Are you optimising the Edge for IoTFollow article
Life at the edge
Putting compute and analytics close to where data is created can provide a versatile and scalable foundation for the industrial Internet of Things
The Internet of Things (IoT) has become a critical enabler of our daily existence. Whether at work, rest or play, the deployment of smart devices that communicate seamlessly over the internet has become woven into the very fabric of modern life and is now so all-encompassing that we rarely tend to give it a second thought.
Indeed, as the IoT proliferates, so the number of ‘things’ requiring connectivity continues to rise. According to a recent report from Strategy Analytics, more than 38 billion devices will be connected to the internet by 2025, rising to 50 billion by 2030.
But there is a problem, in that the IoT runs the risk of becoming a victim of its success. As deployments have become more multi-faceted, and organisations grow to rely on data to undermine new business models, scalability, flexibility, and manageability become even greater considerations. How, for example, will it be possible to handle ever-larger streams of data flowing to the internet without being hindered by issues such as latency? And what happens in circumstances where cellular connectivity is not readily and reliably available? The answer, increasingly, lies with a more distributed computing topology that removes reliance on centralised infrastructure and pushes real-time computation and analytics out to the edge.
Defining IoT at the edge
So, let us take a closer look at edge computing, specifically in relation to IoT. At a top-line level, the edge is commonly used as a phrase which refers to the end of the connected space, where data is generated or consumed. In some instances, that might be considered the last point of high-speed connectivity and high-bandwidth compute availability, such as an internet gateway device. But increasingly in the world of industrial IoT, the edge is more likely to be situated at a remote location where a battery-powered or energy-harvesting sensor or actuator is placed, collating and sending operational data across to the cloud using some form of long-range, ultra-low-power wireless techniques. Here, use of edge infrastructure introduces a degree of local autonomy that provides delegated authority to control specific functions within defined operating parameters, while also providing the environment to run multiple applications without the need for constant cloud connectivity. Edge, therefore, provides a considerable boost in flexibility and agility.
The potential for edge deployment is abundantly clear, with the GSMA’s recent Opportunities and Use Cases for Edge Computing in the IoT report identifying several primary benefits for many implementations. These include:
- Low latency: By placing computation and data storage nearer to the devices out in the field, a shorter round trip for communication is required, reducing latency.
- Longer battery life for IoT devices: With communication channels open for shorter periods, the battery life of battery-powered IoT devices could be extended.
- Access to data analytics and AI: Edge processing power and data storage could all be combined to power in-situ analytics and AI. This requires speedy response times involving the processing of ‘real-time’ data sets that are too large to transmit to centralised systems.
These sorts of performance advantages mean that adoption of edge computing is likely to create myriad opportunities in a vast number of industries as diverse as healthcare to retail, through to automotive, utilities and logistics. According to McKinsey, this uptake will be big business – with a prediction that it could create more than $200 billion in hardware value by 2025. Understanding the potential of the edge is ever-more essential, then, both for electronic product suppliers and for the enterprises which stand to benefit from the new opportunities that it will deliver.
Finding the edge
So, where at the edge might data be generated or consumed? In industrial IoT, it could be sensors, actuators or a range of other embedded devices located in a wide variety of locations such as a factory, an offshore platform, a water processing plant or a vehicle on the road. It can be anywhere, really, where there are the requirement and ability to perform the processing of data and to turn that information into actionable insight, often in a real-time environment.
Then, from the edge, the typical architecture would include communication with an IoT gateway through the use of a variety of field protocols such as Bluetooth, Zigbee and WiFi, and then on finally to an IoT cloud platform. In many cases, the required analytics are performed in the cloud. But increasingly as the miniaturisation of processing and storage technology has gathered pace, this intensive compute is also being carried out in-situ, out at the edge, providing faster and more reliable insight, without the need for always-on cloud connection.
This sort of infrastructure is already widely used and well understand. But some technical and operational obstacles remain. In terms of barriers to implementation, security is perhaps the biggest concern. The more complex and interlinked nature of edge infrastructure would, at first glance, make it seem more open to attack from events such as malware intrusions at the weakest point. However, many advanced security tools and techniques such as strong identity management and authentication can be effectively applied, while update of firmware can be performed through secure transmission links. With malicious intent aimed at single nodes having reduced impact, edge makes it harder for any attack to disrupt the entire IoT network, and the distributed nature of the architecture makes it easier to shut down compromised sections before further damage can be done.
Another potential barrier is establishing a solid business case for the edge and accurately identifying return on investment. As organisations look to implement the edge by pushing data processing away from a centralized platform, there is a requirement for significant investment in the network. But change equals risk – and effective edge requires a carefully structured, gradual approach to implementation requiring full assessment of a range of critical elements. These include existing legacy infrastructure such as industrial control, through to connectivity, analytics and the potential for automation through the application of artificial intelligence. In short, edge IoT should be considered an evolution, rather than a revolution, providing organisations with the confidence to progress at a rate that best suits them.
Applications of IoT at the edge
That is the basic theory behind edge IoT - but what about reality? In which industries are the edge starting to drive actual business change? The list is perhaps already more extensive than you might think, with McKinsey and Company in its report ‘New demand, new markets: what edge computing means for hardware companies’, identifying more than 100 edge computing use cases. The vast majority involve the need for real-time decision-making and the report acknowledges the fact that many of the new applications may need to operate with limited or intermittent connectivity, and that sufficient computing power needs to be available on-device. While it is not possible to outline the full extent of edge opportunity across all sectors, there are some significant hotspots of commercialisation that are worth noting:
The autonomous and connected smart car of the future is perhaps the most apparent application of IoT at the edge. Here, data received by a plethora of on-board vehicle sensors can be collected, analysed and acted upon – securely and reliably – in real-time. The advantage, of course, is speed: even with 5G and ubiquitous cloud computing connectivity, there will be some safety-critical real-time applications that cannot depend on data transmission across cellular networks, even if it was available everywhere and at all times. Intelligent on-board edge computing eliminates latency concerns for autonomous vehicles, providing the basis for accurate and fast decisions making close to the source of data origination.
Edge will power advanced technologies such as autonomous vehicles
Biometrics is an increasingly important technology for law enforcement agencies, with facial recognition now regularly applied as a primary enabler of video-based surveillance, access control, and many other scenarios. Historically, video streams from, say, an airport terminal or railway station would be forwarded to a high-performance backend such as cloud for processing – but this requires high network bandwidth in cases where there are a large number of sensors in use and a high volume of footfall. However, with edge computing deployed for pre-processing and filtering of images, the facial recognition engine in the cloud could be used to identify only the best quality key-frames, thereby reducing computing overheads and improving the reliability of the system.
Industrial assets in sectors such as utilities, rail, oil and gas and construction are often diverse, complex and spread over large geographical areas – often with poor quality connectivity. Monitoring these devices, therefore, can be a difficult task, with cloud-based data processing both expensive and unreliable. By pushing intelligence closer to the edge, in-situ analytics can result in the transmission of much smaller packets of more relevant sensor data, reducing the volume of information that needs to be sent across a network. This makes for more efficient use of network resources while saving money on cloud computing costs. It also opens up the possibility of more widespread monitoring and deployment of assets reducing the importance of internet connection for analytical purposes. Ultimately, the edge can be deployed to underpin more informed maintenance regimes, encouraging the implementation of genuinely predictive methodologies.
Many manufacturers are under a dizzying array of pressures as they strive to meet customer expectation for lower prices, shorter lead times and greater choice. The challenge, then, is to meet these demands without compromising on quality. More so than ever, sophisticated vision systems are being used to automate assembly inspections, reducing dependency on manual checks. But as image quality has improved, so the volume of data produced from video systems has grown, making the storing, processing and analysing of images increasingly burdensome. That is where the power edge computing comes in, specifically through the use of deep learning algorithms which can analyse video streams, identifying quality anomalies and triggering an immediate alert. Doing this in real-time, at the edge, rather than through the human eye can dramatically improve quality control and reduce the impact of quality issues on the production line.
Most applications of wearables technology are linked to the consumer sector, with fitness trackers such as FitBit proving a popular means of encouraging more healthier lifestyles. But wearables have an increasing role on the factory floor, with smart glasses and virtual reality headsets allowing workers to carry out hands-free tasks while having access to all the information they need. At present, most wearables have limited storage capacity, relying on the cloud to access compute capacity to do the hard work. However, as advances in low-power processing capabilities continue to progress, AI-centric edge devices have the potential to improve the user-experience of wearables, providing faster and more impressive visualisation. This is likely to enhance the versatility of wearables, helping them to find an even greater range of applications across industrial facilities.
Edge could power the increased use of augmented and virtual reality within industrial facilities.
There is no doubting, then, that edge IoT is here to stay. Indeed, according to management consultancy Gartner by 2025, around 75 per cent of all enterprise-generated data will be created and processed outside of traditional centralised data centres or cloud. That represents a massive shift from where we are today and illustrates the potential of edge to deliver faster and more versatile performance across a range of industrial applications. The challenge as the proliferation of connected devices grows at an exponential rate is to implement edge in a secure and cost-effective manner, enabling it to deliver on its full potential.