The Industrial world is generating data.
Machines can report on their own operation. Sensors can measure, detect, and monitor. Vehicles and equipment can share information about their location or movement. Production and enterprise systems can capture events as work flows through them.
However, just having a connected device does not mean having useful information. An interesting aspect of AIoT (Artificial Intelligence + Internet of Things) is what happens after information is generated: the connection between physical world facts and analytical insights which identify patterns, potential issues, or other information of interest related to the physical devices.
AIoT Data Flow
A useful way to think about an AIoT architecture is as follows:
Physical Assets
↓
Sensors
↓
Connectivity
↓
Data
↓
AI / ML
↓
Insights
↓
Decision
↓
Action
As you can see, each level is responsible for different aspects of processing information coming from physical assets:
Sensors: typically capture information about the physical world.
Connectivity: provides transport for data between nodes in the systems, including edge systems, gateways, clouds, and more.
Data Systems: capture, store, process, and manage the insights extracted from physical sources.
AI/ML: can process this data to find patterns, potential problems, or other information of interest.
People and Operational Decisions: make decisions and take actions based on the information discovered by data scientists and analysts.
This is important because AIoT is not simply a matter of putting an AI model on a connected device.
AIoT and Raw IoT Data
Let us imagine a scenario in which we have an industrial machine that captures thousands of values throughout the day. The data might be represented in a dashboard with values such as temperature, vibration levels, machine status, and others. This provides some insight into what is going on, but someone would still have to review this information. Instead, we can also provide an AI model which can find patterns of interest: for example, a change in vibration plus a change in temperature could be an indicator of a developing issue with the equipment. The model itself would not stop the machine, but it could draw attention to the fact that something may be happening.
The same approach could be used in a variety of industries and for a wide range of use cases. Connected equipment in a manufacturing plant provides some information about the machinery and production processes. Connected vehicles and equipment can share information about the logistics chain, such as what is being transported and where it is located. Sensors in utility equipment can capture data about their current operating conditions.
The type of information captured and how it is processed will ultimately dictate the type of value that an AIoT initiative will deliver. However, there is one common factor: starting with a problem or a use case.
Thinking about Engineering
Another important point about creating an AIoT initiative is that there are a number of factors which must be considered as part of the implementation. This includes aspects such as:
Data Quality: the quality of information coming from physical assets will impact downstream processes. Missing or corrupted data, missing or incorrect timestamps, or values unrelated to the subject of analysis can impact an AIoT initiative.
Connectivity: industrial environments tend to have a variety of devices with different connectivity needs. Connectivity options can impact decisions about the network infrastructure.
Integration: data collected from these systems will need to be processed and analyzed, possibly with the help of other systems.
Security: exposing physical devices to networks can introduce new attacks surfaces and vectors. Security considerations such as perimeter security, data encryption, and identity management should be considered.
Human Oversight: an analytical result is not always an operational decision. People and established operational procedures should still be considered, especially if an initiative requires taking an action which impacts the physical world.
All of this means that thinking about an AIoT project should start with a particular problem or a use case to address: what operations area is impacted? What do we want to learn about the physical world around us?
From there, it will be possible to determine the best data sources (sensors, equipment, or other connected systems), decide on the type of analysis needed, and design the rest of the AIoT architecture around it. There are a number of other considerations to take into account, but a strong foundation in the operational problem or use case is an essential step in building an impactful AIoT initiative.
For more information on the subject, you can refer to Aperture Venture Studio's article on Connected Industrial Data Intelligence .
In particular, you may find it informative to understand how the opportunities around AIoT are formed and understood in the context of an industrial environment.
Top comments (0)