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.

Collaboration among industrial Internet of Things (IoT) devices and edge networks is critical to support computation-intensive deep neural network (DNN) inference services which require low delay and high accuracy. Sampling rate adaption which dynamically configures the sampling rates of industrial IoT devices according to network conditions, is the key in minimizing the service delay. In this paper, we investigate the collaborative DNN inference problem in industrial IoT networks. To capture the channel variation and task arrival randomness, we formulate the problem as a constrained Markov decision process (CMDP). Specifically, sampling rate adaption, inference task offloading and edge computing resource allocation are jointly considered to minimize the average service delay while guaranteeing the long-term accuracy requirements of different inference services. Since CMDP cannot be directly solved by traditional reinforcement learning (RL) algorithms due to the intractable long-term constraints, we first transform the CMDP into an MDP by leveraging the Lyapunov optimization technique. Then, a deep RL-based algorithm is proposed to solve the MDP. To expedite the training process, an optimization subroutine is embedded in the proposed algorithm to directly obtain the optimal edge computing resource allocation. Extensive simulation results are provided to demonstrate that the proposed algorithm can significantly reduce the average service delay while preserving long-term inference accuracy with a high probability, as compared to benchmark schemes.

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.

With the rapid development of artificial intelligence and industrial Internet of things (IIoT) technologies, intelligent predictive maintenance (IPdM) has received considerable attention. To efficiently predict impending failures and mitigate unexpected downtime, while satisfying the instant maintenance demands of industrial facilities is very important for improving the production efficiency. In this paper, a self-attention based framework, called DeepHealth, is proposed for the instant IPdM. Specifically, the framework is composed of two submodels (i.e., DH-1 and DH-2), which are respectively utilized to perform the health perception and sequence prediction. Considering the potential temporal correlation in time series, we deploy an enhanced attention mechanism to capture global dependencies from the vibration signals, and leverage the long- and short-term sequence prediction of sensor signals to carry out instant decision-making. On this basis, we conduct destructive experiments based on the IIoT-enabled rotating machinery and construct a balanced industrial dataset for model evaluations.

In this article, we propose a multilayer Ka/Q-band satellite–terrestrial integrated network for the Internet of Remote Things (IoRT) to achieve a high transmission rate with communication robustness in dynamic network environments. Under this architecture, we investigate how to jointly manage the offloading path selection and resource allocation to offload computation-intensive and delay-sensitive tasks in the IoRT. Considering continuous low earth orbit (LEO) satellite movements and Markovian rainfall changes, the computation offloading problem is described as a Markov decision process (MDP) formulation with the objective of maximizing the number of offloaded tasks with satisfied delay requirements and minimizing the power consumption of the LEO satellites. A deep reinforcement learning (DRL) approach is leveraged to make optimal decisions by taking account of dynamic queues of IoRT devices, channel conditions that vary with rainfall intensities and satellite positions, and computing capabilities of ground stations. Extensive simulations are conducted to validate the effectiveness and superiority of our proposed scheme.