Article
Computer Science, Artificial Intelligence
Antonino Ferraro, Antonio Galli, Vincenzo Moscato, Giancarlo Sperli
Summary: In recent years, the optimization of maintenance operations has been a major challenge in Industry 4.0. Various predictive maintenance frameworks have been developed to reduce maintenance costs and downtime intervals. However, these frameworks often rely on complex deep learning models, making it difficult for humans to understand their predictions. This paper focuses on using explainable artificial intelligence (XAI) methodologies to explain the predictions made by a recurrent neural network model for estimating the remaining useful life (RUL) of hard disk drives (HDDs). The results show that the SHapley Additive exPlanations (SHAP) method outperforms the local interpretable model-agnostic explanations (LIME) method, providing a suitable and effective solution for HDD predictive maintenance applications.
ARTIFICIAL INTELLIGENCE REVIEW
(2023)
Article
Engineering, Electrical & Electronic
Junjie Shi, Jing Du, Yingwen Ren, Boyu Li, Jinwei Zou, Anyi Zhang
Summary: This paper proposes an improved algorithm for early warning of mechanical hard disk failures, using feature selection, Generative Adversarial Networks (GAN), and Convolution-LSTM (C-LSTM) to enhance the data availability of the IaaS cloud platform.
JOURNAL OF SENSORS
(2022)
Article
Engineering, Civil
Mithun Mukherjee, Vikas Kumar, Qi Zhang, Constandinos X. Mavromoustakis, Rakesh Matam
Summary: In this paper, the task data offloading issue in edge-cloud computing systems is studied. By analyzing hard-deadline and soft-deadline tasks, as well as the average delay and service price of edge and cloud servers, an optimal task offloading policy is proposed to maximize the revenue of both edge and cloud servers. The equilibrium is reached through independent consideration of each task for suitable location offloading.
IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS
(2022)
Article
Automation & Control Systems
Jishan Ahmed, Robert C. Green II
Summary: This study proposes the use of long short term memory networks and a modified focal loss function to address the class imbalance issue in failure prediction. Comparing with traditional machine learning algorithms, the results demonstrate that this approach achieves better performance while maintaining a higher failure detection rate.
ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE
(2024)
Article
Chemistry, Analytical
Tek Raj Chhetri, Anelia Kurteva, Jubril Gbolahan Adigun, Anna Fensel
Summary: Hard drive failure can lead to system crashes and data loss. Current studies on hard drive failure prediction mostly rely on machine learning, lacking context-awareness. Semantic technology provides context-awareness through knowledge graphs, but lacks the learning patterns and prediction capabilities of machine learning. This paper presents a method that combines the advantages of both machine learning and semantic technology, achieving higher accuracy in hard drive failure prediction.
Article
Computer Science, Hardware & Architecture
Yinlong Li, Siyao Cheng, Hao Zhang, Jie Liu
Summary: With the increasing popularity of mobile devices and explosive growth of task processing requirements, edge computing attracts more attention. In a Mobile Edge Computing (MEC) network, the workload offloading problem is important, and efficient algorithms have been proposed. However, most existing algorithms are not suitable for dynamic scenarios with fast-moving devices. This paper proposes a dynamic adaptive workload offloading algorithm based on Lyapunov theories and an FC-LSTM based schedule determining algorithm to balance the workload and minimize energy and time consumption. Theoretical analysis and experiments show high performance of the proposed algorithms in energy consumption, convergence, and latency.
Article
Chemistry, Multidisciplinary
Mingyu Zhang, Wenqiang Ge, Ruichun Tang, Peishun Liu
Summary: This paper proposes a novel failure prediction method that combines machine learning algorithms and neural networks. Through experiments on the publicly available BackBlaze hard disk datasets, it demonstrates the superiority of this method in hard disk failure prediction and solves the problem of low robustness and generalization in traditional machine learning methods.
APPLIED SCIENCES-BASEL
(2023)
Article
Materials Science, Multidisciplinary
Akihiko Aoyagi, Barry Stipe, Roger Wood, Steven Campbell, Xiaodong Che
Summary: This paper describes the history and advantages of helium-sealed HDDs, and discusses the future prospects of sealed-drive technology.
JOURNAL OF MAGNETISM AND MAGNETIC MATERIALS
(2022)
Article
Computer Science, Information Systems
Chong Zheng, Shengheng Liu, Yongming Huang, Wei Zhang, Luxi Yang
Summary: This article proposes an unsupervised and privacy-preserving popularity prediction framework for MEC-enabled IIoT. It introduces the concepts of local and global popularities and models the time-varying popularity of each user. The framework achieves high prediction accuracy and addresses challenges like data deficiency, costly manual labels, and non-i.i.d. user behaviors.
IEEE INTERNET OF THINGS JOURNAL
(2022)
Article
Computer Science, Hardware & Architecture
Miguel A. Guillen, Antonio Llanes, Baldomero Imbernon, Raquel Martinez-Espana, Andres Bueno-Crespo, Juan-Carlos Cano, Jose M. Cecilia
Summary: The combination of IoT and AI is revolutionizing various economic sectors, but the gap between AI and IoT still exists, especially in rural areas where connectivity and power supply are limited. Edge computing is proposed as a solution to bridge this gap, offering new opportunities for scenarios where connectivity is a challenge.
JOURNAL OF SUPERCOMPUTING
(2021)
Article
Computer Science, Hardware & Architecture
Hongliang Zhou, Yifeng Zheng, Xiaohua Jia, Jiangang Shu
Summary: Edge computing facilitates networked services with fast responses and low bandwidth by deploying computing and storage at the network edge. However, EC servers lack security protections compared to centralized data centers, making them vulnerable to security attacks, especially DDoS attacks. In this paper, we propose CoWatch, a collaborative framework for prediction and detection of DDoS attacks in EC scenarios based on distributed SDN architecture. CoWatch can predict and detect DDoS attacks towards EC servers in a timely manner.
Article
Computer Science, Theory & Methods
Pedro Cruz, Nadjib Achir, Aline Carneiro Viana
Summary: Multi-Access Edge Computing (MEC) attracts attention due to its implications in science, technology, and commerce. However, existing MEC initiatives are incomplete, and understanding experimental prototypes and implementations is crucial. This study discusses and surveys existing MEC projects, comparing strategies, limitations, and tools while addressing unresolved issues in practice.
ACM COMPUTING SURVEYS
(2023)
Article
Computer Science, Artificial Intelligence
Ming Yan, XiaoMeng Liang, ZhiHui Lu, Jie Wu, Wei Zhang
Summary: With the rapid development of 5G network, business scenarios such as intelligent service and new retail are becoming popular, leading to an increased demand for flexible and scalable real-time data processing in edge computing. This paper proposes a HANSEL system based on Kubernetes platform, which optimizes the horizontal elastic scaling policy and improves system resource utilization through machine learning and deep learning algorithms.
APPLIED SOFT COMPUTING
(2021)
Article
Chemistry, Physical
Tomasz Skrzekut, Maciej Wedrychowicz, Andrzej Piotrowicz
Summary: The paper presents a comparison study on two methods of recycling aluminum from HDD platters – the melting method and the method of plastic consolidation. The study involved examining the composition and properties of HDD memory components, specifically the data carriers (platters), through various tests and analysis. The results provide insights into the performance characteristics of the materials under different recycling methods.
Article
Physics, Multidisciplinary
Pu Han, Lin Han, Bo Yuan, Jeng-Shyang Pan, Jiandong Shang
Summary: This paper proposes a trajectory prediction model for moving targets in edge networks and a mobility-aware parallelizable task offloading strategy based on this model. Experimental results show that the model effectively predicts the movement trajectory of the targets and the task offloading hit rate is closely related to the user's moving speed. Additionally, network bandwidth occupancy is significantly related to the degree of task parallelism and the number of services running on servers.