The Internet of Things offers high-speed communication among machines without human intervention. It is only natural that developers of industrial processes try to adapt such a promising technology to manufacturing environments. However, industrial manufacturing environments pose different challenges than those found in consumer or commercial scenarios. In many cases the Industrial Internet of Things (IIoT) is a pure Internet of Things in that there is no input from humans nor output to humans as in many consumer or commercial applications: all devices connected to the network operate autonomously without human supervision. Therefore, all parameters related to human user satisfaction are much less relevant, while others, often deemed as secondary, take a much bigger importance
Industrial environment vs. consumer/commercial environment
IoT in consumer and industrial environments has requirements that are not coincident1. Thus, it is Important to keep in mind different requirements for each application, depending on its intended use. The following table portrays some significant differences.
| Consumer IoT | Industrial IoT | |
| Network type | Mostly public networks | Mostly private networks (sometimes public infrastructure for long-distance communication is used) |
| Area traffic capacity | Must be high, since the network is shared | Not an issue, since the network is tailored to and used exclusively by the application |
| Bandwidth | Very important: typically, data is transferred to users in short, big bursts | Not very important. Data transmission is typically low volumes of data transmitted repeatedly |
| Latency | Low importance. Human users experience no discomfort from less than a second wait. | Very important for process control, and critical for applications like unmanned vehicle control in the factory |
| Reliability | Low importance: transfer can be repeated | Very important for the same reasons that latency is and, besides, data preservation. |
| Hardware dependency | Low: typically, they operate without direct access to hardware | Very high: they operate in close contact with hardware. Software is often embedded |
It is thus clear that IoT in industrial environments has a specific set of requirements that normally differ from those applications in non-industrial environments.
In addition to the above differences, it must also be considered that industrial equipment is very expensive and requires a long period of use to ensure profitability. Several generations of equipment with different features often operate together. As a result, industrial applications normally cannot be set up from scratch, and can require additional hardware to allow machines with limited communication and information storage capability to live up to the requirements of IoT.
The costs of stopping a running manufacturing process are high, not only because of the lost production output, but also because of additional costs incurred to stop and then start again the process. In extreme cases, equipment must be partially or totally replaced (e.g. blast and glass melting furnaces). Therefore, development and deployment of the IIoT must be made with the process running and must also not interfere with the system’s normal operation. With the spread of IIoT, it will become increasingly incorporated in the process design, but until all today existing process are replaced, the need to install IIoT on a “live” system will be an important issue.
Key factors of IIoT applications
Latency
A low and foreseeable latency is essential to ensure several critical features of IIoT, mostly related to the need for real-time control. This includes not only control of unmanned vehicles and moving equipment, but also critical industrial processes, where a failure to close the control loop in time can result in damage to goods, equipment, and, worst, physical injury to persons.
Reliability
Reliability is essential for the same reasons as latency. Latency just ensures that the response is generated fast enough. However, to achieve control, it must be ensured that the response arrives at its destinations. A lost response in a real-time environment is equivalent to no response.
In addition to the above factors, reliability is a key issue in an industrial environment since many generated data are volatile. For instance, temperatures, pressures, current consumption, and other process parameters are essential for quality control and vary from machine cycle to machine cycle. Thus, it must be ensured that machine-generated data arrives at its destination, either by providing interim storage for these pieces of data or by ensuring that data reaches its destination first time it is delivered. Also, the need to ensure human safety implies that hard-wired safety mechanisms completely isolated from the IoT network are in place.
Data Storage
Data generated by IIoT applications is often stored for later data analysis. Even though high-volume data bursts are not present in normal operation, data are generated on a regular basis, often 24/7, resulting in huge volumes of generated data. A single factory with 100 machine-tools and 10 cameras generates 204GB of data every day or 72TB per year2. A significant part of this information will have to be stored for big data analytics and ML training.
Data security
Data generated in an industrial process is extremely sensitive. A security breach can disclose valuable know-how to competitors and result in huge losses to the breached company, its suppliers, and its customers.
Positioning accuracy
In industries, hundreds of entities are in the factory at the same time: containers, machines, vehicles (manned and unmanned), production workers, maintenance personnel, etc. Many are movable, and a significant portion move autonomously. Knowing the position and state of all of them is essential for proper planning and execution of manufacturing, to guarantee equipment and components’ integrity and, most important, to ensure human safety. Human safety overrides any other factor, no matter the material costs incurred. Since humans can (and do) move unpredictably, they are ubiquitous in modern factories and they work in even closer conjunction with automatically moving equipment in shrinking areas, so positioning accuracy of humans and moving objects is essential. In IIoT systems, this is achieved by several sensor devices, like cameras, RFID readers, receiving antennae, etc., all connected to the IoT and sharing information.
Distributed computing
Due to real-time control requirements in an industrial environment, reaction time for control loops must be both short -to ensure a timely response- and predictable -to ensure a consistent effect of actions. An all-cloud solution normally offers no such guarantees. it is hence necessary to execute decision logic as close to the control system as possible. This means moving computing operations from the cloud to computers and to computing resources embedded into the physical systems that are being controlled. Tasks requiring more computing power but not a real time response can be left in the cloud while tasks with stricter response time requirements can be moved as close to the physical systems as necessary. The IoT architecture and communication protocols must ensure that all computing resources work together smoothly.
How IoT-NGIN addresses the key factors of IIoT applications
Latency
Requirements for a low latency are fulfilled by the combined use of a 5G-NR network and high-speed Wi-Fi. Even though the theoretical goal for 5G-NR is 1 latency of 1 msec for the entire control loop3, practical applications show that a latency below 6 msec with an availability over 99,9999% can be obtained4. Based on experience with former projects, a latency below 10 msec is considered sufficient for industrial IoT projects at the current state of the art, and it can be potentially improved by using high-speed Wi-Fi together with 5G-NR.
Reliability
As stated before, 5G-NR ensures an availability over 99,9999%. In addition to this network, high-speed Wi-Fi, and Visible Light Communication (Li-Fi) will be used, with the twofold purpose of providing a backup network for added reliability and also evaluate their feasibility as a standalone solution.
Data Storage
Not exclusive of IoT, data storage is a common and old problem for companies that has grown over time as more information was generated during manufacturing processes. There are several storage solutions commercially. IoT-NGIN will help reduce the storage needs by providing distributed AI that will provide early data processing to extract significant information.
Data Security
IoT-NGIN will use Federated Machine Learning. This technology provides many useful features and also addresses risks in the cybersecurity area within IoT-NGIN. To provide data security according to application needs, attack detection models will be developed for use in ML. To train these models, attacks will be analyzed, and synthetic datasets will be created to train the models before deploying them. The model will also be retrained with data from new attacks. In addition, an IoT crawler will be deployed to detect vulnerable nodes before they are attacked.
Positioning accuracy
IoT-NGIN will offer a single digit centimeter precision for positioning accuracy, that will allow control of unmanned vehicles in tightly packed factory areas. Even though current Wi-Fi positioning systems and 5G have difficulties to achieve this precision grade, a combination over 5G network, 6th generation Wi-Fi and visible light positioning (VLP) through Li-Fi is expected to yield the desired position. This precision grade will also allow an evaluation of collision risks and the modification of unmanned vehicles’ route to avoid not only collisions, but also sudden braking caused by the emergency braking systems entering int function.
Distributed computing
IoT-NGIN will use several connected computing devices that will work together to run AI algorithms. These devices will include not only classical computers like workstations and servers, but also several devices and wherever possible, small and low-capable devoces. Distributed computing will use the Federated ML for optimal utilization of computer resources.
1 Pal Varga , Jozsef Peto, Attila Franko, David Balla, David Haja, Ferenc Janky, Gabor Soos , Daniel Ficzere, Markosz Maliosz and Laszlo Toka “5G Support for Industrial IoT Applications – Challenges, Solutions, and Research Gaps” www.mdpi.com/journal/sensors Feb 4, 2020
2 D. Mourtzis*, E. Vlachou, N. Milas “Industrial Big Data as a result of IoT adoption in Manufacturing” Procedia CIRP 55 ( 2016 ) 290 – 295
3 G. Fettweis and S. Alamouti, “5G: Personal mobile internet beyond what cellular did to telephony,” in IEEE Communications Magazine, vol. 52, no. 2, pp. 140-145, February 2014, doi: 10.1109/MCOM.2014.6736754.
4 arXiv:1904.01476v1 (Florian Voigtl ̈ander∗, Ali Ramadan†, Joseph Eichinger†, J ̈urgen Grotepass†, Karthikeyan Ganesan†,Federico Diez Canseco‡, Dirk Pensky§and Alois Knoll∗ 5G for the Factory of the Future: WirelessCommunication in an Industrial Environment)