF system (1); and • downloading operational or performance data from one or more SCF cells (2) stored in the global storage module (3), predictive data generated by the execution of a digital twin model (6) stored in the global storage module (3) and / or data relating to the state, processes or services of the SCF system (1) stored in a repository external to the SCF system (1) connected to said SCF system (1). 2.- Cyber-physical system (1), SCF, according to the previous claim, where, additionally, the global storage module (3) and the first data storage means are configured to house a blockchain type data structure, storing consecutive and non-redundant versions of one or more digital twin models (6). 3.- Cyber-physical system (1), SCF, according to any of the preceding claims, wherein, additionally, the second processing means comprise software means configured to: - manage the transmissions and receptions of digital twin models (6) and / or data made in each SCF cell (2) as transactions of a smart contract (7) executable by the global storage module (3); and - collect operational or performance data from one or more SCF cells (2) and / or predictive data and use them to create one or more training data sets. 4.- Cyber-physical system (1), SCF, according to any of the preceding claims, wherein the intermediate communication module (4) is included in the global storage module (3). 5.- Cyber-physical system (1), SCF, according to any of the preceding claims, wherein the algorithm(s) comprised by each SCF cell (2) comprise a neural network. 6.- Cyber-physical system (1), SCF, distributed according to the previous claim, where the neural network is deep, feedforward, perceptron, multilayer perceptron, convolutional and / or recurrent. 7.- Cyber-physical system (1), SCF, according to any of the previous claims, where the connection between the global storage module (3) and the intermediate communication (4) and / or the communication between the intermediate communication module (4) and each SCF cell (2) is remote, belonging to the same telecommunication network, and where said telecommunication network is wired, wireless or mixed. 8.- Cyber-physical system (1), SCF, according to any of the preceding claims, wherein each SCF cell (2) is connected to the rest of the SCF cells (2) through a telecommunication network (5). 9.- Method for updating in real time a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1), according to any of the preceding claims, which comprises carrying out the following steps in any technically possible order: a) receiving, in a first SCF cell (2): - operating or performance data of one or more of the other SCF cells (2) via the intermediate communication module (4) and / or the global storage module (3) - predictive data of a physical entity, process and / or service associated with one or more of the other SCF cells (2) via the intermediate communication module (4) and / or the global storage module (3); and / or - data relating to the status, processes or services of the SCF system stored in a repository external to the SCF system (1) connected to said SCF system (1); while at least one of said other SCF cells (2) is in operation in the system (1); b) generating, in the first SCF cell (2), a digital twin model (6) of at least a part of the SCF system (1); c) transmitting information about the digital twin model (6) generated in step b) to the global storage module (3) from the first SCF cell (2) through the intermediate communication module (4). 10.- Method for updating in real time a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1) according to the previous claim, which comprises the additional execution of the following steps in any technically possible order: d) training totally or partially, in the first SCF cell (2), the digital twin model (6) generated in step b) using as training data: - the operating or performance data received in step a); - the operating or performance data generated in the first SCF cell (2) itself, from the sensors included therein; and / or - the data stored in a repository external to the SCF system (1) connected to said SCF system (1) obtained in step a); and e) transmitting information about the digital twin model (6) trained in step d) to the global storage module (3) from the first SCF cell (2) through the intermediate communication module (4). 11.- Method for updating in real time a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1), according to the preceding claim, comprising the additional performance of the following steps in any technically possible order: f) executing the digital twin model (6) fully or partially trained in step d) using as input data operating or performance data received in step a), and / or operating or performance data generated in the first SCF cell (2) itself, obtaining first predictive data; and g) transmitting, from the first SCF cell (2), the first predictive data obtained in step f) to the global storage module through the intermediate communication module (4) and / or to a repository external to the SCF system connected to said SCF system. 12.- Method for updating in real time a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1), according to any of claims 9-11, which additionally comprises carrying out the following steps in any technically possible order: h) receiving, in a second SCF cell (2), operating or performance data of one or more of the other SCF cells (2), predictive data of a physical entity, process and / or service associated with one or more of the other SCF cells (2) and / or data relating to the state, processes or services of the SCF system stored in a repository external to the SCF system (1), and an untrained or partially trained digital twin model (6) through the intermediate communication module (4) and / or the global storage module (3), while at least one of the other SCF cells (2) is in operation in the system (1);i) fully or partially train, in the second SCF cell (2), the digital twin model (6) received in step h) using as training data the data of; operation or performance and / or the predictive data received in step h), and / or operation or performance data generated in the second SCF cell (2) itself, from the sensors included therein; j) transmitting information about the digital twin model (6) trained in step i) to the global storage module (3) from the second SCF cell (2) through the intermediate communication module (4). 13.- Method for updating in real time a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1), according to the preceding claim, which additionally comprises carrying out the following steps in any technically possible order: k) executing the digital twin model (6) received in step h) using as input data operating or performance data received in step h), and / or operating or performance data generated in the second SCF cell (2) itself, obtaining second predictive data; and l) transmitting, from the second SCF cell, the second predictive data obtained in step k) to the global storage module through the intermediate communication module and / or to a repository external to the SCF system connected to said SCF system.

14. Method for real-time updating of a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1), according to any of claims 12-13, which additionally comprises carrying out the following steps in any technically possible order: m) executing the digital twin model (6) trained in step i) using as input data operating or performance data received in step h), and / or operating or performance data generated in the second SCF cell (2) itself, obtaining third predictive data; and n) transmitting, from the second SCF cell, the third predictive data obtained in step m) to the global storage module through the intermediate communication module and / or to a repository external to the SCF system connected to said SCF system. 15.- Real-time update method of a digital twin model (6) of a cyber-physical system (1), SCF, and / or of a process and / or service offered by / executed in said SCF system (1), according to the previous claim, where steps a) - n) are repeated iteratively in different SCF cells (2) updating: - the digital twin model(s) (6) generated and / or trained transmitted to the global storage module (3); - the operating or performance data of one or more SCF cells (2) stored in the global storage module (3) and / or in an external repository connected to the SCF system; and - the predictive data set stored in the global storage module (3) and / or in an external repository connected to the SCF system.

Images