Machine learning techniques have been widely adopted to assist in data analysis in a variety of Internet of Things (IoT) systems. To enable flexible use of trained learning models, one viable solution is to leverage all categories of data from different applications to train a general model, which can be further tuned for applications through tuning process. This process incurs additional overhead at the start, but makes later revision and iteration faster and more flexible. Nonetheless, due to limited computing capabilities, IoT devices cannot handle the training process of large datasets. To address this issue, in this paper, we propose a general framework to adopt transfer learning in industrial Internet of Things (IIoT) systems. In our study, we categorize the application space of applying transfer learning to IIoT systems into four generic scenarios: centralized transfer learning with large datasets, distributed transfer learning with large datasets, centralized transfer learning with small datasets, and distributed transfer learning with small datasets. According to the characteristics of each scenario, we design workflows to apply transfer learning technique. To demonstrate the efficacy of the approach, we apply our transfer learning technique to the task of IIoT component recognition. We use the known VGG-16 model and leverage T-Less industrial datasets to evaluate the performance of our approach in different scenarios. Via performance evaluation, our experimental results confirm the efficacy of our approach, which can not only reduce training time, but also achieve higher accuracy, compared with the classical convolutional neural network (CNN) approach.

With the increasing adoption of Industrial Internet-of-Things (IIoT) devices, infrastructures, and supporting applications, it is critical to design schemes to effectively allocate resources (e.g., networking, computing, and energy) in IIoT systems, generally formalized as optimization problems. Nonetheless, because the system is highly complex, operation and networking graph-based environments are time varying, and required information may not be available, it is difficult to leverage traditional optimization techniques to solve the optimal resource allocation problem. In this article, we propose a deep $Q$ -network (DQN)-based scheme to address both bandwidth utilization and energy efficiency in a networking graph-based IIoT system. In detail, we design a DQN model that consists of two deep neural networks (DNNs) and a $Q$ -learning model. The DNN network abstracts the features from the highly dimensional inputs and obtains the approximate $Q$ -function for the $Q$ -learning model. Based on the $Q$ -function, the $Q$ -learning model can generate the $Q$ -table and reward function. After the training process, the DQN model can select appropriate actions for the agents (i.e., robots in a smart warehouse in this study) to improve bandwidth utilization and energy efficiency. To evaluate our proposed scheme, we design a simulation environment to investigate a typical IIoT scenario: the actuation of robotics in a smart warehouse. We then implement the DQN model and conduct extensive experiments to validate the efficacy of our scheme. Our experimental results confirm that our scheme can improve both bandwidth utilization and energy efficiency, as compared to other representative schemes.

Cyber-Physical Systems (CPS), including smart industrial manufacturing, smart transportation, and smart grids, among others, are envisioned to convert traditionally isolated automated critical systems into modern interconnected intelligent systems via interconnected human, system, and physical assets, as well as providing significant economic and societal benefits. The characteristics of CPS include complexity, dynamic variability, and heterogeneity, arising from interactions between cyber and physical subsystems. These characteristics introduce critical challenges in addition to existing and vital safety and reliability requirements from traditional critical systems. To overcome these challenges, Artificial Intelligence (AI) and Machine Learning (ML) schemes, which have proven effective in numerous fields (robotics, automation, prediction, etc.), can be leveraged as solutions for CPS. In particular, reinforcement learning can make precise decisions automatically to maximize cumulative reward via systematic trial and error in an unknown environment. Yet, challenges still remain for integrating complex reinforcement learning systems with dynamic and diverse CPS domains. In this paper, we conduct a thorough investigation of existing research on reinforcement learning for CPS, and propose a framework for future research. In addition, we carry out two case studies on reinforcement learning in transportation CPS and industrial CPS to validate the effectiveness of reinforcement learning in targeted applications. Using realistic simulation platforms, we validate the effectiveness of reinforcement learning for decision making in routing for transportation CPS and production control for industrial CPS. Finally, we outline some future research challenges that remain.

The Industrial Internet of Things (IIoT) is a critically important implementation of the Internet of Things (IoT), connecting IoT devices ubiquitously in an industrial environment. Based on the interconnection of IoT devices, IIoT applications can collect and analyze sensing data, which help operators to control and manage manufacturing systems, leading to significant performance improvements and enabling automation. IIoT systems are characterized by a variety of IIoT applications, which generate different computing tasks depending on their functionalities. Some tasks are time-sensitive, while others are not, and more importantly, some tasks are non-preemptive in IIoT scenarios. Thus, processing the different IIoT applications efficiently in an IIoT environment is key to achieving automation. Since computing resources are limited in IIoT, how to rapidly process time-sensitive tasks is a critical issue. Although some existing scheduling schemes can deal with the latency requirements of time-sensitive tasks, they lack consideration for non-preemptive tasks. To address this issue, in this paper we consider a typical smart warehouse system as an example and propose a generic task scheduling scheme that reserves computing resources to wait for upcoming time-sensitive tasks in such an IIoT environment. In doing so, our proposed scheme is capable of minimizing the overall waiting time for time-sensitive tasks. To evaluate the proposed scheme, we have implemented a simulation platform for a smart warehouse and conducted extensive experiments. Our experimental results demonstrate the efficacy of our scheme, which can allocate computing resources so that the processing time for the time-sensitive tasks can be reduced. Additionally, we discuss some potential research directions toward improving performance in IIoT environments with respect to resource management, machine learning, and security and privacy.

Industrial Internet of Things (IIoT) envisions the tight coupling of numerous critical industrial manufacturing subsystems, such as control, networking, and computing through the ubiquitous Internet of Things technologies. Nonetheless, such interconnectivity poses significant challenges to the successful management and operation of massively distributed industrial manufacturing systems. Without carefully integrated system design, the nonoptimal management and operation of highly intertwined subsystems can lead to the loss of productivity and ultimately the value of factories and plants. To address this issue, in this article, we conduct the integrated design that is capable of simultaneously configuring both control and networking subsystems in IIoT with consideration for their inherent interdependencies. We first analyze the performance of the dynamic backoff exponential (BE) in IEEE 802.15.4 carrier-sense multiple access (CSMA) and show the performance improvement of dynamic BE. We then design a model-free reinforcement learning algorithm to configure the control and networking subsystems automatically via systematic trial and error, as it is impractical to build a model for a highly intertwined complex IIoT system. Considering the time-sensitive characteristics of IIoT systems, we design priority-aware policies based on importance among networking traffic (i.e., sensing traffic and actuation traffic) to improve the convergence speed. The experimental results demonstrate that our priority-aware reinforcement-learning-based integrated design can successfully reconfigure the complex and highly intertwined IIoT system at runtime with a minimal convergence time. Besides, our approach reduces convergence time by 37.5%, and energy consumption by 9.2%, compared to the standard reinforcement learning approach.

Industrial Internet of Things (IIoT) is predicted to drive the fourth industrial revolution through massive interconnection of industrial devices, integrating advances in smart machinery and data analytics driven by computing, networking and artificial intelligence techniques. The massive interconnections between industrial devices increase not only information exchange, productivity, and resource efficiency, but also cybersecurity risks and their impact on the Industrial Internet of Things (IIoT) system. In this paper, we investigate the evolutionary process, which captures the natural interactions and evolutions of both attackers and defenders, as well as their strategies to launch data integrity attacks and protect the IIoT system, respectively. We leverage the wireless cyber-physical simulator (WCPS) as a realistic IIoT environment to implement and evaluate strategies at different levels of evolution. The experimental results indicate the effectiveness of defensive strategies in guarding against various forms of data integrity attacks with increase in time and knowledge throughout the evolutionary process.