Enhancement of Fog/Edge and Cloud based IIoT Architectures
Goiuri Peralta
06/11/2020
- DIRECTORS: Josu Bilbao and Pedro M. Crespo Bofill
- UNIVERSITY: Tecnun
ABSTRACT
Industry 4.0 reflects a technological evolution through the digitalization of industry and manufacturing. The key enabling technologies for this upcoming industrial transformation are mainly Cyber-Physical Systems (CPS) and the Internet of Things (IoT), and they allow the interconnection of millions of devices. When IoT technologies are applied to the industrial domain, Industrial Internet of Things (IIoT) arises, which covers Machine-to-Machine (M2M) and industrial communication technologies. The development of IIoT services is introducing stringent requirements in several industry sectors which require solutions that fully embrace this industrial transformation.
Besides applying novel and advanced technologies, it is crucial to deploy an architecture that supports the emerging IIoT requirements. Cloud-based data processing systems have leveraged the IIoT for scalable real-time data processing. However, cloud computing may not be capable of meeting the requirements of distributed and time-sensitive IIoT applications. In order to overcome those limitations fog computing is one of the emerging topologies, which extends cloud capabilities to the edge of the network. Taking this in mind, in this thesis we have analyzed the potential and capabilities of different industrial architectures or system structures. We have concluded that the deployment of a three-tier architecture (IoT-fog-cloud) is a promising solution since it can be better adapted to the needs of each application and provides flexibility and scalability to the network. Taking the three-tier architecture as reference, we have analyzed different tools and techniques, such as Network Coding (NC) and Machine Learning (ML) among others, and we have observed their potential in order to improve the properties of such systems in order to exploit the inherent advantages of the architecture itself.
The first contribution of the thesis has been to analyze the suitability of NC-based data encoding techniques for distributed storage systems that use low-price cloud services, in order to provide cost-efficiency and reliability. The results show that configurations that seek to minimize the storage yield a higher cost reduction, and that the overall cost decreases as more clouds, i.e. storage nodes, become available. Then, using again NC as the mechanism for data distribution, we have studied the required time to download the data stored in a fog-to-cloud architecture, where multiple nodes provide different performance parameters. To achieve the optimal data distribution that minimizes the overall data download time, we have developed an optimal distribution algorithm which considers system parameters. The conducted analysis prove that the proposed scheme outperforms other distribution strategies, even under poor connectivity conditions. These two contributions demonstrate the potential of NC techniques, which can be extrapolated to enhance other factors, such as the communication delay between different nodes in the system.
The next contribution of this thesis has been the application of ML-based data predictors in a fog-based scheme to enhance energy efficiency of IoT nodes. To do this, we have defined a communication model built over MQTT, which takes into account the energy expense per bit and the total number of bytes in the communication process as well as the accuracy degree of the predictive algorithm. We have demonstrated that the energy consumption decreases with the accuracy degree of the predictor and that with higher QoS levels, the proposed model is more cost-effective, in particular when larger messages are exchanged.
Finally, we have proposed the use of Homomorphic Encryption (HE) together with NC techniques in order to overcome IoT privacy issues, as well as to enhance the benefits that NC can provide over the entire IoT architecture. We have theoretically shown, based on the current state of the art, their applicability over the IoT architecture. We have also discussed the main strengths and limitations of the application of both technologies over this type of architectures.