Research & Design Hub Tech Trends

Box-Level Systems Marry AI and IoT

Figure 1 The Helix 310 and Helix 330 are highly customizable, fanless devices specifically engineered for Industry 4.0, edge computing and industrial IoT applications.
Written by Jeff Child

Intelligence at the Edge

We’re now entrenched in an era where artificial intelligence (AI) is easily done at an embedded systems level. That means system developers can implement local AI decision-making right at the IoT edge using box-level systems.

  • What’s happening in embedded box-level AI technology?

  • Industry 4.0, edge computing

  • Google Coral Edge TPU for AI

  • Fanless rugged edge systems

  • Intel Xeon computing

  • Remote deep learning

  • GPU-based AI systems

  • 5G and the IoT

  • Edge to cloud integration

  • AI and IoT for Drive Thru system

  • Helix 310 and Helix 330 from OnLogic

  • Kontron’s KBox A-203-AI-GC

  • American Portwell’s WEBS-21G0

  • Cincoze’s DS-1300 series

  • Adlink’s DLAP-211 series 

  • Axiomtek’s AIE900-902-FL

  • Advantech’s EI-52

  • Aaeon’s BOXER-8221AI 

Gone are the days when artificial intelligence (AI) was relegated to air-conditioned rooms filled with large computer servers. AI is now routinely done at the edge device level in Internet-of-Things (IoT) implementations. That said, the task of integrating sufficient compute-density remains a challenge because the demand for ever-greater performance is always on the rise.

One place where AI and the IoT have found a comfortable intersection is among embedded box-level systems. Over the past 12 months, vendors of those systems have rolled out a variety of products that offer high-performance computing—using a mix of different technologies—along with robust wireless networking for the IoT, and, in some cases support for interfacing with cloud-based architectures.

INTEL PSE SUPPORT

Exemplifying these trends, in August OnLogic announced two new fanless computing platforms powered by the Intel Celeron N and Pentium J series processors, formerly known as “Elkhart Lake.” The new Helix 310 and Helix 330 are highly customizable, fanless devices specifically engineered for Industry 4.0, edge computing and industrial IoT applications (Figure 1).

Figure 1 The Helix 310 and Helix 330 are highly customizable, fanless devices specifically engineered for Industry 4.0, edge computing and industrial IoT applications.
Figure 1
The Helix 310 and Helix 330 are highly customizable, fanless devices specifically engineered for Industry 4.0, edge computing and industrial IoT applications.

The units support triple independent 4K displays via DisplayPort interfaces, a 0°C to 50°C operating temperature range and a wide set of configuration options. Both the Helix 310 and Helix 330 embed either a dual-Core Celeron N6211 or Quad-Core Pentium J6425 processor. I/O includes 3x USB 3.2, 3x USB 2.0, and 2x COM ports. Networking is done via a 1Gb LAN port (2Gb LAN is available with Pentium CPU). Power input is 12V to 24V and 32GB of DRAM is provided. Optional features include digital I/O, two additional COM ports, CAN Bus and three additional antennas. The Helix 330 comes equipped with two additional Gb LAN ports standard.

IoT-specific features of the Helix 300 Series include the Intel Programmable Services Engine (Intel PSE). The Intel PSE is a dedicated offload engine for IoT workloads powered by an Arm Cortex-M7 microcontroller, which enables enhanced real-time computing. The Helix 300 Series also features OnLogic’s unique ModBay expansion technology, which allows users to customize systems with additional connectivity options via available M.2 and mPCIe slots.

The Helix 300 Series can be configured with a range of Windows or Linux Ubuntu operating systems and can be paired with OnLogic’s suite of OEM services, including custom branding, software imaging, custom fulfilment services and lifecycle management support.

— ADVERTISMENT—

Advertise Here

INTEGRATED CORAL EDGE TPU

The Google Coral Edge TPU (Tensor Processing Acceleration Unit) is one way to pack in the necessary compute-density needed for AI. In April Kontron announced an AI-platform as a KBox PC based on an Intel Atom x7-E3950 processor. Integrating a Coral Edge TPU, the KBox A-203-AI-GC is ready-to-use and can be integrated directly into existing x86 landscapes to quickly and efficiently implement application scenarios around machine learning (ML) and AI at the intelligent edge (Figure 2). This includes, for example, image and video data processing for visual inspection, quality assurance, predictive maintenance, sorting or object recognition—as a stand-alone device or gateway.

Figure 2 Integrating a Coral Edge TPU, the KBox A-203-AI-GC is ready-to-use and can be integrated directly into existing x86 landscapes to quickly and efficiently implement application scenarios around machine learning (ML) and AI at the intelligent edge.
Figure 2
Integrating a Coral Edge TPU, the KBox A-203-AI-GC is ready-to-use and can be integrated directly into existing x86 landscapes to quickly and efficiently implement application scenarios around machine learning (ML) and AI at the intelligent edge.

With the integrated Google Coral Edge TPU with up to 4 TOPS (trillion operations per second), the speed increases significantly compared to applications with simple USB cameras without TPU. In addition, the TPU supports small and low power applications and requires only 1W for 2 TOPS.

Unlike the Kontron AI-platform based on the NXP i.MX8M, which is available as a board product, the KBox A-203-AI-GC can be placed directly into the control cabinet or attached to a machine without further preparation. The gateway solution is based on the proven KBox A-203 system, which is used in practice for collecting, processing and distributing sensor data. It networks different end devices and ensures smooth and secure transmission to a higher-level server infrastructure. With the AI-accelerator, visual information can now also be processed. Two front USB 3.0 ports, connections for displays, two Gb Ethernet and an industrial 24 VDC power input are available as interfaces. In addition, there are serial interfaces and a Fieldbus option.

In the deep learning environment, the close conceptual connection of hardware and software is particularly important in order to optimally solve computationally intensive tasks. The AI-platform runs on Debian Linux, is compatible with TensorFlow Lite and pre-equipped with application examples. Users can benefit from the wide range of applications offered by Google-AI right from the start. On this basis, software solutions can be developed quickly and intuitively with a short time-to-market, based on pre-trained models for respective use cases.

FANLESS “BRICK” DESIGN

What the “edge” actually is can vary among different IoT implementations, but often it can mean a harsh environment where rugged computer gear is needed. Along such lines, in April American Portwell Technology released the WEBS-21G0, a fanless embedded system featuring the 8th generation Intel Core processor product family with a low 15W thermal design power (TDP) (Figure 3). Its rugged, compact design and high performance make the WEBS-21G0 well suited for applications in factory automation, automated test equipment, semiconductor equipment, robotics, unmanned vehicles, medical/healthcare equipment, digital signage, industrial IoT gateway and more.

Figure 3 The WEBS-2000 embedded system series is built upon a “brick” design concept to simplify system customization. It implements an intelligent structure by building the system chassis using three simple elements: wall, pillar and cover.
Figure 3
The WEBS-2000 embedded system series is built upon a “brick” design concept to simplify system customization. It implements an intelligent structure by building the system chassis using three simple elements: wall, pillar and cover.

The rugged WEBS-21G0 is equipped with Portwell’s NANO-6051, a Nano-ITX embedded board based on the 8th generation Intel Core processor product family. Processors available in this family combine low power consumption with high processing power and improved performance compared to previous generation processors. The WEBS-21G0 also features DDR4 2400MHz non-ECC SO-DIMM up to 32GB and storage interface in the form of 1x M.2 Key M 2280 socket for SSD.

For functionality extension, it provides 3x USB 3.2 Gen 1 on rear I/O to ensure fast data transmission with low-power consumption. The WEBS-21G0 has a M.2 Key E 2230 socket for wireless module connectivity including Wi-Fi and Bluetooth, making it well suited for for communication and IoT applications. Intel I210AT and Intel I219LM Ethernet controllers provide dual Gbit Ethernet LAN access via the two RJ-45 connectors. The is also a RS-232/422/485 post that is selectable by BIOS adjustment.

The Intel Gen 9.5 graphic engine supports dual Mini DisplayPort (DP) on rear I/O with resolution up to 4096×2304. It provides a selection of multiple connections such as displays, graphic cards, cameras, storage and more on the same system. Moreover, WEBS-21G0 offers audio port and DC 12V input on rear I/O, on-board TMP 2.0 for application security and supports multiple operating systems including Microsoft Windows 10 IoT Enterprise, Ubuntu, real-time Yocto Project (YP) and Wind River.

The WEBS-2000 embedded system series is built upon a “brick” concept design created by Portwell to simplify system customization. It implements an intelligent structure by building the system chassis using three simple elements: wall, pillar and cover. This design of the WEBS system series enables flexible and easy customization. The Portwell WEBS-21G0 further benefits from the brick concept chassis with a top cover/heatsink mechanism that facilitates efficient heat dissipation. The rugged and compact WEBS-21G0 supports a temperature range from 0ºC to 50ºC, while at the same time, its fan-less design ensures silent operation, reliability and low maintenance rate and costs.

— ADVERTISMENT—

Advertise Here

XEON-BASED SYSTEM

Combining AI-level processing and IoT functionality requires high-performance processing in an embedded form factor. That can mean designing server class computing—like Intel’s Xeon processor—into an embedded box-level system. Cincoze did just that in June, rolling out a 10th generation Intel Xeon/Core high-performance, high-expansion, rugged industrial computer: the DS-1300. The three-model lineup includes the DS-1300, DS-1301, and DS-1302, which vary by the number of PCIe expansion slots. The series is especially suitable for deployment in industrial equipment, AIoT, logistics automation, cobots and applications requiring high computing power in harsh environments (Figure 4).

Figure 4 The DS-1300 series can support a 10th generation Intel Xeon/Core (Comet Lake-S) CPU, up to 10-core at 80W. It supports two sets of DDR4 SO-DIMM for a total of 64GB, and storage options include a set of high-speed M.2 NVMe storage slots, three sets of mSATA slots and two sets of 2.5" hard drive trays
Figure 4
The DS-1300 series can support a 10th generation Intel Xeon/Core (Comet Lake-S) CPU, up to 10-core at 80W. It supports two sets of DDR4 SO-DIMM for a total of 64GB, and storage options include a set of high-speed M.2 NVMe storage slots, three sets of mSATA slots and two sets of 2.5″ hard drive trays

The DS-1300 series can support a 10th generation Intel Xeon/Core (Comet Lake-S) CPU, up to 10-core at 80W, representing a 31% performance increase for multitasking compute-intensive applications. It supports two sets of DDR4 SO-DIMM for a total of 64GB, and storage options include a set of high-speed M.2 NVMe storage slots, three sets of mSATA slots and two sets of 2.5″ hard drive trays. One set of drive trays is arranged in the front maintenance area for emergency access, providing fast storage and convenient repairs.

The DS-1300 series supports up to two PCI/PCIe expansion slots, a 110W power supply capacity and can be connected to various commercially available high-speed I/O cards, image capture cards, motion control cards or GPU cards. In order to strengthen the stability of the expansion card, Cincoze’s patented “Adjustable PCIe Retainer” uses a two-stage precision adjustment mechanism that can be tightly adjusted and strengthened according to the size of the external card (up to 235mm × 111mm), providing reliable support under high-vibration environments, and maintaining safe and reliable continuous operation at the site.

The DS-1300 series has an array of native high-speed I/O, including 2x GbE LAN, 2x COM, 6x USB3.2, 2x USB2.0 and more. It can also be used with Cincoze’s modular design CMI, MEC and CFM, for additional functions. CMI and MEC provide up to 12x GbE LAN, 2x 10GbE LAN, 32x Digital I/O, 8x M12, 4x COM or 4x USB3.2. CFM modules can add power ignition sensing and PoE functions. In addition, there’s a built-in MiniPCIe expansion slot that supports commercially available Wi-Fi/4G/GPS modules.

The full DS-1300 range is certified against shock and vibration, adhering to the stringent MIL-STD-810G military equipment standard. It also meets the specifications of rail standard EN 50155 (EN 50121-3-2 only) and maintains the tough design of the rest of the Diamond series. It supports wide operating temperature (-40°C to 70°C), wide voltage input range (9VDC to 48VDC) and boasts overvoltage, overcurrent and electrostatic discharge protection.

REMOTE DEEP LEARNING

Deep learning is a powerful technology, but an ability to do it over networks to remote locations makes it even more powerful. Facilitating exactly that, in August Adlink Technology entered into a partnership with Allxon to offer a remote device management solution for the Adlink DLAP-211 series of deep learning acceleration platforms (Figure 5). The partnership has enabled an online portal for easy monitoring, control and update of a DLAP-211 anytime, anywhere, without an onsite visit.

Figure 5 After a DLAP-211 is deployed at the edge, IT can use Allxon Device Management Solutions (Allxon DMS) to remotely perform a number of functions, such as managing devices according to group settings and custom-set polices.
Figure 5
After a DLAP-211 is deployed at the edge, IT can use Allxon Device Management Solutions (Allxon DMS) to remotely perform a number of functions, such as managing devices according to group settings and custom-set polices.

After a DLAP-211 is deployed at the edge, IT can use Allxon Device Management Solutions (Allxon DMS) to remotely perform a number of functions. It can monitor CPU usage and GPU performance and receive alerts on DLAP-211 performance. It can manage devices according to group settings and custom-set polices. The system also collects logs for troubleshooting. It also controls DLAP-211 reboots, scheduling, screenshots and commands. Over-the-air updates can be performed for applications, containers, AI models and firmware

As an example of the system in action, consider security and performance improvements that may require a system update. IT can provision a deployed DLAP-211 to run different applications or update DLAP-211 with improved AI models for feature, speed and accuracy enhancements. As deep learning acceleration platforms deployed at the edge are scattering across different locations, countries or even continents, large-scale remote manageability keeps deep learning acceleration platforms healthy and up to date, avoiding the tremendous time and efforts onsite visits require.

Adlink’s deep learning acceleration platform, the DLAP-211 series, has versatile AI for applications at the edge, including factory floors, logistic centers, warehouses, restaurants, fish farms and more. Allxon DMS will be supported on the DLAP-211 so that the edge platform can be efficiently monitored in real time, and managed and updated remotely.

GPU-BASED SYSTEM

GPUs like Nvidia’s Jetson Xavier have become a well-established processing choice for AI at the edge. Using that technology, in July Axiomtek introduced the AIE900-902-FL, its edge AI computing system (Figure 6). This edge AI system uses the Nvidia Jetson AGX Xavier platform, which has an 8-core NVDIA Carmel Arm v8.2 (64-bit) processor and a 512-core Nvidia Volta GPU with 64 Tensor cores. The AIE900-902-FL is well suited for AI-powered autonomous machine applications such as 3D vision guided robots, autonomous mobile robots (AMR), intelligent video analytics, domain-focused robot assistants, intelligent roadside units and more.

Figure 6 The AIE900-902-FL edge AI computing system uses the Nvidia Jetson AGX Xavier platform, which has an 8-core NVDIA Carmel Arm v8.2 (64-bit) processor and a 512-Core Nvidia Volta GPU with 64 Tensor cores.
Figure 6
The AIE900-902-FL edge AI computing system uses the Nvidia Jetson AGX Xavier platform, which has an 8-core NVDIA Carmel Arm v8.2 (64-bit) processor and a 512-Core Nvidia Volta GPU with 64 Tensor cores.

The AIE900-902-FL features a rugged design for operation within harsh environments and supports a wide temperature range of -30°C to +50°C and vibration of up to 3Grms. It comes with four PoE ports and two LAN ports for 3D LiDAR and high-speed intelligent video surveillance applications. This embedded system is equipped with a 32GB 256-Bit LPDDR4x onboard and has one M.2 Key M 2280 SSD slot with a PCIe x4 NVMe interface, one Micro SD slot and one 2.5″ SSD/HDD drive bay for massive data processing and AI applications.

The system also supports one full-size PCI Express Mini Card slot (USB + PCI Express signal), one M.2 Key E 2230 slot and one SIM slot for 3G/4G, GPS, Wi-Fi and Bluetooth connections. To reduce the effort and shorten the deployment process, the Nvidia JetPack has been pre-installed on the edge AI system for quick development. In addition, the AIE900-902-FL is certified to CE and FCC Class A.

The AIE900-902-FL offers multiple I/O options including two lockable HDMI 2.0 ports with 4k2k supported, two 10/100/1000 Mbps Ethernet, four 10/100/1000 Mbps PoE, two USB 3.1 Gen2 ports, two USB 3.1 Gen1 ports, two USB 2.0 ports, one Micro USB port, one 8-CH DIO, two RS-232 default (or 2x CAN by jumper settings) and four SMA-type antenna openings. There are eight LED indicators showing for power/storage/LAN/PoE active status alert. The advanced system also features one recovery switch and one 24VDC power input connector.

5G AND THE IoT

In many ways, 5G cellular technology is more about IoT and machine-to-machine interfacing than it is about smartphones. In June Advantech released its latest edge intelligence system: the EI-52. This compact, high performance system leverages an 11th Gen Intel Core i5/i3/Celeron processor and plug-and-play system design. EI-52 is designed for edge-to-cloud interconnection and 5G + AI solutions.

The EI-52 comprises a hardware and software integrated package with EdgeX Foundry IoT plug-and-play open software framework and Advantech’s WISE-DeviceOn IoT edge intelligence software (Figure 7). EI-52 empowers 5G and AI applications by supporting AIW 5G modules, Wi-Fi kits, VEGA AI acceleration modules and the FaceView facial recognition I.App. Advantech’s EI-52 speeds AIoT smart application deployment, data acquisition application development and remote device management.

— ADVERTISMENT—

Advertise Here

Figure 7 The EI-52 comprises a hardware and software integrated package with EdgeX Foundry IoT plug-and-play open software framework and Advantech’s WISE-DeviceOn IoT edge intelligence software. EI-52 empowers 5G and AI applications.
Figure 7
The EI-52 comprises a hardware and software integrated package with EdgeX Foundry IoT plug-and-play open software framework and Advantech’s WISE-DeviceOn IoT edge intelligence software. EI-52 empowers 5G and AI applications.

Advantech’s EI-52 is powered by a 11th gen Intel Core i5/i3/Celeron processor and supports Windows 10 IoT or Ubuntu 20.04. This compact, high- performance system (156mm × 112mm × 60mm) includes diverse I/Os ports such as 2x GbE, 2x COM, 6x USB, 8/16GB DDR4 memory and a 64GB SATA slim SSD. It provides broad operating temperatures at -10°C to 50°C. EI-52 is also integrated with Advantech’s WISE-DeviceOn. This software supports zero-touch onboarding, remote device monitoring/management and visualized user interfaces. When combined with a plug-and-play design, these features help IT operators remotely monitor and manage EI-52 in real time, thus reducing deployment times in diverse applications.

EDGE-TO-CLOUD INTEGRATION

Advantech’s EI-52 features an edge-to-cloud integrated architecture that shortens development time. Using the pre-installed Edge X on EI-52 enables developers to avoid delving into different data formats and device APIs for connecting edge data to cloud services. Edge X supports over 15 types of protocols for sensing devices—including common OPC-UA/Modbus/REST—and has been tested on more than 20 heterogeneous devices. It further provides open device SDK to ease device integration using proprietary protocols.

This intelligent system enables edge data pre-processing and analysis, and comes with pre-configured tools for quick connection to Ali/AWS/Azure cloud applications. This solution also provides the software/hardware integration capabilities and microservices architecture needed to shorten development times by up to 50%, augmenting the efficiency and flexibility of importing system applications.

Advantech’s EI-52 is compatible with a selection of optional packages that enable 5G, AI and quick-start AIoT applications. The EI-52 can support an AIW-355 5G module with an optional thermal kit. This solution can also support additional wireless capabilities using the EWM-W189H02E Wi-Fi module with Wi-Fi 802.11ac/a/b/g/n and Bluetooth 5.0. The EI-52 can use VEGA-330 AI acceleration modules with Intel Movidius Myriad X 2x VPU to empower AI applications. This low-power module equips EI-52 with additional AI inference computing power. Finally, EI-52 supports Advantech’s FaceView facial recognition I.App—enabling contactless access control applications for home offices and smart buildings, as well as VIP management for self-service kiosks.

AI AND IoT FOR DRIVE THRU

In example of AI and IoT in action, in September Aaeon announced that it teamed up with a US manufacturer to help develop and deploy an AI system designed to help speed up drive-thru service at a wide range of businesses. The system is powered by the BOXER-8221AI compact edge system with Nvidia Jetson Nano.

With the vast majority of communities in America designed around car-centric transportation, drive-thru service is common across a wide range of businesses, including fast food, banks, pharmacies, coffee shops and even florists, says Aaeon. During high-traffic times, such as lunch and rush-hour, lines in drive-thru lanes can often become quite long. The current pandemic has exacerbated the situation with many businesses choosing to close their lobbies or dining rooms in favor of drive-thru only. Even as governments seek to re-open, businesses are maintaining these practices due to cautious health practices or labor shortages causes by the health crisis.

Aaeon says these issues cause what were already long lines prior to the pandemic to become even longer, with both customers and delivery drivers having to wait in long lines for food, which can cause impatient drivers to leave before they are served, and increase air pollution from idling vehicles. One company in the US is helping to alleviate the situation with a solution that uses AI to help speed up drive-thru service.

Aaeon and its partner’s solution leverages both AI technology and big data analysis to help with allocation of workers and tasks by predicting customer demand based both on the number of vehicles in line, but also the type of vehicles. By cross referencing vehicle types with a database of past orders, the system can promote the kinds of meals or purchases the driver is more likely to make, reducing ordering time, and further reducing the time spent in line. Businesses can set conditions based on vehicle type to also help create a preparation plan before customers even place their orders.

Figure 8 A US manufacturer used the BOXER-8221AI from Aaeon, along with a camera to perform AI analysis at the drive-thru ordering kiosk.
Figure 8
A US manufacturer used the BOXER-8221AI from Aaeon, along with a camera to perform AI analysis at the drive-thru ordering kiosk.

The manufacturer of the system uses the BOXER-8221AI from Aaeon, along with a camera to perform AI analysis at the drive-thru ordering kiosk (Figure 8). The BOXER-8221AI is powered by the Nvidia Jetson Nano, providing energy efficient performance that delivers processing speeds up to 472 GFLOPS, perfect for object recognition and data processing at the edge. The BOXER-8221AI is compact and features a dust-resistant fan-less design, with flexible I/O loadout, well suited for deploying at the edge, wherever it’s needed. 

RESOURCES
Aaeon | www.aaeon.com
Adlink Technology | www.adlinktech.com
Advantech | www.advantech.com
American Portwell Technology | www.portwell.com
Axiomtek | https://us.axiomtek.com
Cincoze | www.cincoze.com
Kontron | www.kontron.com
OnLogic | www.onlogic.com

PUBLISHED IN CIRCUIT CELLAR MAGAZINE • NOVEMBER 2021 #376 – Get a PDF of the issue

Keep up-to-date with our FREE Weekly Newsletter!

Don't miss out on upcoming issues of Circuit Cellar.


Note: We’ve made the May 2020 issue of Circuit Cellar available as a free sample issue. In it, you’ll find a rich variety of the kinds of articles and information that exemplify a typical issue of the current magazine.

Would you like to write for Circuit Cellar? We are always accepting articles/posts from the technical community. Get in touch with us and let's discuss your ideas.

Sponsor this Article
Former Editor-in-Chief at Circuit Cellar | Website | + posts

Jeff served as Editor-in-Chief for both LinuxGizmos.com and its sister publication, Circuit Cellar magazine 6/2017—3/2022. In nearly three decades of covering the embedded electronics and computing industry, Jeff has also held senior editorial positions at EE Times, Computer Design, Electronic Design, Embedded Systems Development, and COTS Journal. His knowledge spans a broad range of electronics and computing topics, including CPUs, MCUs, memory, storage, graphics, power supplies, software development, and real-time OSes.

Supporting Companies

Upcoming Events


Copyright © KCK Media Corp.
All Rights Reserved

Copyright © 2022 KCK Media Corp.

Box-Level Systems Marry AI and IoT

by Jeff Child time to read: 14 min