With the tremendous revival of artificial intelligence, predictive maintenance (PdM) based on data-driven methods has become the most effective solution to address smart manufacturing and industrial big data, especially for performing health perception (e.g., fault diagnosis and remaining life assessment). Moreover, because the existing PdM research is still in primary experimental stage, most works are conducted utilizing several open-datasets, and the combination with specific applications such as rotating machinery is especially rare. Hence, in this paper, we focus on data-driven methods for PdM, present a comprehensive survey on its applications, and attempt to provide graduate students, companies, and institutions with the preliminary understanding of the existing works recently published. Specifically, we first briefly introduce the PdM approach, illustrate our PdM scheme for automatic washing equipment, and demonstrate the challenges encountered when we conduct a PdM research. Second, we classify the specific industrial applications based on six algorithms of machine learning and deep learning (DL), and compare five performance metrics for each classification. Furthermore, the accuracy (a metric to evaluate the algorithm performance) of these PdM applications is analyzed in detail. There are some important conclusions: 1) the data used in the summarized literatures are mostly from public datasets, such as case western reserve university (CWRU)/intelligent maintenance systems (IMS); and 2) in recent years, researchers seem to focus more on DL algorithms for PdM research. Finally, we summarize the common features regarding our surveyed PdM applications and discuss several potential directions.

With the great evolution of human society from the information era to the smart automation era, intelligent production and maintenance have become the core orientation of Industry 4.0. Data-driven intelligent predictive maintenance (IPdM) offers potential solutions to satisfy the urgent demand of predicting impending failures and mitigating unplanned downtime of industrial machinery, and is envisioned as a critical enabler for future industrial Internet of Things. In this article, we provide a comprehensive survey of deep learning (DL)-based IPdM for the researchers and practitioners who focused on the promotion of fault diagnosis/prognosis. First of all, four mainstream algorithms of DL are reviewed around the specific IPdM applications of recent five years. Some prominent discussions and insights on the DL-based IPdM are given here in terms of data characteristics and model performance. Particularly, as an important contribution, a new industrial dataset of dual-bearings rotating machinery is published for opening research to address one of the main challenges for the development of DL-based IPdM, i.e., the shortage of high quality industrial datasets. Compared with current datasets, our dataset is multisensory, which is closer to reality and more suitable for building advanced DL-based IPdM models. We also give a DL-based case study to demonstrate how to build a predictive model for maintenance decisions based on this dataset. Finally, potential research trends and challenges are also discussed for enhancing the intelligence of IPdM in the industrial scenarios.

Network slicing technology and the Open Radio Access Network (O-RAN) architecture provide solutions for resource isolation and dynamic resource optimization in resource allocation models for heterogeneous services. However, the existing resource allocation models employ uniformly sized radio Resource Blocks (RBs) and fixed-length time slots, which incur low resource utilization and a high packet loss rate, especially when the packet arrival rate is high. In this paper, we propose a multi-cell, multi-dimensional resource, and multi-timescale O-RAN slicing framework, named M3O-RANS, which meets the Quality of Service (QoS) requirements of heterogeneous services and reduces resource costs for network service providers. Specifically, the M3O-RANS framework partitions multiple timescales into a large timescale and multiple small timescales, using radio RBs of different sizes for inter-slice operations at the large timescale and reusing radio RBs across the multi-cell network for intra-slice operations at the small timescale, thereby improving resource utilization. To optimally allocate radio and computing resources according to time-varying network conditions, we formulate an optimization problem with the objective of minimizing resource costs for network service providers while satisfying delay and reliability requirements. Since inter-slice and intra-slice resource constraints are mutually coupled, it is difficult to solve the problem via traditional Deep Reinforcement Learning (DRL) methods. Therefore, we decompose the original problem into two subproblems and propose a Parallel Hierarchical DRL-based Resource Allocation (PHDRA) algorithm. The proposed algorithm can make decisions regarding the size and number of radio RBs and the amount of computing resources at the upper layer while making radio RB selection and computing resource allocation decisions at the lower layer. Simulation results demonstrate that the PHDRA algorithm can improve the QoS of heterogeneous services and reduce resource costs for network service providers compared with state-of-the-art benchmarks.

The demand for accurate localization in complex environments continues to increase despite the difficulty in extracting positional information from measurements are not yet complete. Like any other process, localization also has security requirements. The use of ultra-wideband (UWB) indoor localization systems have recently grown quickly in industries, with a reliable, fast, and have high accuracy performances. In particular, time difference-of-arrival (TDOA) is one of the widely used localization models. However, as TDOA measurement errors increase, the accuracy of the localization decreases. The accuracy of the TDOA measurement is influenced the accuracy of the localization and affected by multiple factors such as time synchronization, errors in sensor positions, missing data is caused by network attack. To reduce the influence of a sensor measurement error in localization, this paper proposes an improved localization algorithm for source localization using deep learning to address a TDOA measurements error or missing data in an asynchronous localization called DeepTAL. Unlike the conventional algorithm, DeepTAL can obtain highly accurate localization data in the presence of TDOA measurement errors or missing data. The algorithm starts with TDOA measurements without time synchronization. A network based on Long Short-Term Memory (LSTM) is then applied to achieve a stronger learning and better representation of the determined target state and TDOA prediction. The network can express features more abstractly at higher levels and increase recognition accuracy. After that, the target node is accurately located by TDOAs. We implement the DeepTAL algorithm on an asynchronous localization system with UWB signals. The experiments show that the proposed DeepTAL algorithm is efficient in improving the localization precision as measurement errors or missing data.

Localization has emerged as an attractive solution to enable new business models that rely on location-based services in wireless networks for communication, sensing, and control. In particular, time difference-of-arrival (TDOA) is one of the widely used localization models. However, the conventional TDOA requires precise time synchronization between a target node and anchor nodes for measuring the time difference, which leads to a large number of packets for communication. To reduce packet transmission, we propose a model for measuring the time difference without time synchronization called ASync-TDOA. Different from the conventional model, ASync-TDOA can measure the time difference in a one-way-based ranging by introducing the reference node. Specifically, the time difference between the target node and anchor nodes can be directly measured by the server based on the timestamps from the reference node. After that, the target node is accurately located using least squares and brute force for the time difference. We implement the ASync-TDOA model on a localization system with ultra-wideband signals to estimate the localization of the target node, which is easy to operate in practical engineering. The experiments show that the proposed ASync-TDOA is efficient in reducing the packet transmissions and improving the TDOA measurement and localization precision.

As a key component of rotating machinery, the health perception of bearings is essential to ensure the safe and reliable operation of industrial equipment. In recent years, research on equipment health perception based on data-driven methods has received extensive attention. Overall, most studies focus on several public datasets to verify the effectiveness of their algorithms. However, the scale of these datasets cannot completely satisfy the representation learning of deep models. Therefore, this paper proposes a novel method, called CarNet, to obtain a more robust model and ensure that the model is sufficiently trained on a limited dataset. Specifically, it is composed of a data augmentation method named equitable sliding stride segmentation (ESSS) and a hybrid-stacked deep model (HSDM). The ESSS not only amplifies the scale of the original dataset but also enables newly generated data with both spatial and temporal correlations. The HSDM can, therefore, extract shallow spatial features and deep temporal information from the strongly correlated 2-dimensional (2-D) sensor array using a CNN and a bi-GRU, respectively. Moreover, the integrated attention mechanism contributes to focusing limited resources on informative areas. The effectiveness of CarNet is evaluated on the CWRU dataset, and an optimal diagnostic accuracy of 99.92% is achieved.