The Future of Embedded IoT and Edge Computing: Making AIoT Real with Flexible, Open, Plug and Play Edge Platforms

Author: Jeroen Baerents, European Business Development Manager /Product Sales Manager, Advantech Europe

System integrators need edge computing platforms that are flexible, open, off-the-shelf, and plug-and-play to unleash the full potential of the Artificial Intelligence of Things (AIoT)

Big Data is the new oil for industry, and the parallels between the two commodities are striking. Unleashing the latent value within data is impacting every aspect of life and work, and it is changing the way infrastructures are managed. It makes new activities possible, and it transforms the way businesses operate, conceive, and deliver new products and services. 

The way data is handled – from its extraction, refinement, transportation, and the processes applied to derive value-added products – critically affects its value. Even now, the data revolution is only just beginning. The possibilities are potentially endless, and many have not yet become apparent. Computing underpins and defines each stage in the data lifecycle, and the ability to move the data to the right processing platform easily, efficiently and quickly is essential.

The Emerging AIoT

The IoT provides the fundamental infrastructure that connects them all together. To deliver the needed flexibility, adaptability, security, robustness and reliability, IoT brings together a multitude of technologies including established Internet technologies as well as emerging technologies such as AI, 5G, and also Wi-Fi 6, which is engineered for much greater efficiency and scalability than preceding generations.

The entire solution, from sensors and aggregators to the central cloud, must be conceived as whole to ensure optimal performance and efficiency. What happens at the edge is increasingly pivotal to the outcome. 

Low latency and real-time response are critical to specific scenarios, such as industrial IoT (IIoT) implementations. Moreover, it is important to ensure that network bandwidth and compute resources, as well as energy, are utilised efficiently. Both of these reasons are driving the move towards more processing of raw data locally, near to the source, enabling time-critical responses to be communicated back quickly and forwarding only essential data to the central cloud. Gartner has predicted that 75% of data will be processed at the edge by 2025. 

Infusing AI into edge and embedded computing creates the AIoT, or Artificial Intelligence of Things, where AI adds value to IoT through machine-learning capabilities while IoT adds value to AI through connectivity, signalling and data exchange.

A complete AIoT solution brings together many elements and demands a wide range of competencies. With so many facets to the value chain, system integration can appear complicated, time consuming, expensive, and potentially uneconomical.

To make AIoT real and deliverable, effectively leveraging edge and embedded technologies in an effective blend, system integrators need software and hardware building blocks that are conceived for AIoT applications. To minimise development risks and time to completion, these building blocks should also be off-the-shelf; they should leverage open standards while offering plug-and-play convenience and simplicity. 

Choices for Edge Computing

Advantech is addressing this demand by creating plug-and-play computing platforms and software frameworks that simplify and streamline deployment of new AIoT applications in edge and embedded scenarios. These include software platforms such as Edge AI Suite and WISE-DeviceON (figure 1).

Edge AI Suite includes the Intel® OpenVINO toolkit™ and provides features that ease deep learning inference on edge devices, such as a model optimiser, inference engine, pre-trained models such as rapid object detection and human pose estimation, and a deployment wizard for launching AI models. It is also ready for third-party SDKs such as vehicle classification and license-plate recognition. In addition, support for the open-source, vendor-neutral EdgeX Foundry provides a software framework for interoperability between applications and devices such as sensors, which helps users establish easy plug-and-play connections.

pastedGraphic.png

Figure 1. WISE-DeviceON applications simplify connected-device management challenges.

(picture from https://www.advantech.com/resources/news/advantech-delivers-remote-device-operation-and-management-to-aiot-applications-using-the-wise-deviceon-series) 

WISE-DeviceON provides real-time device and data management capabilities and enables instant intelligent edge onboarding, data acquisition, and status visualization. There is also a set of WISE-DeviceOn industrial apps targeted for specific domain requirements, such as DeviceOn.DM for operating and managing industrial IoT devices. DeviceOn DM’s easy-to-use interface helps monitor device health, real-time actions to power on/off, troubleshoot problems, and send software and firmware updates over-the-air (OTA) on-site and remotely. There is also DeviceOn.CommBridge, which provides data acquisition, protocol conversion, rule engine, and easy-to-integrate APIs to quickly respond to the requirement of edge intelligence industry and help customers achieve digital transformation and upgrade in IoT era.

Other DeviceON apps include DeviceOn/iEdge for data and edge management, DeviceOn/Kiosk+ for remote kiosk management, DeviceOn/ePaper, DeviceOn/Display, and DeviceOn/SQ Manager for the remote control and management of related peripheral modules. These functions help customers manage connected IoT devices while building edge-to-cloud applications. 

Edge+ enhanced edge computing platforms incorporate processing, wireless connectivity, and cloud integration capabilities, and leverage domain-focused software including DeviceOn apps and others. The ARK-3532 edge computer is optimised for machine- and factory-automation applications and comes IoT-management ready with industry-standard communication interfaces, built-in security features, and WISE-DeviceOn pre-loaded. The ARK-3531 is featured for remote kiosk applications with a powerful compute engine and DeviceOn/Kiosk+ software.

On the other hand, the EIS-S230 Edge Cloud Solution provides a scalable on-premises platform that features built-in Kubernetes and on-demand microservices. There are also Edge+ digital signage solutions such as the DS-085 quad 4K-signage player, designed for energy efficiency and slim outline to meet the needs of the application environment.

There are various Edge AI systems, such as the AIR-101 optimised for visual AI with two Intel® Movidius™ VPUs and software including Edge AI Suite. For real-time AI inference and high-performance computing in industrial applications, the AIR-300 Edge AI system can be specified with up to a Core i7 CPU and provides a PCIex16 slot for a high-performance graphics card. There is also a large storage capacity to hold massive numbers of images, as well as sufficient bandwidth to handle data transmission.

Such a system is used in applications like industrial visual inspection, where traditional machine vision with rule-based algorithms often struggles to detect all types of defects. In addition, traditional systems cannot be easily updated to recognise new defects. 

Edge Intelligence in Action

Leveraging AI and deep learning technology in the AIR-300, system integrators working on an inspection solution for a ceramics company were able to overcome these familiar limitations and perform visual defect inspection in real-time. Sending captured images to the edge AI system for real-time inferencing enabled defective products to be quickly identified. The system can also be used as a local training server when the defect inspection system needs to be updated to inspect new products. AI-based defect inspection can be changed simply by preparing training datasets of the new defect types. hence updating the defect inspection system no longer requires the assistance of costly professional engineers.

AIR series edge AI inference system have also been deployed to assist epidemic prevention and containment in schools. Traditional facial-recognition technologies have suffered from low recognition rates and high costs, typically calling for a centralised server and facial recognition software. Cost is a key issue for schools, which usually cannot quickly raise the funds to finance large technical projects. Edge AI technology now allows facial-recognition models to be run on edge computers, which significantly reduces total cost while also sharpening system response times by capturing videos and enabling responses close to the event source. 

At the height of the pandemic, Advantech helped a Taiwanese system integrator create a combined facial-recognition and thermal-imaging solution leveraging the AIR edge AI system with Advantech VEGA-300 series AI-acceleration modules and the Advantech FaceView industrial app. Leveraging the high-performance computing platform, FaceView provides access-control, face-mask detection, and contactless body-temperature screening in real-time.

The edge computers, each connected to up to five infrared thermal-imaging cameras, scan entire classrooms automatically every few hours. The total solution came with APIs and helped the Taiwanese system integrator quickly integrate the school management system, student information database, and LINE push notification to record abnormal body temperatures and generate alerts as necessary.

By working with co-creation partners to create a broad selection of edge computing solutions, and with system integrators to help bring the right combinations together, Advantech is enabling the next generation of AIoT solutions to come forward. 

Conclusion

To meet performance and efficiency demands, intelligence in the IoT is moving out from central cloud services and increasingly into edge computing platforms handling real-time decision making and data filtering tasks. AI has a leading role in this emerging intelligent edge, and system integrators need help to deliver powerful, flexible solutions without needing deep expertise in neural networks or data science. Plug-and-play hardware and software building blocks that are also open and flexible are helping to build the new AIoT.