Building a Layer for AIoT and Automotive Manufacturing

Modern auto factories are distributed software systems.

PLCs control the machinery, MES manages the production processes, RFID/RTLS keeps track of the location of things, AGV transfers materials within the facility, SCADA monitors the equipment, and ERP stores the business-relevant information on these processes.

How does one tie all these systems together in a way that reliably brings the information these systems need to perform their function, without turning it into an unmanageable set of point-to-point integrations?

Industrial AIoT integration layer presents us with an alternative.

Architecture

One can imagine the architecture which utilizes an integration layer as follows:

Machines / PLCs / Sensors
|
v
Edge / IoT Gateway
|
OPC UA / MQTT
|
v
Integration Layer
/     \
/      \
MES      RLS
|       |
+------ + -----+
|
v
Analytics / ERP

The precise way it would be implemented will differ, but there is a common idea of collecting, normalizing, correlating and distributing events.
We are not seeking to build an alternative to the systems present in the factory, but a way for these systems to share the information needed for their operation.

OPC UA as standard

Factory equipment comes from various sources.
Modern robotic cells may expose their data in a highly-structured manner while legacy PLCs and machines reveal little information about their internal state.
OPC UA can help address this concern by providing a common language for exposing industrial information to other systems.
Rather than point-to-point integrations, an organization can pick up an industry-standard set of information exchange interfaces and employ them across its operations wherever available.
This is especially helpful as the factory grows and acquires more equipment.

Role of MQTT

MQTT is helpful in cases when the architecture needs lightweight publish/subscribe messaging.
For example, it could be employed in an edge gateway which would have subscribers which listen for certain types of information.
An example of the data could be the following:

{
"machine": "assembly-cell-07",
"event": "cycle_completed",
"timestamp": "2026-09-23T15:20:31Z"
}

This way, various applications can listen for different messages instead of the machine having to talk to every interested party.

Adding manufacturing context

Raw machine events often aren't enough.
To understand what occurred, a production application may need to know what car it was, which production order it belonged to, which workstation it was at, what component was being installed, what happened before, and if the needed piece of hardware was available when it needed to be.
This is where integration with the MES becomes essential.
Rather than a machine event being a simple occurrence, it can represent several related manufacturing and logistics events:

Machine Event
|
v
Workstation
|
v
Production Order
|
v
VIN
|
v
Component / Quality Event

Such approach greatly enhances the value and utility of any event stream.

RTLS and location data

We also sometimes need to know where items are in our facilities.
A RTLS platform can provide details about vehicles, tools, or materials.
Assume, for example, we get an event that component X needs to be at workstation A.
Now imagine correlating that with other information:

1. MES - the factory needs component X
2. Inventory - the factory has component X
3. RTLS - the location of component X
4. AGV - transport vehicle availability
5. Production - workstation readiness Rather than five separate pieces of information, we can combine them into a single manufacturing context. ## Edge processing and bandwidth considerations In some cases, having the data processed closer to the source is beneficial. The decision to use an edge gateway can be helpful when we need to identify certain patterns of interest or take specific actions. Such cases could include detecting production events, coordinating an AGV, making decisions based on local sensor observations, safety-related events, or conditions which require data samples at a higher rate than is practical to transmit. An edge gateway which performs some processing can help filter the data and only send what other systems actually need, saving both precious bandwidth and processing power.

Challenges of the point-to-point paradigm

At some point, it becomes impossible to manage additional point-to-point connections.
Adding more devices and systems makes the architecture unstable:

PLC -> MES
PLC -> ERP
PLC -> Analytics
PLC -> RTLS
MES -> ERP
MES -> Analytics
RTLS -> ERP
RTLS -> Analytics

As opposed to that, a more structured approach would give us shared events and functionality:

+--> MES
|
Machines -> Edge -> Integration Layer -> ERP
|
+--> RTLS
|
+--> Analytics

Of course nothing in the world of software is ever as simple as it appears.
A different structure can provide us with more flexibility and allow us to evolve the architecture as the operations on-premises evolve.

What developers should think about

Developers of industrial AIoT systems should not think about APIs first and foremost.
Among the problems which they would need to address are:

• Data models and how information on the same asset or event may differ in different systems
• Time and the need for timestamps to be trustworthy and consistent across all systems to enable time-based correlation
• Identity and the connection between objects, for example, the relationship between a VIN and a production order or a component
• Connectivity and the fact that the internet is not available everywhere and the factory cannot rely on it being present
• Securing the environment both for operations and for protecting against threats
• Observability and the ability to detect and diagnose issues when they occur
• Scalability and the ability to add more production lines and facilities while keeping the same level of quality and with minimal infrastructure investment We can then proceed to examine an example system. ## Practical example OEMNex AI presents an approach based on integration with MES and ERP as well as PLCs, SCADA, RTLS, RFID, Edge, OPC UA and MQTT. Let us analyze how an integration layer could be applied in the scenario. First off, the layer itself serves as an intermediary between the production environment and other systems. Events flow through it, correlating data from different systems such as MES or RFID/RTLS with machine events or sensors. Now, why is this architecture beneficial? Let us consider a few practical examples of production-relevant information which might want to be integrated with some form of event-driven architecture. "We need to know where a critical component is and if it will arrive where it needs to be when it needs to be there." This single use case would already require integration with the MES, inventory, RFID/RTLS and possibly the AGV. With the basic data model built and the information flow established, adding other production events would become a matter of identifying new sets of correlated information. OEMNex AI describes automotive AIoT integration for automotive manufacturing with the involvement of MES, ERP, PLCs, SCADA, RTLS, RFID, Edge solutions, OPC-UA and MQTT. What it really boils down to is, we are not seeking to gather all the information from all the systems. Rather, we want to provide exactly the data needed in such a manner that it is available where and when it is needed. All in all, industrial AIoT is much more interesting than adding a few sensors on the factory floor. The real challenge, rather, lies in building an equally reliable integration layer which ties machinery together with all the other relevant systems and processes.