Cloud to Edge Engines



Designers targeting markets from medical to transportation to machine learning to digital signage are using Intel® processors to streamline data gathering and improve efficiency.

Controls and sensors are key components in automation and robotics. They provide the data for the intelligence around the network. In edge computing, processing that data is performed not in the Cloud but by thermostats, robots, or other devices located at the edge of the network.

ADLINK is evangelizing about edge computing for industry. CEO Jim Liu sees the company’s role as “First, connect the unconnected and then visualize, collect data and monitor in real-time to understand the factory status. Then analytics are used to add intelligence,” he tells EEcatalog.

Vision processing will increase, he believes, and it will become more important than data processing. He sees the Industrial Internet of Things (IIoT) as consisting of the Central Processing Unit (CPU) for data processing, the Graphics Processing Unit (GPU) for video and image, and a Field Programmable Gate Array (FPGA) for sensing functions. “This complicates the platform,” he says. “ADLINK does the data and video processing convergence to support future Artificial Intelligence (AI) apps.”

AI is going to move into every industry—automotive, home, and healthcare. For this, reliable, low latency data sharing and connectivity are critical.

Data and Image Processing
“ADLINK is supported by NVDIA so is in a unique position to understand how to put the CPU and GPU into an optimized architecture,” says Liu. He went on to say that partnering with vendors from across industries allows the company to work with technology partners to complete end-to-end solutions, whether AI, 5G, or sensors and to provide the last mile for system integrators, who are also partners, he adds.

The partnership with NVIDIA brings what the company terms ‘AI at the edge,’ combining embedded systems and connectivity with NVIDIA’s AI and deep learning technologies.

Figure 1: ADLINK demonstrated object recognition at this year’s Embedded World, in Nuremberg, Germany.

Examples are object recognition software and the sharing of data between mobile robots, or camera technology that can scan barcodes on irregularly shaped objects, differentiating shapes and objects for classification and traceability.

“AI is going to move into every industry—automotive, home, healthcare,” predicts Liu. For this, reliable, low latency data sharing and connectivity are critical.

At Embedded World 2018, ADLINK demonstrated that not only the monitoring of the device, but also its management, can be performed at the edge of the network. The company demonstrated an open source fog computing platform prototype that provides end-to-end virtualization of compute, storage, and communications from Cloud services to connected, embedded devices.

In March at the NVIDIA GPU Technology Conference in San Jose, the company showcased reference designs for ‘factory of the future,’ smart cities, and networking and communications. ADLINK’s Accelerated Multi-access Edge Computing Development Kit (AMEC DEVKIT) with custom GPU is for the design of telecom carrier-grade platforms to support edge computing. Also on display:

  • Smart camera technology for deep learning inference, defect inspection, and object classification
  • A smart city platform, calculating vehicle flow to improve traffic management
  • An overview of IBM’s Accessible Olli, a self-driving, electric, and cognitive shuttle to be used for mobility within a neighborhood.
    • Edge computing is responsible for processing the driving and sign language translation and Olli’s “360° 3D immersive bus ride” experience, says the company, although this is usually used in entertainment or marketing applications.

The Vision for the Future
Connect Tech offers 12 standard products based on NVIDIA’s Jetson TX1 and TX2 computing modules. Its mission is to bring AI to edge devices, such as robots, cameras, or enterprise collaboration devices, exploiting the low-power Jetson platform.

One of the most recent additions, the OrbittyBox, is a metal enclosure, consisting of two parts, designed to house the company’s Orbitty Carrier, Jetson TX2 or, for cost-sensitive applications, the Jetson TX1, and a heat sink.

For machine vision systems, the company also offers the Cogswell Vision System. It allows up to five Gigabit Ethernet (GbE) cameras to be connected, and four can be powered by on-board Power over Ethernet (PoE). The vision system is supplied with USB 3.0, USB 2.0, and USB on-the-go ports and RS-232, Mini PCIe, and mSATA ports for expansion.

The company also offers a custom design service to encourage innovation and encourages start-up companies to reach out to it for custom chassis and design support. Michele Kasza, Vice President of Sales and Marketing at Connect Tech, explains that the company’s design service can open up application areas. “The rugged, small form factor modules for NVIDIA Jetson TX1 and TX2 packages can be changed to accommodate the environment,” she says. She proposes they can be used in long distance vision systems, for example in agriculture, where they can be used in drones to identify weeds from plants and specific areas in which to apply pesticides or water. The TX2 in particular is designed to bring Artificial Intelligence computing to the edge of the network. The TX1 and TX2 are the same shape and size and use NVIDIA’s Maxwell GPU microarchitecture.

The company recently added a COM Express Type 7 + GPU embedded system to its portfolio (Figure 2). It combines Intel® Xeon® D, server class, x86 processors with NVIDIA’s Quadro and Tesla GPUs. The module measures just 8.5 x 6.45-inches or 216 x 164mm yet has two Mini PCIe slots and two M.2 slots, as well as interconnect options for 10GbE and GbE, USB 3.0, USB 2.0, HDMI, SATA III and I²C, with PC-style connectors.


Figure 2: Connect Tech integrates GPUs for vision systems at the edge of the network.

 

This addition to the company’s offering is also suitable for deep learning and AI applications. The module’s GPUs can be used for four independent display outputs or as a headless general-purpose GPU processing system using CUDA cores.

Another edge computing application was demonstrated by WinSystems, which announced the SBC35-C427 Single Board Computer (SBC) based on the Intel E3900 series processor for IIoT applications (Figure 3).


Figure 3: The Intel E3900-based SBC35-C427 from WinSystems targets industrial IoT applications.

The SBC, with up to 8-Gbyte of Error Code Correction (ECC) Double Data Rate 3 (DDR3) Random Access Memory (RAM), is designed for processing at or near the data sensor. It has onboard Analog to Digital Conversion (ADC) input, General Purpose Input/Output (GPIO) and serial interfaces to collect data for IIoT gateway applications, such as industrial control, smart transportation, energy, and digital signage.

The SBC is semi-customizable and can be configured to meet specific applications’ requirements without extending the time to market, points out the company. Available options are additional sensors and data acquisition choices through the Mini PCI Express, M.2 and the company’s modular I/O interfaces. Standard and custom modules can be added to collect data for edge processing and control.

The dual DisplayPort and Low Voltage Differential Signal (LVDS) interfaces support up to three video displays. There are also digital backlight and touchscreen control interfaces available.

Operating temperature range is -40 to +85°C. The SBC is also shock- and vibration tested, making it suitable for transportation, smart cities, and agriculture as well as industrial automation.

Industrial Strength
Capitalizing on the long-term availability of Intel processors, ADL Embedded Solutions has based its ADL120S industrial SBC on Intel’s sixth and seventh generation processors (Figure 4).

The SBC measures 4.72 x 4.72-inches or 120 x 120mm. It is based on the Intel Q170 chipset with the Intel HD Graphics 630 series graphics core. There are also 24 Execution Units (EUs) and up to 32-Gbyte DDR4 dual channel host memory, designed to support OpenCL computing to perform the increasing data volume and complexity of parallel tasks in industrial applications.

Figure 4: The ADL 120S industrial SBC is compact and thermally efficient.

It is designed for high resolution image and video pre-processing on signal acquisition industrial PCs or servers.

For maintenance-free operation in harsh or difficult to reach environments, the SBC has passive heat spreaders and heat pipes for cooling, which ADL claims is more efficient than conventional active cooling systems using fans. As a result, the operating temperature is -20 to +70°C, or with external cooling air supply it can operate between -40 to +85°C.

Manufacturers are responding to the increased vision processing that is the entry point to AI, initially in industrial applications, but which will extend to other areas where machine learning and deep learning will require edge computing to be reliable and efficient.


Caroline Hayes has been a journalist covering the electronics sector for more than 20 years. She has worked on several European titles, reporting on a variety of industries, including communications, broadcast and automotive.

Share and Enjoy:
  • Digg
  • Sphinn
  • del.icio.us
  • Facebook
  • Mixx
  • Google
  • TwitThis

Tags:

Extension Media websites place cookies on your device to give you the best user experience. By using our websites, you agree to placement of these cookies and to our Privacy Policy. Please click here to accept.