Dynamic traffic patterns and shifts in traffic distribution in Open Radio Access Networks (O-RAN) pose a significant challenge for real-time network optimization in 5G and beyond. Traditional traffic analytics methods struggle to remain accurate under such non-stationary conditions, where models trained on historical data quickly degrade as traffic evolves. This paper introduces AIDITA, an AI-driven Digital Twin for Traffic Analytics framework designed to solve this problem through autonomous model adaptation. AIDITA creates a digital replica of the live analytics models running in the RAN Intelligent Controller (RIC) and continuously updates them within the digital twin using incremental learning. These updates use real-time Key Performance Metrics (KPMs) from the live network, augmented with synthetic data from a Generative AI (GenAI) component to simulate diverse network scenarios. Combining GenAI-driven augmentation with incremental learning enables traffic analytics models, such as prediction or anomaly detection, to adapt continuously without the need for full retraining, preserving accuracy and efficiency in dynamic environments. Implemented and validated on a real-world 5G testbed, our AIDITA framework demonstrates significant improvements in traffic prediction and anomaly detection use cases under distribution shifts, showcasing its practical effectiveness and adaptability for real-time network optimization in O-RAN deployments.

With the continuous evolution of digital twins, the requirements of interconnection and interoperability have led to the creation of the term Distributed Digital Twin, where commonly, different components of the same digital twin operate on different devices. This article addresses this new gap in the field by combining Digital Twins and Distributed systems technologies and introduces a re-definition of the architecture of OpenTwins, an open-source platform designed to develop generic next-gen 3D-IoT-AI-powered digital twins. This approach enables the distribution of digital twins across different infrastructures by seamlessly integrating multiple instances of OpenTwins working together. Distributing Digital Twins across networks and devices can offer dynamic and collaborative simulations, artificial intelligence techniques, yet poses synchronization and scalability challenges. The platform re-definition involves the definition of a lightweight, synchronized, and distributed version of the original OpenTwins architecture to tackle these issues. As the field of digital twins is closely linked to the Internet of Things environment and Industry 4.0, this architecture has been designed to be compatible in IoT devices, being compatible with ARM architectures, consuming fewer resources than the original platform as it has fewer components, thus reducing the number of messages and the bandwidth consumed.

Machine learning data privacy has been improved with Federated Learning approaches. However, some obstacles to guaranteeing traceability, openness, and participant contribution incentives prevent its widespread use. In this study, Ethereum blockchain technology is integrated into the data stream Kafka-ML framework, presenting a novel asynchronous and blockchain-based Federated Learning approach. By utilising Ethereum for transparent and auditable participant tracking, this integration overcomes some shortcomings such as auditability and model sharing reliability. Furthermore, Ethereum smart contracts allow for automatic reward distribution systems, which promote equitable incentive systems and increased involvement in the Federated Learning process. To demonstrate its potential, an extensive evaluation has been carried out on a wireless net-work technology detection use case. By improving transparency, traceability, and incentive structures of Federated Learning, it is expected to strengthen the robustness of flexible machine learning collaboration with data streams.

With the increasing development and deployment of new systems for efficient and clean mobility, Electric Vehicles (EVs) are becoming more and more common among people. Those produce large amounts of data streams that need to be collected and analyzed to understand user needs and improve their performance. For this purpose, Artificial Intelligence (AI) techniques are playing a very important role. Within this context, Kafka-ML is a Machine Learning (ML) framework that enables the consumption and processing of data streams and allows the flexible management and deployment of neural networks throughout their entire life cycle. Kafka-ML can work with Distributed Neural Networks (DNN) which reduce latency and response times, perform incremental training over time allowing models to adapt to data on the fly, and carry out Federated Learning (FL) processes for this type of algorithms so a more robust global model can be created while maintaining data privacy and security, but all this separately. This work has considered the joint implementation of FL, for anonymous data sharing, incremental learning for continuous training of the models, and DNN for distribution of the models across different points on the map. All this applied within a Vehicle-to-everything (V2X) domain where EV usage and charge data can be shared to improve the user experience, as well as to better understand the behavior of this type of vehicles and their charging points to achieve savings, and how it affects people daily lives. An evaluation of the system related to this EV use case is presented to demonstrate the viability of the tool.

Abstract Machine Learning (ML) applications need large volumes of data to train their models so that they can make high‐quality predictions. Given digital revolution enablers such as the Internet of Things (IoT) and the Industry 4.0, this information is generated in large quantities in terms of continuous data streams and not in terms of static datasets as it is the case with most AI (Artificial Intelligence) frameworks. Kafka‐ML is a novel open‐source framework that allows the complete management of ML/AI pipelines through data streams. In this article, we present new features for the Kafka‐ML framework, such as the support for the well‐known ML/AI framework PyTorch, as well as for GPU acceleration at different points along the pipeline. This pipeline will be described by taking a real Industry 4.0 use case in the Petrochemical Industry. Finally, a comprehensive evaluation with state‐of‐the‐art deep learning models will be carried out to demonstrate the feasibility of the platform.

A pipeline of constant data streams is being built by the Internet of Things (IoT) to monitor information about the physical environment. In parallel, Artificial Intelligence (AI) is constantly developing and enhancing industrial, economic, and academic endeavors as well as quality of life thanks to these IoT data. In streaming contexts, Kafka-ML is our open-source framework that enables the management of Machine Learning (ML) and AI pipelines over data streams. Accordingly, it simplifies the deployment of Deep Neural Networks (DNNs) in practical applications. Nonetheless, this framework did not support the possibility of carrying out an Online Learning (OL) process, which is needed when new data are continuously arriving, and the models need to adapt to them on the fly. In this work, we have extended our previous work, the Kafka-ML framework, to enhance the management of ML/AI pipelines with OL features to enable both ML/AI distributed and centralized models to learn indefinitely over time. These models are continuously upgraded thanks to a process where automatic and flexible inference is carried out when improvements in the model performance are achieved. This opens up a large number of new possibilities within different fields of application, development, and work under the premise of incremental learning with ML models such as Electrical Vehicles and Industry 5.0. We have validated these new features by adapting and deploying state-of-the-art DNN models in different online scenarios, for both single and distributed configurations. The results show the capability of Kafka-ML to execute effective online training processes for ML models, improving their performance over time as new data becomes available.

Abstract The use of serverless computing has been gaining popularity in recent years as an alternative to traditional Cloud computing. We explore the usability and potential development benefits of three popular open‐source serverless platforms in the context of IoT: OpenFaaS, Fission, and OpenWhisk. To address this we discuss our experience developing a serverless and low‐latency Distributed Deep Neural Network (DDNN) application. Our findings indicate that these serverless platforms require significant resources to operate and are not ideal for constrained devices. In addition, we archived a 55% improvement compared to Kafka‐ML's performance under load, a framework without dynamic scaling support, demonstrating the potential of serverless computing for low‐latency applications.

Abstract The realm of digital twins is experiencing rapid growth and presents a wealth of opportunities for Industry 4.0. In conjunction with traditional simulation methods, digital twins offer a diverse range of possibilities. However, many existing tools in the domain of open‐source digital twins concentrate on specific use cases and do not provide a versatile framework. In contrast, the open‐source digital twin framework, OpenTwins, aims to provide a versatile framework that can be applied to a wide range of digital twin applications. In this article, we introduce a re‐definition of the original OpenTwins platform that enables the management of custom simulation services and the management of FMI simulation services, which is one of the most widely used simulation standards in the industry and its coexistence with machine learning models, which enables the definition of the next‐gen digital twins. Thanks to this integration, digital twins that reflect reality better can be developed, through hybrid models, where simulation data can feed the scarcity of machine learning data and so forth. As part of this project, a simulation model developed through the hydraulic software Epanet was validated in OpenTwins, in addition to an FMI simulation service. The hydraulic model was implemented and tested in an agricultural use case in collaboration with the University of Córdoba, Spain. A machine learning model has been developed to assess the behavior of an FMI simulation through machine learning.

Currently, wind energy plays a fundamental role in the process of generating energy in a sustainable and environmentally friendly manner. However, their infrastructures require ongoing maintenance tasks that involve considerable risk. This is why a predictive maintenance system for the surface inspection of wind turbine blades based on machine learning techniques has been developed. Specifically, convolutional neural networks have been applied to detect and classify turbines and their blades, as well as the surface defects that may appear on them. The system comprises a mobile application that makes use of a telescope to take pictures with certain precision, a computing edge node responsible for processing the images that are captured, and a motorized mount that allows the telescope to move. The objective of this open-source project is to detect and classify different surface defects on the blades of wind turbines and carry out the maintenance of these infrastructures. The system is responsible for undertaking a complete sweep of the surface of the turbine blades in an autonomous way and finally presents the defects found to the user. The deep neural networks also help the system to decide which movements the motorized mount has to make together with the telescope to perform the inspection. Accuracies of around 97% for label predictions and 90% for bounding box coordinate predictions have been achieved for the convolutional deep learning models. Two possible approaches have been considered for the project: the first is to carry out all the necessary computation on a mobile phone to have a portable solution, and the second option considers a edge node to balance the load and thus not overload the mobile device. Tests show that the edge node approach gives better results overall. The proposed system for detecting surface damage on blades was experimentally validated on a wind farm.

Machine Learning (ML) and Artificial Intelligence (AI) depend on data sources to train, improve, and make predictions through their algorithms. With the digital revolution and current paradigms like the Internet of Things, this information is turning from static data to continuous data streams. However, most of the ML/AI frameworks used nowadays are not fully prepared for this revolution. In this paper, we propose Kafka-ML, a novel and open-source framework that enables the management of ML/AI pipelines through data streams. Kafka-ML provides an accessible and user-friendly Web user interface where users can easily define ML models, to then train, evaluate, and deploy them for inferences. Kafka-ML itself and the components it deploys are fully managed through containerization technologies, which ensure their portability, easy distribution, and other features such as fault-tolerance and high availability. Finally, a novel approach has been introduced to manage and reuse data streams, which may eliminate the need for data storage or file systems.

Considering the recent transition towards renewable energy sources such as off-shore wind turbines, solar farms, and hydroelectric power plants, Structural Health Monitoring (SHM) of these novel infrastructures using Machine Learning (ML) methods has become extremely attractive. However, the strong dependence of energy-thirsty ML approaches on cloud computations limits their application at the edge, which is significantly important for SHM in remote locations. Therefore, development of edge-oriented machine learning models and their integration with edge-computing technologies such as neuromorphic platforms is vital for the real-time and on-site processing of sensory signals without cloud computations for SHM. Therefore, the objective of this work was to develop a neuromorphic-compatible ML model for time-series analysis of accelerometer data acquired from a wind turbine gearbox for fault detection purposes. The hardware-friendly model in this work provided an accuracy of 83.2% for the recognition of healthy and damaged gearboxes, providing promising results for off-cloud SHM using neuromorphic-computing technologies.

The current dependency of Artificial Intelligence (AI) systems on Cloud computing implies higher transmission latency and bandwidth consumption. Moreover, it challenges the real-time monitoring of physical objects, e.g., the Internet of Things (IoT). Edge systems bring computing closer to end devices and support time-sensitive applications. However, Edge systems struggle with state-of-the-art Deep Neural Networks (DNN) due to computational resource limitations. This paper proposes a technology framework that combines the Edge-Cloud architecture concept with BranchyNet advantages to support fault-tolerant and low-latency AI predictions. The implementation and evaluation of this framework allow assessing the benefits of running Distributed DNN (DDNN) in the Cloud-to-Things continuum. Compared to a Cloud-only deployment, the results obtained show an improvement of 45.34% in the response time. Furthermore, this proposal presents an extension for Kafka-ML that reduces rigidness over the Cloud-to-Things continuum managing and deploying DDNN.

In the last few years, the Internet of Things (IoT) has emerged as the new disruptive technology to change the world. Cloud computing has accompanied this field to overcome its processing and storage limitations. However, this evolution has originated a huge increase in IoT devices and data that will create a bottleneck for current networks, in addition to a lack of low latency in cloud communications. Edge computing has been developed to address this challenge, moving the processing to the edge of the network. In this paper, an edge computing architecture is presented to overcome these challenges. The architecture, based on our previous work on the λ-CoAP architecture, covers the whole vision of an edge computing deployment, from IoT devices, to the edge Smart Gateways and up to a cloud infrastructure.

In the Internet of Things (IoT) vision, conventional devices become smart and autonomous. This vision is turning into a reality thanks to advances in technology, but there are still challenges to address, particularly in the security domain e.g., data reliability. Taking into account the predicted evolution of the IoT in the coming years, it is necessary to provide confidence in this huge incoming information source. Blockchain has emerged as a key technology that will transform the way in which we share information. Building trust in distributed environments without the need for authorities is a technological advance that has the potential to change many industries, the IoT among them. Disruptive technologies such as big data and cloud computing have been leveraged by IoT to overcome its limitations since its conception, and we think blockchain will be one of the next ones. This paper focuses on this relationship, investigates challenges in blockchain IoT applications, and surveys the most relevant work in order to analyze how blockchain could potentially improve the IoT.

In the Internet of Things (IoT) a worldwide network of sensors and actuators transmit data and actuate over the Internet. Nevertheless, the deployment of sensors and actuators usually requires tools to program and configure them before they can work, and not everyone has such access. Current smartphones are provided with a large set of sensors and actuators that can be incorporated in the IoT. With the aim of achieving seamless integration, this paper presents SocICoAP, a system that enables the sharing of sensors and actuators present in smartphones and installing custom ones at run-time in microcontrollers. In this way smartphones are integrated into the IoT, not only a sensor providing data from its built-in sensors but also as a gateway for nearby deployed sensors. Furthermore, the sensor data can also be globally shared and analysed through a cloud computing integration.