The Industrial IoT (IIoT) is beginning to move beyond concept to reality. The simple concepts of cheap-sensors-everywhere, data-to-the-cloud-for-analysis, and dashboards-on-your- browser are giving way to more nuanced and realistic models. Recent articles have explored this trend with discussion of fog or edge computing . This is a good first step, but it is not the complete picture. Mature IIoT systems will also include mist computing. This is computing close to the sensors and beyond what would normally be considered fog computing. In any given application, mature IIoT deployment will likely include a mix of these architectural concepts, including mist-cloud, mist-fog, and mist-fog-cloud. Driven by the increasing capabilities of microcontrollers, system-on-chip (SoC), and low-cost communications, mist computing will be an important component of millions of solutions.
The dog got the controller
A recent TV ad shows a family scrambling to find the TV remote. In the background, one family member asks Amazon's Alexa to select the correct channel. Alexa processes the request and sends commands to the Dish TV Hopper box. Within a short time, the family has settled down to watch, and the remote is forgotten.
The time taken to process the TV control loop is probably faster than it takes to find the mislaid remote controller (especially if the dog has carried it off to a different room). This is good, but do you want a nuclear power station controlled this way? How about the air traffic control system or the emergency alert system at your chemical plant?
This example highlights that the IoT splits into at least two major models: The consumer model and the industrial model. The consumer model usually involves the centralization of data followed by decision-making in the cloud. This is a perfect model for offering free or low-cost functions or services where the bulk data gathered can be monetized in other ways. The classic example is Google where search or email is offered for free, but the collected data is monetized in advertising services. There are other less-obvious examples, such as smart electric meters. The meter is on your house, but the data is gathered centrally, and you only get access via a carefully created utility web site. Most of the data is used by the utility for other purposes, primarily cost reduction, demand response, fault diagnosis, and system planning (Figure 1).
IIoT is different
With IIoT applications, there are other concerns: latency in the control loop, lower reliability due to a long chain of elements involved in the decision process, cost of data transmission and storage, and security of sensitive operational data [See sidebar].
To alleviate these concerns, Arm, Cisco, Intel, Microsoft, and others have proposed edge computing as an alternative . Here, an edge device or a set of edge devices contain business logic and can make decisions locally or regionally, without reference to, or cooperation with, a central core. The concept has become known as fog computing to note the decentralized nature and distinguish it from the centralized cloud . Fog computing may involve a single edge device or multiple edge devices operating together. There are many combinations, and most examples will work in conjunction with the cloud resources (Figure 2).
There is, however, another factor at play. By gathering up all sensor data, we risk overwhelming our systems with a mix of relevant and irrelevant data. We receive so much data that it is hard to figure out what to do with it. This is similar to the problem of “alarm fatigue” suffered by pilots in aircraft cockpits and medical staff in hospital intensive care units .
The better approach is to derive intelligence from the sensor data and only transmit the intelligence to the decision-making site (fog or cloud). In ideal circumstances, the intelligence is derived close to the sensors not in edge or cloud-computing locations. This concept is becoming known as mist computing (Figure 3).
Here, the idea is to use low-cost micro-controllers to do more than data conversion and simple communications. The processing power is used to look at streams of data from multiple sensors and derive inferences and complex insights. We can also look at the condition of the sensors themselves. This approach may yield additional understanding of what is happening at the location or assist with maintenance cycles.
Sensor platforms to the rescue
Fortunately, advances in sensor platforms, such as the CRATUS/FUJITSU BlueBrain system , and powerful micro-controller families, such as the Arm Cortex series, make this an economic and straightforward approach. Platforms such as these contain a mix of sensors, I/O, computing resources, communications, and development resources, making it easier to prototype a solution for an individual problem or application. If the required volume is low, the sensor platform can be used as a final solution. If the volume is high, a custom design can be crafted by cost-reducing the platform hardware and software (Figure 4).
At the Sensors Expo & Conference (June 2017, San Jose, California), CRATUS and Fujitsu demonstrated an example of mist computing by directly tying two BlueBrain sensor platforms to a Microsoft HoloLens headset. This demonstration of augmented reality (AR) used sensor data from the BlueBrain platform directly overlaid onto the visual seen through the HoloLens, and was achieved on a Cortex M4 processor without additional edge, handheld, or cloud computing. This is an example of how feedback and control may be provided in an industrial setting where complexity in the visual field makes it hard to distinguish cause and effect. Direct systems of this nature support human-machine collaboration and improve safety in dangerous environments.
Cloud platforms are readily available, so one may ask, "Why bother?" (see sidebar). Recent work gives us clues: There are savings in communication, power and, by implication, storage costs . In a recent IEEE article , Markakis et al, compared three different approaches from pure cloud to pure edge, and the resulting savings were significant.
Similar benefits can be gained by using computing resources close to the sensors. In this case, the communications will usually be by wireless, so in addition to the benefit of bandwidth, we will be reducing the radio frequency noise and interference levels from the billions of IIoT devices expected to be deployed.
The bottom line: Mist is already here
Current IoT and IIoT solutions are one-dimensional–they are typically deployed to address one need or use case. The true benefits of IoT technology will appear when we have these multiple systems cooperating to help with a bigger picture. This work cannot be done in the cloud for all use cases. Fog and mist devices need to be flexible. They must be open to having additional functions overlaid long after initial deployment.
The increasing power of microcontrollers and recent advances in software-defined sensors by my company  will provide capabilities at the mist-computing level that are hard to grasp in the current market– just like the smart phone was hard to grasp in the earlier cell phone market.
The best example for understanding future architectures is to look at the vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) systems being rolled out now and for the coming decade. This infrastructure looks like the cloud, yet vehicles will communicate with each other and make mutual decisions without reference to the infrastructure. These are fog devices. Drilling further down into automotive systems, individual ECUs in the car will also make decisions about their own subsystems without reference to the overall car systems. This is mist computing–and it is already here.
- “Internet of Things Leaders Create Open Fog Consortium to Help Enable End-to-End Technology Scenarios for the Internet of Things.” (2015, November 19). https://iotbusinessnews.com/2015/11/19/80306-internet-of-things-leaders-create-openfog-consortium-to-help-enable-end-to-end-technology-scenarios-for-the-internet-of-things/.
- McMillin, Bruce et al. (2017, February). “Fog Computing for Smart Living.” IEEE Computer Magazine, Vol 50, No 2, page 5.
- Wald, Matthew L. (2010, July 31). “For No Signs of Trouble, Kill the Alarm.” New York Times. http://www.nytimes.com/2010/08/01/weekinreview/01wald.html.
- “FUJITSU Component Sensor-Based System BlueBrain(r) Interface Board.” (2017, August 2). http://www.fujitsu.com/downloads/MICRO/fcai/wireless-modules/bluebrain-interface-board.pdf.
- Markakis, Evangelos K. et al. (2017, July). “EXEGESIS: Extreme Edge Resource Harvesting for a Virtualized Fog Environment.” IEEE Communications Magazine, pp 173-179.
- Gunay, Z. (2017, October, 31) “Software Defined Sensors for Industrial IoT and Industrie 4.0.” www.cratustech.com/downloads/.