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.

In this paper, we aim to make the best joint decision of device selection and computing and spectrum resource allocation for optimizing federated learning (FL) performance in distributed industrial Internet of Things (IIoT) networks. To implement efficient FL over geographically dispersed data, we introduce a three-layer collaborative FL architecture to support deep neural network (DNN) training. Specifically, using the data dispersed in IIoT devices, the industrial gateways locally train the DNN model and the local models can be aggregated by their associated edge servers every FL epoch or by a cloud server every a few FL epochs for obtaining the global model. To optimally select participating devices and allocate computing and spectrum resources for training and transmitting the model parameters, we formulate a stochastic optimization problem with the objective of minimizing FL evaluating loss while satisfying delay and long-term energy consumption requirements. Since the objective function of the FL evaluating loss is implicit and the energy consumption is temporally correlated, it is difficult to solve the problem via traditional optimization methods. Thus, we propose a “ Reinforcement on Federated ” (RoF) scheme, based on deep multi-agent reinforcement learning, to solve the problem. Specifically, the RoF scheme is executed decentralizedly at edge servers, which can cooperatively make the optimal device selection and resource allocation decisions. Moreover, a device refinement subroutine is embedded into the RoF scheme to accelerate convergence while effectively saving the on-device energy. Simulation results demonstrate that the RoF scheme can facilitate efficient FL and achieve better performance compared with state-of-the-art benchmarks.

Performing deep neural network (DNN) inference in real time requires excessive network resources, which poses a big challenge to the resource-limited industrial Internet of things (IIoT) networks. To address the challenge, in this paper, we introduce an end-edge-cloud orchestration architecture, in which the inference task assignment and DNN model placement are flexibly coordinated. Specifically, the DNN models, trained and pre-stored in the cloud, are properly placed at the end and edge to perform DNN inference. To achieve efficient DNN inference, a multi-dimensional resource management problem is formulated to maximize the average inference accuracy while satisfying the strict delay requirements of inference tasks. Due to the mix-integer decision variables, it is difficult to solve the formulated problem directly. Thus, we transform the formulated problem into a Markov decision process which can be solved efficiently. Furthermore, a deep reinforcement learning based resource management scheme is proposed to make real-time optimal resource allocation decisions. Simulation results are provided to demonstrate that the proposed scheme can efficiently allocate the available spectrum, caching, and computing resources, and improve average inference accuracy by 31.4$\%$ compared with the deep deterministic policy gradient benchmark.

In this paper, we present a three-layer (i.e., device, field, and factory layers) deterministic federated learning (FL) framework, named DetFed, which accelerates collaborative learning process for ultra-reliable and low-latency industrial Internet of Things (IoT) via integrating 6G-oriented Time-sensitive Networks (TSN). Utilizing dispersive local data, industrial IoT devices distributively train a deep neural network (DNN) model, and the updated model parameters are aggregated at their associated field servers every round or at a centralized factory server every a few rounds. Aiming at optimizing the learning accuracy of FL without affecting the co-transmission of burst traffic (e.g., safety-critical traffic), an integrated TSN is considered to establish connections among the three layers, where a cyclic queuing and forwarding mechanism is deployed in each switch to support deterministic model parameter transmission with microsecond-level delay and near-zero packet loss requirements. To improve the FL performance, we formulate a multi-objective stochastic optimization problem to simultaneously maximize the scheduling success ratio and learning accuracy while satisfying the deterministic requirements of delay, jitter, and packet loss. Since the objective function is implicit and the available time slots of the considered TSN in each FL round are temporally correlated, the problem is difficult to solve in real time. Therefore, we transform the problem into a Markov decision process formulation and propose a dynamic resource scheduling algorithm, based on deep reinforcement learning, to make optimal resource scheduling decisions while adapting to device heterogeneity and network dynamics. Experimental results based on real-world dataset demonstrate that the proposed DetFed significantly accelerates FL convergence and improves learning accuracy as compared to state-of-the-art benchmarks.

Deterministic transmission guarantee in time-sensitive networks (TSN) relies on queue models (such as CQF, TAS, ATS) and resource scheduling algorithms. Thanks to its ease of use, the CQF queue model has been widely adopted. However, the existing resource scheduling algorithms of CQF model only focus on periodic time-triggered (TT) flows without consideration of bursting flows. Considering that the bursting flows often carry high-priority data in real systems, in this paper we investigate the mixed-flow (i.e., TT and bursting flows) scheduling problem in CQF-based TSN aiming to maximize the number of schedulable flows and system load balance while satisfying the deterministic demands of delay, jitter, and reliability for both TT and bursting flows. Unfortunately, it is challenging to schedule the mixed flows with the original CQF model because of the huge difference between TT and bursting flows. To resolve this problem, we firstly design an enhanced Multi-CQF model to satisfy the basic demands of bursting flows sent at any time without affecting the deterministic transmission of TT flows. Given the complexity of mixed-flow scheduling and the proposed queue model, it is difficult for traditional algorithms to fully utilize network resources. Thus, we further propose a time-correlated DRL resource scheduling (TimeDRS) algorithm to optimize the resource allocation. TimeDRS can be extended to other time-related resource scheduling scenarios, such as TDMA-based scheduling. Experimental results demonstrate that our proposed approaches can greatly reduce frame loss and end-to-end latency for bursting flows, and well balance runtime and schedulability compared with state-of-the-art benchmarks.

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.