AIoT — the union of Artificial Intelligence and Internet of Things — is becoming increasingly important to industrial software.
IoT provides the connectivity layer: sensors, RFID, location, machines, and other physical assets generate operational information.
AI provides the intelligence layer: algorithms can analyze the data, find patterns, find outliers, and support decisions.
The interesting question is what happens if these technologies have to work together in a real manufacturing environment.
The basic AIoT architecture
A simplified view of the AIoT architecture can be represented by multiple layers:
Physical Environment
↓
Sensors / RFID / RTLS / Machines
↓
Edge / IoT Connectivity
↓
Data Processing
↓
AI / Analytics
↓
Business & Manufacturing Systems
↓
Operational Decisions
Each layer plays a certain role: sensors and tracking technologies gather data from environment, IoT or edge infrastructure transports and processes the information, AI and analytics turn it to usable knowledge, and the final information becomes useful to operational decisions.
Why manufacturing makes AIoT interesting
A manufacturing facility is, essentially, a distributed physical system.
Materials are transported from location to location, machines are running, work-in-progress passes from one stage to another, forklifts and automated guided vehicles move around, inventory is forming and deforming during the day.
This information comes in large amounts, and, possibly, from many sources and technologies.
For instance, an organization may have:
RFID data
BLE or UWB location data
Machine sensor data
ERP data
MES data
WMS data
EAM data
It is not enough to just collect this information: the system has to make it usable.
From raw events to information
Let us imagine a simple material tracking scenario.
An RFID reader sees an item entering a certain zone.
The event that is sent to the IoT layer may look something like that:
Item detected
Location: Zone A
Timestamp: 10:32:15
What this event means, however, is something that the application needs to determine.
It may want to enrich the event with additional information:
Item location
+
Production order
+
Expected process stage
+
Historical movement
+
Inventory status
The information can then be passed to AI or analytics to identify patterns.
This is when AIoT becomes more interesting than a simple tracking system.
Edge computing reduces distance between data and actions
Industrial facilities can generate a large amount of data.
It may not always be feasible or efficient to send every single event to a remote environment for processing.
Edge computing can help by performing some operations closer to the source.
One example of such a scenario can be represented by the following architecture:
Sensor
↓
Edge Device
↓
Local Processing
↓
Relevant Event
↓
Cloud / Central Platform
Instead of sending every raw event for processing, an edge layer can perform some filtering and only send relevant information for further processing.
The specifics will, of course, depend on the requirements and the environment.
Latency, connectivity, throughput, and security are just a few factors that have to be taken into account when designing such a solution.
AI adds context to IoT
IoT can inform about occurrences.
AI, however, can help determine if and what value these occurrences have.
Some examples may include:
Detecting unusual movement patterns
Finding anomalies
Supporting demand or replenishment analysis
Recognizing recurring delays
Evaluating utilization
Identifying potential bottlenecks
It is important to note, however, that AI should not be mistaken for the underlying data architecture.
Poor quality of sensor data, for instance, will inevitably lead to poor quality of analysis.
One of the interesting integration challenges
One of the interesting questions in industrial AIoT is data integration.
A manufacturing environment rarely uses a single application, but rather a set of interconnected solutions.
IoT Devices
↓
Data / Integration Layer
↓
ERP ↔ MES ↔ WMS ↔ EAM
↓
Analytics / AI
↓
Applications
The goal, in this case, is not to unify everything into a single solution, but rather create a connected layer that will allow using information from various systems along with the data from IoT devices.
This is particularly crucial for in-plant logistics, where physical movements and digital processes are deeply interconnected.
Users need a good presentation of information
Another important note concerns the presentation of information.
AI can identify patterns, but it is people who have to understand them.
A good industrial application, therefore, should be able to provide context and explain what is happening in terms that make sense to an operator.
Instead of hundreds of raw events, a user may want to see something along the lines of:
Potential exception detected
Material: Component-482
Expected zone: Production Line 3
Current zone: Staging Area
Status: Delayed movement
The more information that is understandable to a person, the more value the technology brings to the process.
Security and reliability
AIoT technologies connect software to the physical world, which means that the principles of security and reliability should be designed into the solution from the start.
Some of the areas that need to be considered include:
Device authentication
Data integrity
Access control
Network security
Failure handling
Data validation
System monitoring
Backup and recovery
Industrial systems, in particular, should have well-defined behavior in case of connectivity issues or device failures.
AIoT and in-plant logistics
In-plant logistics is one area where many of the above points come together.
Tracking materials, inventory, WIP, vehicles, and other assets can generate a substantial amount of data.
The challenge, however, is to turn this information into actionable insight.
PlantLog AI focuses specifically on applying AIoT technologies to in-plant logistics, including material tracking, asset visibility, inventory management, WIP, and other areas. Developers and engineers interested in the application can take a closer look at the platform here: PlantLog AI.
What developers should focus on
Creating an AIoT system is not a matter of simply adding an AI layer to an existing IoT solution.
The challenge is, rather, the entire pipeline that leads to patterns:
Physical event → reliable data → connectivity → processing → context → AI/analytics → application
If any of these stages is not handled properly, the entire system will fail to deliver.
As a result, developers working on industrial applications have an opportunity to think beyond AI and consider such factors as data architecture, edge processing, integration, security, physical processes, and users.
AIoT becomes valuable when all of these elements combine to create an environment in which connected activity transforms into information that can then improve operations.
Top comments (0)