How are companies using connected devices and real-world data to engineer intelligent systems?
Think about a manufacturing plant with hundreds of pieces of machinery, each producing a steady stream of temperature and performance data. Getting that information is only part of the equation, the greater challenge lies in understanding it and knowing how to respond.
This is where the Internet of Things (IoT) and Artificial Intelligence (AI) come into the picture. These two technologies work in concert IoT gathers and shares data; AI analyzes and understands it. Artificial Intelligence of Things (AIoT) is the emerging space where AI and IoT converge.
What are companies building with AI and IoT? How do these technologies complement one another? What are typical AIoT implementations?
These are the questions we will address by building the foundation of understanding.
1. Companies Shaping the Space
There are a variety of established technology companies in the AI, IoT space, many of which are building computing infrastructure, robotics, or intelligent systems.
NVIDIA: Compute and Graphics Architecture
NVIDIA makes GPUs, AI computational software, and tools for robotics and computer vision.
As an application developer, one would consider NVIDIA-based AIoT applications for robotics and computer vision use cases around warehouses, fulfillment centers, or even autonomous vehicles.
These applications range from performing inferencing based on imagery and sensor data to using vision to find a defect on a manufacturing line what those applications can do are limited only by the training data, context, and environment in which this AIoT application would be deployed.
Microsoft: Software, Enterprise Intelligence
Microsoft builds some of the world’s largest and most influential cloud and enterprise systems in the world, and they power many of the tools we use on a daily basis.
IoT architectures often feature a cloud component that allows the connected products and sensors to get updates, report statistics, and connect business intelligence tools together. Microsoft components could enable that IoT architecture one would consider them when designing a system that could analyze data arriving from thousands of connected products and bring it together into a meaningful whole.
The choice of architecture would be informed by the data, size, scalability expectations of the system, and business needs.
Siemens: Industrial Automation
Siemens makes industrial automation products, tools, and systems. Industrial applications usually involve the convergence of IT and OT — operational technology, or connected devices and machinery.
By analyzing this data, teams could monitor performance, identify issues and bottlenecks, or improve their understanding of what is happening on the factory floor. That is important to developers because an industrial application typically runs in a more complex environment which requires deeper domain expertise about how systems, processes, and devices function.
IBM: Enterprise Systems and Tools
IBM’s offerings span across enterprise platforms, big data, automation, and industrial applications.
By combining those, IBM enables organizations to analyze their data in novel ways, find anomalies and patterns that suggest something unusual is happening in their business, but the results of that investigation need to be interpretable by people within the system.
Google: AI and Language Models
Google’s end to end systems power much of the modern world, from cloud infrastructure hosting to AI and machine learning architectures.
These would likely serve as the backbone of an AIoT application that requires data science, advanced analytics, or perhaps another external application that would connect to this system. One would choose this architecture based on the particular AIoT use case and data requirements.
The choice of architecture would be informed by the use case, current infrastructure, cost expectations, security needs, team expertise, and not brand preference.
This is a highly simplified architecture, but the general idea follows one designs the system around the data and what they want to achieve.
2. AI and IoT in Concert
A typical AIoT architecture has elements of physical machinery, and actions, and the software and AI that analyze the data and suggest or execute actions in the real world.
Physical World –> Sensors –> Connectivity –> Data processing –> AI / ML –> Decision –> Action
Within this architecture one can see six stages:
Sensors –> these measure physical quantities like temperature or vibration
Connectivity –> some form of transmission, either physical or wireless, that moves this data to a device or cloud
Data processing –> where this data is validated, filtered, organized, and prepared for analysis
AI and ML –> models analyze and detect patterns in data, identifying known events or forecasting likelihood of an undetermined event
Decisioning –> decision logic analyzes model output and operational constraints and identifies recommended or required actions
Action –> recommendation, alert, or an automated action executed by this system
Not every AIoT system includes AI. Often, decision trees and basic logic can accomplish what an AIoT application would attempt but a simple threshold value could suffice when a particular physical condition is met.
One builds an AIoT application the same way as any other: identifies a situation where current tools, actions, and technologies are insufficient an unmet need and then designs a system that delivers value.
The application might use AI or it might use simple logic what is more appropriate to this setting?
3. A Working Example
Let’s consider a working example: predictive maintenance within a manufacturing plant.
There are a number of large motors that turn during the manufacturing process, and attached to them are vibration and temperature sensors. Every so often, these sensors transfer their information to an edge device or cloud, where a machine learning model analyzes if these measurements are normal or not then it notifies the maintenance team if there is a concern.
Such applications need to take context into account, be it physical location or operational constraints.
Other concerns also become relevant data integrity and quality, latency or responsiveness if the connection is slow, false positives reducing credibility of these alerts,
security of this connection and the data itself, and possibly human review before an issue is escalated. All these are practical concerns when implementing the application. It is not enough for an AIoT application for it to simply work it should function reliably and consistently under a variety of environmental conditions.
There is also the question of edge AI versus cloud AI one can implement the system so that sensor data is processed near the edge, where it receives near real time updates based on physical conditions, limited bandwidth, and other operational constraints. Alternatively, one could implement a system that uses the cloud, where the processing power and memory is less constrained, though there might be latency. Many applications use a combination of both, with edge processing being used for immediate concerns and the cloud providing longer-term insights, analysis, and reporting.
The choice between cloud AI, edge AI, or a combination would depend on the needs of the application, its response and latency requirements, budget, privacy, and reliability expectations.
4. Looking Ahead at AI and IoT
Artificial Intelligence in Industrial Applications: Intelligent, Connected, and Adaptive AI and IoT are making that possible across a new range of applications and are enabling the rise of Physical AI: systems that are aware of and adaptive to the physical world. This spans sensing and perception of equipment and surroundings to advanced logistics or process intelligence. These are the technologies that can make systems around asset availability, safety protocols, and even the very nature of industrial processes evolve. This kind of transformation is what companies such as Aperture Venture Studio are pursuing as an area of strategic interest by building AIoT companies in physical industrial markets.
That being said, there is a long road ahead before such an application sees production. There is a great deal more than choosing a technology stack, but it is always best to start at the beginning: by identifying the problem the technology will solve.
Conclusion
Companies such as NVIDIA, Microsoft, Siemens, IBM, and Google, each offer distinct capabilities that power intelligent applications, enabling modern developers and organizations to accomplish a great variety of objectives. There is no single application or purpose that these technologies serve but rather, they complement a growing array of use cases and domains. Their selection becomes informed by what the application requires to accomplish its stated purpose. These are the concepts we explored in this series.
When designing AIoT system, what considerations would you give to the choice between edge AI versus cloud AI? Why?
Top comments (0)