Automotive manufacturing plants produce an incredible amount of data.
PLCs report machine status, SCADA captures process information, MES captures production data, ERP tracks business processes, RTLS tracks assets, RFID and UWB provide location information, and quality systems capture inspection data...
The challenge is not only the collection of this data
but
integrating these sources and providing enough context for it to actually be useful information
This is where an AIoT data architecture would come in handy.
What is AIoT? What would an example architecture look like in a factory?
AIoT is an extension of Industrial IoT that applies AI and analytics capabilities to this data.
In a simplified form, it could look something like this:
Factory Equipment
|
v
PLCs / Sensors / SCADA
|
v
Edge & Industrial Connectivity
|
v
Data / Event Integration Layer
|
+-------- MES
+-------- ERP
+-------- RTLS
+-------- RFID / UWB
+-------- Quality Systems
|
v
Analytics & AI
|
v
Operational Applications
The architecture would not necessarily require replacing existing systems (MES, ERP, etc.), but rather focus on how to connect disparate systems together reliably.
The integration layer is critical.
An automotive plant is likely to house equipment from multiple vendors, spanning generations of technology.
Some may communicate over industrial protocols, others might expose APIs, databases, or messaging systems, etc.
Things like OPC UA, MQTT, Modbus, industrial Ethernet, APIs, and event-streaming can all be leveraged as part of the integration architecture.
The important thing is that there's an understanding of what data each system exposes, how often it updates, where it should be processed, normalized, how downstream applications will consume it, and how to handle disruptions in connectivity.
It's particularly important for production environments where disruptions can cause operational issues.
Not all manufacturing events need to go to the cloud.
Depending on the required processing power, latency, reliability, and bandwidth needs, systems in a manufacturing plant could be designed to handle events at the Edge or in a centralized cloud environment.
Edge processing can be useful for applications that have low-latency requirements.
Other drivers could be a need for local decision making, network reliability, reduced bandwidth costs, data sovereignty, or operating in disconnected / harsh environments.
On the other side, centralized cloud environments can offer more processing power for complex analytics, AI, and modeling, as well as better organization-wide visibility and reporting, and easier model management and deployment.
A hybrid approach could look like this:
Cloud
|
Analytics / Models
|
Factory Gateway
|
+-------------+-------------+
| | |
MES SCADA RTLS
| | |
PLCs Sensors Trackers
It's a matter of deciding what fits best for each application based on their requirements around latency, security, connectivity, and others.
What value does contextualizing industrial data provide?
Industrial equipment tends to report fairly basic information.
Temperature:
temperature = 82C
...but that becomes much more valuable once you add context:
Machine: Assembly Station 12
Vehicle: VIN XXXXX
Production Cycle: 4382
Process: Battery Installation
Timestamp: 10:42:17
Quality Status: Passed
Contextualizing data is one of the most important activities in manufacturing analytics.
For most practical purposes, it's not as useful to look at isolated values, but rather analyze events and relationships.
As a general rule of thumb, the more valuable data is, the more it has been massaged and enhanced with contextual information.
What other data sources can add context?
RTLS systems can provide location tracking of vehicles, AGVs, tools, racks, containers, materials, workers, equipment, etc. Again, that's valuable data, but not always easy to consume in its raw, coordinate-based form.
An analyst might need to ask questions such as
How long was this vehicle in the station?
How often was this AGV waiting at this location?
Is material arriving at this line consistently on time?
Are workers spending too much time picking parts?
That's where the value of an AIoT architecture comes in - to provide these insights in an operational, easily consumable format.
Top comments (0)