What you’ll learn:
- Advantages of AI in rugged industrial applications.
- How to identify critical design principles in extreme environments.
- What computing at the edge means for remote data processing.
When a system can run high-performance deep-learning-based inference engines, it can reliably perform advanced data- and video-processing tasks such as object detection and image segmentation of multiple video image streams. Data and video inputs can span the gamut in embedded systems, capturing streams through HD-SDI, Ethernet, USB3.0 cameras, and the like, interfaced through high-speed connectors.
Throughput speeds are increasing, with new system designs capable of 25 Gb/s per lane supporting high-speed PCIe Gen 3 and Gen 4 designs. And sensors will start to make use of 100 GbE to transfer in and between chassis. This increase in data transmission has not only positioned artificial intelligence (AI) to become a mainstream element in embedded computing, but it’s set the stage for AI to be used deep in the field as well as in more remote locations.
As advanced data-processing capabilities move farther away from a traditional data-center environment, factors such as temperature, dust, humidity and vibration become less controlled. Ensuring reliability in a more volatile environment is a huge contributing factor to the success of rugged computing that uses AI at the edge –and it requires a focused design approach to guarantee dependable operation (Fig. 1).
What’s Driving the Data Push?
Edge AI has taken off as one of the most notable trends in today’s technology innovations within embedded computing, since it allows system designers to run AI processes with less security concerns and reduces latency in data transmission. Platforms can react quickly to inputs at the sensors themselves, pre-process data, and more efficiently transmit relevant information across the computing network, resulting in exponential data throughput within the system itself.
The combination of powerful hardware, AI software frameworks, and real-time analysis on the edge empowers remote and mobile applications with these advanced capabilities. And by employing ruggedized hardware within harsh applications, users can achieve faster, more efficient solutions that are well-suited for the unique challenges and requirements of these environments.
Current Processing Implementations Using AI
Rugged AI platforms using the NVIDIA Jetson AGX Orin are delivering up to 275 TOPS, while incorporating numerous sites for standard and custom I/O as well as increased storage requirements.
The powerful Orin system-on-module (SOM) features a 2048-core NVIDIA Ampere architecture GPU with 64 Tensor cores and a 12-core Arm Cortex-A78AE, as well as deep-learning and vision accelerators plus a video encoder/decoder that all deliver impressive processing capabilities. The framework required by these platforms must withstand harsh environments to increase performance and enable reliable AI-based computing operations.
The new Jetson AGX Orin Industrial SOM provides impressive control over the power it consumes, and consequently the heat it generates, by providing states for four power budgets: 15 W, 30 W, 50 W, and up to a max of ~60 W. If an application doesn’t require a 100% duty cycle from the processor, the developer can set the module to a lower power budget and reduce the deployable system’s overall heat load. This can mean the difference between needing an active cooling kit (i.e., fans) or relying on passive, fanless convection cooling.
One Orin-based example is Elma’s JetSys-5330, designed for several rugged and mission-critical applications utilizing wireless multi-access edge computing as well as those in remote or outdoor environments. Easily configurable and expandable to meet mission requirements, this type of system can handle multiple concurrent video pipelines. Thus, it’s able to provide enhanced data handling and processing power for rugged systems, too, meeting the increased input streams of modern embedded computing (Fig. 2).
Advanced AI systems that facilitate data processing from the edge redefine the possibilities for using rugged, compact technologies in autonomous, harsh, and mobile environments. NVIDIA’s latest SOMs open the door to address new rugged deployable deep-learning and streaming I/O platforms as well as edge AI capabilities that provide operational efficiency and reduce risk and cost.
Design Considerations for Harsh Environments
In addition to server-class AI processing, today’s rugged-embedded-systems designers crave mission-critical SFF autonomy they can deploy in remote locations, while overcoming challenging connectivity. These systems need real-time responsiveness, minimal latency and low power consumption.
To properly design a rugged deployable AI-based system, one must evaluate the performance needs against the different options available to ensure that the rugged platform will reliably operate under even the most demanding situations (Fig. 3).
Using open-standards hardware architectures is a means to provide access to a wide range of interoperable and cross-compatible components with common operating systems and interfaces. This enables more flexibility and scalability in an integrated system and improves and empowers operational efficiencies, while ensuring rugged requirements are inherent in design.
Some other considerations that help define crucial rugged design parameters include:
Thermal profile and design considerations
Clearly applications in-the-field and in remote locations will require consistent operation across a wide range of temperatures. A mid-winter in northern Canada can expose vehicles to the −40°C range, while electronic systems in Australia or central Africa must contend with +45°C ambient plus the heat generated by the vehicle itself.
The options for managing high temperatures are widely known—convection, conduction, liquid- and air-flowthrough—to help dissipate both heat generated by the systems as well as heat from components themselves. But the low-end, −40°C requirement poses its own challenges.
Electrical components begin to act unpredictably when they’re started after a long cold soak, often disrupting designed-for power sequencing and simply not starting correctly. This can be addressed by introducing pre-heating mechanisms, but it requires time, power, and extra components. To manage any added costs or engineering to accommodate these parameters, it’s much better to design for a reliable cold start at these low temperatures.
Contaminant ingress and environmental elements
In addition to the temperature ranges, rugged systems need to deal with extremes of vibration and shock, humidity, rain and icing, blowing sand and dust, electrostatic discharge (ESD), and exposure to liquid contaminants such as fuel, hydraulic fluid and even something as mundane (but surprisingly damaging) as coffee or soda.
Requirements generally mandate that these systems operate in environments needing up to IP67-levels of ingress protection (no ingress of dust, and total immersion into 1 meter of water for 30 minutes). Additional testing for liquid contaminants is often a requirement, too.
Various standards, such as MIL-STD-810, define both the environmental requirements that electronic components must reliably operate in as well as the test methodologies needed to validate that the components do, in fact, meet those requirements. Ensuring a system meets these industry bellwethers can greatly improve the reliability of a rugged design.
Rugged AI for Tomorrow’s Intelligence Advantage
Advances in AI and computationally intensive data processing for deep learning are enabling faster adoption of capabilities like object detection and recognition, intelligent automation, and advanced navigation systems.
Increasingly, such processing is happening at the edge, meaning data is processed at the device location itself, instead of being offloaded to a centralized cloud-computing facility or separate data center. This provides a faster response to actionable data, significantly reducing latency in critical real-time applications, though it also puts strain on a system operating in a remote or harsh environment.
The freedom to place a system wherever it’s needed without concern for exposure to environmental contaminants is imperative for rugged systems deployed in remote areas. Rugged platforms generally live under a constant barrage of stressors, so long-term stamina and resilience is key. Reliable equipment should be tested to, and comply with, accepted industry standards and regulations to ensure system operation for long durations in extreme environments.
