From Data to Decision: The Case for IoT Edge Analytics in Predictive Maintenance
Most IoT devices don’t need—or aren’t equipped—to send data to the cloud. Faster decisions can be made locally with predictive maintenance systems and software.
Whether it’s 20 billion or 200 billion connected IoT devices, the accepted model for the Internet of Things is data flowing to the cloud—and control, information and business analytics flowing back to users, their machines and connected devices. According to IBM, the majority of IoT systems today are cloud-centric. Yet robust, low latency connectivity cannot be guaranteed for all installations and use cases. Moreover, many industrial and business IoT installations neither need nor want cloud connectivity for security, geographical, or business reasons.
In the case of industrial predictive maintenance making decisions using IoT data, there are issues to consider. The manufacturing floor LAN is intentionally kept separate from the corporate enterprise. Data from machines and IoT-enabled devices is in formats and databases inaccessible to the traditional higher stack and cloud infrastructure. And besides secure connectivity, networking together industrial machines with the business enterprise is often neither acceptable nor practical.
“After all, this is the sweet spot of the IoT. The solution is to keep the IoT local and mostly separate from the cloud.”
Yet the promise of IoT analytics to predict machine failures, to optimize production, to apply machine learning to process flows, and to garner improvements and efficiency is just too compelling to ignore. After all, this is the sweet spot of the IoT. The solution is to keep the IoT local and mostly separate from the cloud. Data collection, storage, processing, analysis, and sharing or control should be done at the edge or within the “fog” of local processing resources.
The Industrial IoT: Use Case
By now, the notional IoT topology diagram is well understood (Figure 1). It shows connected IoT node devices such as pumps, valves or wind mills spewing data to the cloud where enterprise machines run software that presents the device data in dashboards that operators can then act upon (Figure 2).
For example, ADLINK and Intel, companies partnered together providing hardware, software and processors for the IoT, present a use case where smart vending machines not only report sales in real time but also give operators inventory and maintenance status. Route technicians can restock, collect coins from and repair the machines most needing attention first while servicing less critical machines at a lower priority. Figure 3 shows the variables involved with typical IoT route logistics.
A broader description of “predictive maintenance” is shown in Figure 4. The promise of the IoT to “monetize data” is clearly depicted in this diagram. For example (from the top and going clockwise), reducing machine failures and operating equipment at the highest possible speed increases output/time and maximizes profit. Lower risk exposure might be accomplished via enhanced safety in public places through the use of surveillance cameras and facial recognition.
Workforce productivity can be optimized through better machines, or through machine learning that identifies extra steps that can be eliminated. Enabling smarter replacement and inspection can be handled by machine sensors that recognize wear-out signals, such as noisy bearings or increased motor current consumption. Repairing or replacing equipment before it fails—especially during off hours—can keep productivity high, be it a factory floor or a traffic signal.
Despite the IoT, Problems
Yet despite the compelling nature of the notional solutions presented by the promise of the connected IoT, real-world problems prevail. Not the least of which is that capital and operational budgets are tight. While devices, machines and what is generically described as “IoT nodes” can make data available, that data often remains locked inside the device or trapped on a separate network.
Operators may recognize that the availability and reliability of their assets lag behind baseline performance expectations, but they don’t know by how much nor what “levers” to pull to make the entire system work more efficiently. Operators see that they are plagued by breakdowns, ill-timed maintenance and quality stoppages, but can’t see a way ahead to run equipment in a less tactical, but more strategic fashion.
Worse: they have no data to present to management to justify trying a different approach. Their mountains of data are considered “isolated data,” and are often stranded in disconnected systems. Or, devices with data are unconnected—usually because they’re legacy machines or nodes that talk proprietary or bespoke protocols. Getting them connected requires both hardware and software conversion from the old to IP-based networks.
Remember that notional diagram in Figure 1 of the IoT showing everything flowing to/from the cloud? Interestingly, according to research experts at IDC, Goldman Sachs, IMS Research and MC/EDC, “…40% of IoT data will be stored, processed, analyzed…at or near the edge…” And despite the perception of everything flowing to the cloud, those groups predict that “…50% of IoT networks will be network constrained…”. That is: cloud connectivity?…maybe half the time. The solution is more processing at the edge with predictive maintenance software running on individual nodes, collections of nodes in a fog arrangement, or in dedicated edge equipment. A topographic description of predictive maintenance at the edge is shown in Figure 5.
IoT, the Cloud and Weather Thou Art?
In 2015 IBM purchased the Weather Channel. To most, it seemed an odd marriage, but IBM foresaw value in aggregating the world’s weather data with cloud computing, the company’s Watson supercomputer, the IoT and data-centric software. IBM’s cognitive solutions come in several packages entitled: IBM Predictive Maintenance, Predictive Quality, Predictive Warranty, and Foundation for Energy. Of most interest to embedded edge analytics is the company’s Predictive Maintenance product, PMQ. The product’s value proposition aims to solve the problems and use cases so far described, as shown in Figure 6.
Despite IBM’s data center and super computer capabilities, PMQ applies to the edge and to embedded systems. The company cites four key reasons when edge processing trumps cloud computing:
- Insufficient bandwidth to push data to the cloud. With so many devices at an installation, networks can be insufficient to service them all within a predefined time slice.
- Connectivity cannot be guaranteed. Many IoT nodes are autonomous, geographically distant, off the grid, and moving. Agricultural equipment is a good example: is there 4G wireless in all areas of remote North Dakota?
- Latency is a critical parameter. IoT protocol/database vendor PrismTech is a pioneer in the publish-subscribe standard that became the Object Management Group’s (OMG) Data Distribution Service (DDS). PrismTech cites that IoT nodes with latencies in the millisecond range might have difficulty relying on cloud analytics and predictive maintenance.
- Linking IoT nodes and networks to the cloud can be difficult, due to enterprise incompatibilities, security, and local protocols such as CANbus, SCADA and others.
We’ve so far described some examples of real-world problems that can benefit from predictive maintenance, and have established when edge processing might be a preferred IoT solution to cloud computing. How best to realize embedded edge predictive maintenance?
Life on the Edge, with Peers and a Little Fog
If IoT edge nodes can’t depend on deterministic, reliable cloud connections to provide predictive maintenance, then individual nodes, collections of nodes, and local gateways will provide the necessary processing. Peer-to-peer, peer-to-fog, fog-to-fog, and fog-to-cloud become the primary data sharing models.
Data can be easily shared among these entities using OMG’s DDS, a standard relied on by defense, industrial, and transportation industries. Vortex Cloud, a DDS distribution implementation by PrismTech, manages the distributed databases between resources while handling protocol conversion among entities (Figure 7). “Fog” computing refers to clusters of edge nodes sharing data and compute resources.
A predictive maintenance system, using edge resources and DDS data sharing is shown in Figure 8. Here, a machine vision inspection sensor (camera, lower left) connects to a SCADA network and passes measurement data to a “cognitive gateway.” Shown in this image is ADLINK’s MXE-200i, which passes data upwards to an edge “cognitive industrial appliance.” The hardware shown is ADLINK’s SETO-1000 rugged, industrial server.
Either the cognitive gateway or industrial appliance can run the IBM PMQ software and its “scoring engine.” In most use cases, the business analytics software can run on the lower power Intel Atom processor-based gateway, reserving the higher performance Intel Xeon processor-based industrial appliance for more intense computing needs, such as on-going scoring model updates. Predictive maintenance decisions are made, and data is passed back to the original gateway plus an additional gateway elsewhere on the SCADA network.
The predictive data reaches one or more pieces of control equipment and the “actionable intelligence” promise of the IoT is realized. That is: source data from the camera run through PMQ, and the scoring engine algorithm within the software is used to manipulate the robot via its controller. Data sent to the second gateway is used to predict other downstream events and can be used to adjust other controllers/robots.
A sample dashboard created by PMQ running on the SETO-1000 is shown in Figure 9. What’s different in this example from the typical IoT explanation is that all the decisions and resources are local, at the edge. The dashboard is also local, although the gateways’ and server’s data and analyses could also have been sent to the cloud for the benefit of a remote enterprise, such as the factory’s engineering or accounting departments.
Still Room for Clouding Around
We’ve shown that in several real world use cases, like factory automation, smart vending machines, or route delivery logistics, predictive maintenance and the promise of IoT can best be performed at the edge. This offers the flexibility of keeping data within the confines of the factory or at least separate from the corporate enterprise for security reasons. As well, local processing is faster and probably more available than inconsistent cloud connections.
Yet the IoT is a vast concept and it continues to evolve. Though predicting analytics at the edge has its place, vast amounts of data will still flow to the cloud for users to dissect and act. For these cases, companies like ADLINK offer SEMA Cloud for remote control and to present information in concise, programmable dashboards.
This article was sponsored by ADLINK Technology.