The industrial edge needs intelligence for many Industrial Internet of Things (IIoT) projects to produce successful results, both technically and financially. To make the edge “intelligent,” end users must understand what constitutes the real edge close to or even at the data sources, what both the edge and the cloud can and cannot accomplish, the key challenges in creating an intelligent industrial edge, and the significant technical and, most importantly, economic benefits this transformation will deliver. An intelligent edge can fill in the gaps to follow through on problem identification, time-to-resolution, and access for new pathways to success. In the IIoT, it all starts with the business case, and this paper provides several examples of successful business cases.
ABI’s latest report, entitled “The Business Case for Industrial IoT Edge Intelligence” explores the technical and economic benefits of IIoT Edge intelligence.
Design considerations include:
- Leveraging a hyper-efficient complex event processor (CEP) for real-time analytics on streaming industrial data.
- Harnessing IIoT AI, including continuous, closed-loop Edge to Cloud machine learning.
- Lowering data persistence and transport requirements through edge intelligence.
- Improving security posture by eliminating the need to transmit sensitive OT data across networks.
- Deploying cloud-agnostic edge intelligence to facilitate multi/hybrid cloud strategies.
- Seamless bridging edge intelligence with OT tribal knowledge, transferring operator domain expertise into analytic expressions and ML models.
- Leveraging small footprint edge computing and controller hardware to minimize new computer investments.
- Benefiting from subscription, not consumption, based pricing for much more controllable/predictable operating costs and radically lower charges for data-intensive applications.
Read on this article and get expert analysis on the technical and economic benefits of adding edge intelligence into industrial operations, including several real-world use cases.
What is Industrial Edge?
How Edge Intelligence Augments the Cloud
What are the key challenges of creating an Intelligent Edge?
The future of Edge Intelligence
The Economic of Edge Intelligence
What is Industrial Edge?
Different types of service providers and vendors define the edge in different ways. A Mobile Network Operator (MNO) might say the edge means cellular base stations. A gateway supplier might say it is a gateway. An Information Technology (IT) networking company might point to routers and switches, and an industrial radio networking company might think of remote field devices. Where does real edge intelligence truly reside?
When discussing edge intelligence, the edge is more than a “thing.” It is the fabric of networked devices close to or even at the data source being analyzed. This fabric includes a variety of both IT and Operational Technology (OT) devices that run on a variety of Operating Systems (OSs), processors, and protocols, which might exist in an automobile, on an oil rig, on a factory floor, along miles of pipeline, throughout a building, in an elevator, or in any other industrial environment.
Enterprises may transform any of the following devices into an edge intelligent device:
- Distributed control systems
- Control systems
- Motion sensor kits
- Robot arm controllers
- Internet of Things (IoT) gateways
- Industrial Personal Computers (PCs)
- On-premise single and multi-core servers
- Other small-footprint devices, e.g., the UP² board
How Edge Intelligence Augments the Cloud
Application enablement platforms and other cloud-based IIoT solutions can help discover and highlight long-term trends in data and assist in strategic planning across geographies, but this only covers a fraction of the potential benefits of the IIoT. As the volume of sensor data and the number of data sources grow, enterprises will develop more uses, applications, and services built on data. Whoever stores these data will attract high-value application add-ons for everything from management and security to analytics and consulting services. ABI Research refers to this growing phenomenon as the theory of data gravity.
The challenge for IIoT applications is that the “speed limits” imposed by each analytic value-chain component and/or process also extend to the network. The “holy grail” of data management is to create an environment that enables high-velocity data streaming alongside real-time analytics, at cloud scale, with zero manual tunings. For example, if an application requires instant responses to the data, e.g., local safety mechanisms and contextually-aware devices, then keeping the appropriate processing task at the edge is typically the only way to deploy it (versus a round-trip query to the cloud and back). The idea is to ship code to data (also known as serverless computing) rather than data to code. There are several key ways that edge intelligence augments the cloud:
Substantially Reduces Costs Driven by the Volume of Data: Edge intelligence can leverage the existing compute resources in industrial environments to reduce the volume of data sent to the cloud by providing analytics on-premises. Already, a large retail shop generates an average of 0.8 Terabytes (TB) of data/day; a large oil refinery generates 1 TB; an automated manufacturing facility generates 24 TB; a mine can generate 1 Petabyte (PB). Only a small percentage of these data drives decisions because moving these amounts of data to the cloud requires too much bandwidth. More video and audio sensors will only exacerbate the problem without an intelligent edge reducing this volume of data.
Relying solely on cloud computing increases communications and cloud processing and storage costs. By conducting more processing on existing on-premises compute infrastructure, these costs will go down.
Improves Security and Regulatory Compliance: Transmitting data off-premises inherently presents more security risks than processing data on-premises, and regional regulations may require that data stay on-premises. The IIoT will connect many legacy devices and pieces of equipment never previously networked running decades-old software with no security features. Limiting exposure to the Internet and the cloud through an intelligent edge also limits the security risk and helps end-users stay compliant with regulations.
Creates High Fidelity Data Analytics: Processing data close to the source ensures that the application gets the most data possible for that source’s use case in its most complete form. The effectiveness of the application increases because it runs on near real-time streamed data rather than batched data. Batched processing in the cloud tends to present “blind spots” or “jitters” that do not occur in streamed data, as seen in Chart 1 below.
Enables Contextual Machine Learning Models: Any Machine Learning (ML) or Artificial Intelligence (AI) model running on an intelligent edge will develop algorithms for its specific use case. Each IIoT edge has unique characteristics that the model should consider. These AI/ML models can also be used to test, train, and triage new AI/ML models; when done right, they can greatly improve automation.
Lowers Latency: The speed of light limits the cloud so that no matter how much computing power it harnesses, it will still face some delays in transmitting data and insights. Enterprises must process their data close to the source to obtain actionable analytics with the lowest possible latencies. Industrial control applications often require micro-second latencies currently impossible via cloud connectivity.
Delivers OT-Centricity: OT includes the computers and systems that monitor or control physical assets, devices, and processes in industrial environments using field bus or industrial ethernet protocols. IT includes the computers and systems that store, retrieve, process, and transfer data and information using Internet Protocol (IP) and Hypertext Transfer Protocol (HTTP). IT and OT do not easily communicate and exchange data. The cloud necessarily communicates using IT protocols, but many applications need to process data originating from OT equipment. An intelligent edge processes data on OT equipment.
To truly maximize the utility of data, enterprises can choose to implement an “intelligent edge” and find the optimal balance of edge and cloud computing.
What are the key challenges of creating an Intelligent Edge?
In any industrial edge environment, enterprises generally have:
- Lots of legacies, small compute footprint gear (control systems, Programmable Logic Controllers (PLCs), etc.)
- A variety of processors and OSs
- Tons of open and proprietary communication protocols
- Limited to no IT staff at the end-user
- Limited to no connectivity
- Scalable discovery and fusion of data from digital, audio, and video sensors
- High volume/velocity streaming data analytics
These factors result in greater demands for edge computing and edge analytics.
Table 1 below shows some of the protocols with which IIoT applications must interoperate.
Cloud solutions process the data from these environments after the end-user collects and transmits it to the off-site data center in batches. Edge intelligence empowers end-users to take this existing complex infrastructure and fuse high-volume digital, audio, and video data streams to produce near-real-time analytics at low latencies. It cleanses and enriches the data on-premises to improve operations and business outcomes on-site. Because the intelligent IIoT edge does all this on-premises, it can operate in limited bandwidth situations or even locations lacking an Internet connection entirely, such as remote or highly secure locations.
The future of Edge Intelligence
An intelligent edge will require the following functionalities and capabilities:
Ease of Deployment and Orchestration: An intelligent edge provides a Complex Event Processor (CEP), ML capabilities, and real-time stream processing with actionable analytics to any range of computing devices. It needs to run seamlessly on any of the industrial devices listed in Section 2, most of which have constrained compute footprints. To run on any of these devices, it needs to run on any of their OSs. An edge platform does not require a standard, full-feature OS or hypervisor, but instead works on pre-existing software. The edge management system must have auto-discovery of sensors and an OT-intuitive, easy-to-use Graphical User Interface (GUI) that empowers end users to define analytics without programming skills.
Real-Time Stream Processing: Technology drag means business drag. Most cloud solutions analyze data as they are sent up in batches at regular intervals for offline analysis. This takes time and results in a lag between the upload of data and the download of results. Edge intelligence, on the other hand, cleanses, enriches, and processes data in streams as the sensors make their readings. Its CEP engine can derive near-real-time insights and high-fidelity inferences through contextual sensor fusion without waiting on batched processing.
Edge to Cloud Closed-Loop ML: Every operational use case at the edge has a slightly different context. As a result, different algorithms are needed to process different kinds of data for different use cases. Small differences can have a big impact on the bottom line. Manually updating each algorithm is too costly. ML enables standard algorithms to adapt to live data streams. World-class edge intelligence must have pre-built ML model libraries and closed-loop ML that can run by itself at the edge or in concert with leading cloud computing solutions to empower autonomous operations as seen in Graphic 3 below.
Data Science Services/Expertise Available: Edge intelligence should help end users solve operational business problems. To do that, intelligent edge algorithms must produce results that prescribe an obvious call to action, but these algorithms may require data curation tasks that demand dedicated personnel. In this situation, edge intelligence vendors must offer a managed service option.
- Edge Intelligence Consideration: Real-time CEP-based streaming analytics Technical Benefits: Built for advanced, low latency, closed-loop industrial applications
- Edge Intelligence Consideration: Iterative machine learning (ML) on live industrial data Technical Benefits: Continuous inferencing on all sensor data (including video, audio) for closed-loop ML
- Edge Intelligence Consideration: Radically lower data persistence and transport requirements Technical Benefits: Processing live data at the source reduces data networking and storage resources
- Edge Intelligence Consideration: Enhanced security posture Technical Benefits: Edge processing eliminates need to transmit critical OT data across networks
- Edge Intelligence Consideration: Cloud agnostic Technical Benefits: Avoids cloud provider lock-in and facilitates multi/hybrid cloud strategies
- Edge Intelligence Consideration: Taps into OT tribal knowledge Technical Benefits: Translates operator domain expertise into analytics expressions and ML models
- Edge Intelligence Consideration: Leverages small footprint edge computing and controller HW Technical Benefits: Runs on industrial control systems or highly constrained edge compute devices
- Edge Intelligence Consideration: Subscription, not consumption, based pricing Technical Benefits: Easier to project scaling requirements after initial PoCs
The Economic of Edge Intelligence
Computing capabilities only matter if they generate business results at scale. Although by now, most industrial enterprises have at least experimented with IIoT applications, few have scaled. They do not scale because of speed and ease of deployment, and because the costs start to outpace and outweigh the benefits. There are several different ways intelligent edge computing can change the bottom line.
Easy to Set up/Deploy, Leveraging Existing Infrastructure
Rather than requiring additional investment in data centers or extra gateways, edge computing can run on the already installed base of devices. If gateway suppliers charge ~US$1,000 per gateway and servers cost even more, then cutting hardware, overhead, and infrastructure costs can make an enormous difference. Real edge computing takes advantage of existing resources to minimize additional expenses and should reach the time to value in months.
- Edge Intelligence Consideration: Real-time CEP-based streaming analytics Economic Benefits: Faster actionable insights for greater operating efficiencies (uptime, yield, energy savings)
- Edge Intelligence Consideration: Iterative machine learning (ML) on live industrial data Economic Benefits: Higher quality predictive insights to drive asset performance and process improvements
- Edge Intelligence Consideration: Radically lower data persistence and transport requirements Economic Benefits: Reduces cloud storage and communications costs by 100-1000x
- Edge Intelligence Consideration: Enhanced security posture Economic Benefits: Reduces security infrastructure, risk mitigation, and regulatory compliance costs
- Edge Intelligence Consideration: Cloud agnostic Economic Benefits: Increases bargaining power with cloud providers and reduces sourcing costs
- Edge Intelligence Consideration: Taps into OT tribal knowledge Economic Benefits: Cheaper and faster than PLC reprogramming, Avoids expensive cloud-based AI exercises
- Edge Intelligence Consideration: Leverages small footprint edge computing and controller HW Economic Benefits: Minimizes investments in heavy compute or new industrial control systems hardware
- Edge Intelligence Consideration: Subscription, not consumption, based pricing Economic Benefits: More controllable/predictable operating costs, radically cheaper for data-intensive applications
Highest Value from Data
The low-latency, high-fidelity data processing in intelligent edge computing leads to faster decisions and more immediate actions based on more accurate and more immediate information with fewer “blind spots.” By acting quicker or planning maintenance further ahead, the end-user can prevent costly machine failures or downtime and reduce scrap, resulting in higher yield. Preventing downtime can save a company US$10,000 per minute or more, depending on the industry and use case.
Reducing Communication Costs
If a facility or connected asset sends every sensor reading or even only the anomalies to the data center or cloud, this can result in massive bills from the Communication Service Provider (CSP). With an intelligent edge platform, enterprises can increase flexibility and make smarter decisions about what they publish to their cloud platforms. They can limit the data they send to the cloud to insights that help find long-term trends or help make strategic decisions. The edge platform will already have processed and routed more immediate actionable insight on-site.
Reducing Cloud Costs
If enterprises reduce communication costs by limiting the data they send to the cloud, they will also face smaller bills for cloud processing and storage. Cloud storage might look very cheap at face value, US$0.021 per Gigabyte (GB) per month for Amazon S3, but if an enterprise generates hundreds of TB per month, that can quickly add up to hundreds of thousands of dollars per year. On the other hand, if end users only publish data necessary for long-term or strategic decisions to the cloud, the amount of cloud processing and storage they use each month will shrink and so will their cloud service bill.
Workload Consolidation, Lowering Total Cost of Ownership
Edge intelligence not only leverages existing infrastructure, but it also normalizes, filters, and enriches data for seamless app integration and workload consolidation. If each piece of hardware runs its own OS for its own application without intelligence, these applications will remain siloed. These siloed applications make it difficult to coordinate maintenance alerts with enterprise systems and make it more expensive to repair, maintain, and replace the hardware in the field, especially in geographically dispersed or mobile environments. Edge intelligence makes it easier to consolidate multiple workloads from discrete devices onto a single device that can process multiple streams of data, thereby lowering the total cost of ownership.
Real edge intelligence requires stream processing, ML, and actionable analytics running on a tiny footprint on various OT OSs. FogHorn Systems, one of the early leaders in edge intelligence software, brings the above-mentioned capabilities and benefits to its clients, as demonstrated in the use cases identified below.
Smart Buildings: Optimizing Elevator Performance
One client hired FogHorn to monitor the sensor data on its millions of elevators and escalators to reduce the inspection and repair costs at a rate of more than US$2,000 per event. FogHorn was installed on the existing motion sensor kits, with a footprint of less than 1Gb. FogHorn’s software, executing several machine learning models, was leveraged to predict maintenance needs, and send alerts to the appropriate personnel. The client then reduced the number of regularly scheduled maintenance inspections and instead scheduled technicians to inspect and repair equipment when the machine learning models sent alerts. This reduced the company’s overall repair and servicing costs and generated new managed service revenue.
Smart Manufacturing: Improving Capacitor Production Yield
One client’s capacitor winding machine could not meet its yield targets and produced increasing amounts of scrap. While the machine-generated large amounts of data, the client could not find the root cause or predict future issues. Further, they had no way to monitor the data in real-time or deliver the analytics to the shop-floor workers.
FogHorn’s CEP engine analyzed the sensor data in real-time, and its EdgeML machine learning software adapted the algorithms to multiple data streams, producing analytics and failure alerts delivered to the OT staff. This empowered them to reduce maintenance costs by catching problems earlier and only calling in technicians when they had a problem to fix. The solution also increased the yield of the machine and reduced scrap work.
Smart Manufacturing: Optimizing Steel Manufacturing
Another client, a steel manufacturer, had installed vibration sensors to identify bottlenecks in the production process, but the data exceeded the available bandwidth for transmissions to the cloud. Without analyzing this data, the manufacturer could not predict process failures. FogHorn’s CEP engine enriched the multivariate vibration data as they streamed from the sensors, and EdgeML conducted a Fourier analysis at the edge on an OT-centric operator dashboard. It then sent the results to the cloud for vendor-specific analysis. In this case, FogHorn delivered actionable edge analytics and reduced data transfer costs while enhancing the cloud-based solution. FogHorn’s software reduced the data sent to the cloud by 90%. The resulting analysis enhanced production KPIs to avoid bottlenecks and a foundational platform for output optimization.
Oil & Gas: Early Pump Failure Detection
A global oil & gas company had a large volume of industrial pumps that needed to monitor with limited communications and computing resources. It asked FogHorn to reduce cavitation events and minimize equipment damage in the hopes of reducing maintenance lifecycle costs by 30-40%. Cavitation occurs with air bubbles form in the pumps. If the air bubbles implode, it can generate a shock wave and significantly damage parts of the pumps. Deploying FogHorn on existing control gateways, its CEP engine monitored pressure, vibration and acoustic sensor data and generated early alerts if any of the sensors produced readings outside the pumps’ normal operating parameters. Edge ML successfully predicted failure conditions with significant lead time by interpreting patterns in the sensor data. The client managed all of this on an OT-centric platform without any data scientists. The solution significantly reduced pump lifecycle costs, cavitation and equipment damage and empowered the client to institute predictive rather than regularly scheduled maintenance.
Oil & gas: Automated Flare Stack Monitoring
FogHorn has also detected problems at natural gas processing plants by analyzing the size of flares and smoke coming off the many flare stacks. These plants must comply with strict environmental regulatory requirements. Before deploying FogHorn’s solution, the client spent a large amount of maintenance and compliance. Using FogHorn, the client compared video data to pressure, acoustic, and vibration sensors from a compressor in near real-time. Its convolutional neural networks (CNN) for deep learning worked at the edge to find the types of bad sounds or vibrations that lead to bad smoke. The solution helped the client monitor its operations in real-time for compliance while lowering operational expenditure (Opex) and maintenance costs and improving safety.
Smart Rail: Locomotive Operational Efficiency
A train operator leveraged FogHorn’s software to optimize fuel usage, detect sub-optimal operating conditions and reduce the bandwidth and networking costs of monitoring its trains. FogHorn installed its software into existing on-board hardened data collection systems. The software analyzed idling and throttle data based on location, speed and time in near real-time. Before, it streamed most of the raw data straight to the command center. Now, the system sends proactive alerts with video only on abnormal conditions to command centers for quick resolutions to problems and operational optimization. These changes reduced fuel and cellular communication costs while optimizing crew and train performance and ensuring safe operating conditions.
Electric Vehicle Charging: Charging Station Predictive Maintenance
An operator of electric vehicle charging stations deployed FogHorn to monitor its stations and improve performance. Before, it had faced an extremely high % of false-positive alerts for +/- input phase volts and lacked any anomaly detection based on historical data and values. This resulted in high maintenance costs. FogHorn installed its software at the stations for real-time condition monitoring of battery conditions. It reduced the % of false positives, and its predictive maintenance app reduced maintenance costs and downtime.
As more companies in various industries deploy edge intelligence, the IIoT will finally live up to its potential. Edge intelligence fills the gaps in the IIoT to make it financially viable. Real-time stream processing, ML, and actionable analytics on existing compute infrastructure leads directly to reduced communication and cloud costs and quicker resolution of issues with less downtime. Meanwhile, edge intelligence also makes it easier to consolidate multiple workloads from discrete devices onto a single device that can process multiple streams of data, thereby lowering the total cost of ownership. Industrial companies need real edge intelligence to maximize the utility of their data in almost any use case. Rail, smart buildings, smart manufacturing, oil & gas, and EV charging stations use cases that have already started to benefit from reduced bandwidth and cloud costs, as well as lower latencies for IIoT apps. World-class edge intelligence directly contributes to the bottom line in all of these cases. Any company hoping to reduce maintenance costs, communication costs, cloud costs, operational costs, and/or downtime, and to improve efficiency, safety, and performance of any asset with sensor data should deploy real edge intelligence as an integral part of the IIoT solution.
Source: FogHorn Systems