In 2019, we are no longer debating whether industrial IoT (or Industry 4.0) will happen. Instead, we’re looking at what shape it will take as the manufacturing industry becomes increasingly digitized – primarily to reap benefits related to production optimization and reduced maintenance costs. While these benefits may sound mundane, they drive real investment in the industrial IoT infrastructure because the payoffs are so measurable and immediate.
According to IDC, “the industries that are expected to spend the most on IoT solutions in 2018 are manufacturing ($189 billion), transportation ($85 billion), and utilities ($73 billion).” All of these are industrial applications of IoT. In fact, as IDC says, “Consumer IoT spending will reach $62 billion in 2018, making it the fourth largest industry segment.” 1
There is a massive amount of data in a manufacturing process that can be harnessed to achieve very substantial goals – predict faults, optimize equipment lifetimes, derive new revenue streams, and even optimize the production process to better align with market needs. The first job of any industrial network is to facilitate the collection of this data, implement the right balance between local, real-time data processing and longer term, off-line data storage, and, finally, to take effective action to optimize the industrial process.
In fact, many times the need to process the data locally, in real time, drives the need for a different type of an edge computing device. As McKinsey writes in a recent article, “This movement of computational capacity out of the cloud—to the edge—is opening up a new sector: edge computing.”2
This article argues that the need and the increasing capability to move the processing to the edge of the network create a different type of industrial IoT network – a network which may not have a strict hierarchy and will exhibit a variety of connectivity and processing options amongst myriad forms of edge devices.
Industrial IoT: At the Edge
In an industrial world, there is a shift of data processing to the edge. Let’s illustrate this with a few examples:
- Instead of a temperature sensor that would traditionally trip when the temperature went above a certain threshold, we now have sensors that can accurately measure and log the temperature locally. The sensor can compare a running series of temperature data points against a set normal to determine when the equipment under test may start exhibiting signs of wear and tear. A sensor like this may never communicate with the central control system if the machine is running within a pre-set limit. But it can also flag preventative maintenance as required.
- A vibration sensor can continuously record the vibration of a servo motor. By doing local fast Fourier transform (FFT) analysis on the vibration data and comparing the dominant vibration frequencies, it can only flag the controller if the vibration frequency is out of spec.
A real-life example of this is explained by Enterprise IOT Insights magazine3. In this case study, they describe how Intel decided to monitor the health of its fan filter units (FFUs) in its semiconductor production facilities. FFUs filter and clean the air inside industrial machines. They are everywhere on the factory floor. Intel placed an accelerometer at the top of each FFU to measure variations in the fan’s function and did all the processing in a local gateway as shown in Figure 1. Only the summary data and reports were sent to the cloud.3
Industrial IoT: A Different Model
In a traditional IoT model, sensors/hardware are at the bottom. Their job is to collect the data and through built-in network connectivity, send the data up on the stack to an IoT server/platform. Data analysis, data visualization, and application development happens on the collected data. The result of this analysis is then used by management to determine the right course of action. This can be scheduled machine maintenance, production process optimization, or something else. This very hierarchical network structure is shown in Figure 2.
In fact, most networks in industry are not like this at all. First, it makes no sense to put in a 5G radio or Wi-Fi in each sensor, especially those operating in rugged conditions like in a coal mine, underwater, or next to a hot, vibrating servo motor.
Second, most of the instantaneous data that the sensor is collecting has only real-time value. A proximity detector that detects a dangerously close human presence needs to shut off the motor (STO) or bring it to a safe speed (SLS). This data does not need to go up the stack for a management decision.
Lastly, in some cases sending second-by-second vibration data via a network, for example, is a waste of bandwidth. A summary report (on the FFT analysis) may be relevant for down-line analysis. Real-life industrial IoT networks are more closely approximated in Figure 3.
Much of the data processing is being done locally and even multiple sensor data may be analyzed using edge servers. The connectivity is not always via a TCP/IP network connection, rather traditional rugged industrial networks are used. These can be industrial fieldbus, RS-485, BACNet, or some other industrial standard. Generally, the summary data is put on the network from the edge servers. This is the data which is then stored on the IoT servers to be analyzed and visualized and for which different applications can be developed. A representative factory network is shown in Figure 4.
Edge processing will become important in various industrial IoT applications. The systems that will be used for edge processing must exhibit small form factors and low-power characteristics. But other than that, they can differ widely. A sensitive temperature measurement device in the process industry will vary from an occupancy detector in a smart building, which itself will differ from a sensor deployed to enable precision agriculture.
The development of different edge devices will be done by different companies that are experts in that field. As edge devices proliferate, the architecture of the industrial IoT will change from what has been generally promoted. A lot of value generated will flow from the cloud equipment makers to companies that are experts in an edge equipment for a specific application.
About the author
Suhel Dhanani is a director of business development for the Industrial & Healthcare Business Unit at Maxim Integrated. Prior to joining Maxim, Suhel led the industrial automation segment marketing at Altera Corp. Suhel has over 20 years product/segment marketing experience in various Silicon Valley companies such as Xilinx, Altera and Tabula. He has published various articles and is an author of a book titled, Digital Video Processing for Engineers. Suhel holds MSEE and MBA degrees from Arizona State University and a Graduate Certificate in Management Science from Stanford University.