Artificial intelligence is frequently associated with applications that take the form of software. There are, of course, plenty outside of that sphere.
Factories, warehouses, transportation logistics, and manufacturing facilities create and consume a tremendous amount of data on a daily basis. Physical sensors measure, detect, and otherwise quantify the conditions of the environment, the machinery, and other criteria.
The question is “Do Something With It,” isn’t it?
Enter the concept of AIoT — using IoT infrastructure, technologies, and capabilities to create value from this information. In other words, connecting everything from AI to analytics to operations and logistics, in order to understand exactly what’s going on in the physical environment and using that information to optimize performance.
The IoT Side Of The Equation
An IoT-based system typically has:
• Sensors
• RFID readers and tags
• Cameras
• GPS or location tags
• The equipment itself
• Edge devices
• Gateways to data repositories
• And more
These devices and sensors provide the data that will then be used by AI and/or analytics processes to detect and predict trends and behaviors.
If a particular system’s purpose is to manage the physical logistics of an operation, the data it generates might tell you (or the next layer in the process) where an asset was, where it is now, and when it moved. If it’s a system that’s focused on monitoring equipment, it might provide vibration and temperature data, for example.
But having a dashboard or collection of data is only part of the puzzle. What can you do with it?
That’s where “AI” comes in.
AI Adds An Additional Layer
AI and machine learning processes and algorithms are useful and sometimes necessary. To analyze a vast amount of information, detect patterns that would otherwise be time-consuming or nearly impossible to identify by human eyes, and act on those insights.
Here’s a generalized sample architecture:
Physical Environment
– Sensors / RFID / Devices
– Edge Devices & Gateways
– Collect / Integration
– Analytics / ML
– Insights / Alerts
– Operational Side – Take Some Action
Keep in mind that this is a generalized look at how these technologies might come together. There may be other aspects, depending on the overall scope of the project.
Let’s look at one potential example.
Predictive Maintenance For Machinery
One typical example is predictive maintenance — in this case, how AIoT can be used to detect issues with machinery, based on a combination of current and past data. That way, repairs can be made before a total system failure occurs, and/or the engineering and operations team can take preventive measures as needed.
This application might work like this:
Sensors measure relevant physical properties — vibration and temperature, for example.
This data is collected and analyzed by an AI process
Past data is compared, and an anomaly is noted.
The maintenance team is alerted
The maintenance team takes the appropriate action
The truth is this particular application might not require the most sophisticated AI process. A more simplistic approach might be used.
Asset Tracking Across Systems Or Facilities
Industrial companies and manufacturing facilities may have a variety of assets, from tools, to parts and containers to entire pieces of machinery and equipment. It can be challenging to keep track, not only of where they all are but to make sure that physical assets match their digital twins.
It can be useful for the digital side of operations to “see” where an asset actually is in the real world.
By adding RFID tags and other sensors to assets, data can be collected. AIoT adds another layer. It can potentially detect unusual movement or patterns and provide additional insights regarding specific assets that are either misplaced often or not used.
One is strictly logistics or operations; the other is analytics. The two can work together in tandem.
Data Quality Can Easily Be The Major Challenge
The primary risk or failure point in any AIoT initiative is often data quality. The AI algorithm itself probably isn’t the issue — more often, it’s the quality of data being fed in. Some common data-quality challenges include:
• Missing sensor readings or erroneous timestamps
• Sensor accuracy and calibration issues
• Connectivity errors or failures
• Duplicate data entries
• Inconsistency in device or sensor ID tagging
• Legacy systems that weren’t designed to communicate with other systems
• Lack of historical data for training or identifying patterns
This is just another reason why the data collection layer is just as important as the actual AI process.
AIoT Requires More Than A Machine-Learning Algorithm
Edge-Based Infrastructure Can Enhance An AIoT Architecture
The AIoT stack typically relies on the IoT infrastructure, which generates and captures massive amounts of data. There may be situations — especially if there’s a need for low-latency or localized processing and decision-making — where it’s beneficial to move some of that data to the edge.
A simple example might be:
Sensors
– Edge Gateway
– Local Processing
– And Then? Immediate Alert
Plus:
Selected Data Stream
– Edge Gateway
– Into The Cloud
– For Broader/Deeper Analysis
Latency, bandwidth, and overall scope all play roles in determining whether and when this makes sense.
Start With The Question — It Will Drive Your Technology Stack
One of the most common pitfalls in developing AI processes or AIoT applications is to start with the AI. Begin with the question or the potential benefit. Then determine the data, processes, and infrastructure you need in order to achieve the outcome you want.
Problem -> Data -> Infrastructure -> Analytics -> Action
For example, if a company’s primary concern is predictive maintenance for certain equipment, the relevant data would be temperature and vibration data for that equipment. That would drive the infrastructure decisions — what sensors to use and how to deploy them. That leads to the analytics piece, identifying what kind of model would be helpful. And it culminates in taking the relevant action. And by building it in that order, it’s easier to ensure that the technology is being used to drive real value.
Security Needs To Be Part Of The Conversation
A typical aspect of implementing any kind of AIoT initiative is networking existing or new devices and sensors onto a common platform. That might introduce new security considerations. The AIoT architecture must account for physical security, network security, and more.
There isn’t always a perfect balance between the level of security and the resources required to implement and maintain it. That said, security must be considered at the architectural design stage, not an afterthought.
AIoT And Its Real-World Applications
This space is all about the connections between artificial intelligence and the physical world. That means an array of technologies, from sensors and GPS to RFID tags and the “edge.” All of this connects to analytics, data, and machine learning processes.
The companies involved in this space include some that are using AIoT technologies to drive insights across a range of applications, from asset visibility to industrial intelligence. But the common thread runs through that fundamental connection — what can be learned from the data about the physical environment, and how that drives insights and decision-making in real-world operations. Aperture Venture Studio
The question should always be: how does this data serve to answer a question or lead to an insight?
Top comments (0)