Modern automotive manufacturing facilities are complex distributed systems. There are many sources of information, ranging from machine telemetry to business applications. The challenge is not necessarily with capturing that information and rather using it effectively in operational decision support
The intersection of Artificial Intelligence and the Internet of Things is an interesting space to explore in that respect, and this article tries to show what that could mean in an automotive manufacturing context.
A Look at Some Architecture
An AIoT architecture can be viewed as a stack, with layers addressing specific aspects:
Physical Assets
↓
Sensors / PLCs / RTLS
↓
Industrial Connectivity
↓
Edge / Data Pipeline
↓
Data Platform
↓
AI / Analytics
↓
Operational Insights
↓
Human or Automated Action
IoT and other industrial systems provide information from the physical world (or the digital representations of it). Connectivity makes that information accessible to various systems. Platforms store it or distribute it to interested parties, and AI/Analytics processes it or identifies actionable intelligence. The bottom line, in many ways, is still an operational decision that could be supported by information.
Information Available Within a Manufacturing Environment
There are different kinds of data and information that could be collected within the manufacturing space:
Equipment telemetry
Vehicle identification
Asset tracking/location
Production information
Inventory movement
Environmental conditions
Machine status
Materials flow
Workflow or personnel movement
Different aspects generate different information, and various technologies enable the capture of that data: sensors, RTLS, PLCs, SCADA, industrial networks, and so on. Not all of it is applicable at once, and it's more useful to think about particular problems and what information supports them.
Leveraging Existing Industrial Infrastructure
Systems and applications might provide relevant information, even though that data might not always be in a format that's easily consumable:
PLC / SCADA
│
├── Machine Data
│
MES ─┼── Production Data
│
ERP ─┼── Business / Inventory Data
│
RTLS ── Asset Location
│
IoT ─── Sensor Telemetry
From an architectural perspective, an AIoT implementation would be capable of providing a data lake, accessible to both operational and business intelligence applications. It may involve industrial connectivity technologies to access that information if the applications that need it are not collocated on the same platform. It could include everything from industrial messaging and data formats, APIs, edge computing, relational data, distributed data, and cloud data. All of it would present unique challenges addressed differently in terms of technology.
A lot depends on the environment and the specific set of data that needs to be accessed.
Edge Computing Makes Sense in Some Contexts
Cloud-centric architectures are not the only option in industrial Internet of Things implementations because some applications require low latency, local information processing, and even availability when the cloud connection is unavailable. Edge computing makes that information available in a timely manner and, in some cases, could support filtering and further processing.
A basic implementation could be:
Machine
↓
Edge Gateway
↓
Local Processing
↓
Filtered / Relevant Data
↓
Central Platform
↓
AI / Analytics
Specific use cases determine the architecture to use.
Another consideration is the volume of information. High-frequency or large data volumes that do not require further processing might make it more efficient to keep data at the edge instead of transmitting it to the cloud. If further processing is needed, it might be resource-intensive for a gateway to perform.
AI and Analytics Bring the Intelligence to the System
Once relevant and actionable information is available, the next step is to process it in a meaningful way. This is typically where tools like Artificial Intelligence and analytics come into play, with applications like:
Pattern recognition
Anomaly detection
Production analysis
Equipment analytics
Asset usage
Operational intelligence
Forecasting
Optimization and so on
This is again a matter of defining the requirements and selecting the information that supports solving the problem at hand. It then becomes a matter of identifying any relevant additional data and making sure all of it is available at the location where that processing will take place. It is one thing to say that an asset has moved from point A to point B; it is something else to combine that information with production data to identify that a particular vehicle production might have issues or that equipment is utilized in a non-optimal fashion.
Vehicle Traceability Use Cases in Manufacturing
A common use case for automotive manufacturing environments is vehicle traceability. During the different manufacturing steps and throughout the supply chain, different information could be captured regarding production, testing, logistics, and so on. Associating vehicle identifiers to relevant asset data helps in creating a production history and can be used in downstream applications like logistics, retail, and service. This area is covered by OEMNex AI as well, covering concepts spanning from the industrial IoT to vehicle production genealogy, RTLS, and manufacturing analytics.
Problem-First Approach Instead of Model-First
One of the errors often observed in industrial AIoT applications is that the approach is often model-first instead of problem-first. There is nothing wrong with experimenting and playing with different AI models, but an operational implementation typically has a specific set of problems that it needs to address. That perspective is also necessary when designing such systems and ensures that requirements stay in scope.
Examples of potential problems addressed by AIoT in automotive manufacturing:
Lack of visibility into production
Difficulty in asset location
Difficulty in associating production data
Missing information regarding inventory or material flow
Poor coordination of the production supply chain
Poor system integration and interoperability
Once the problems are identified, it becomes possible to define what information is needed and what additional constraints might exist.
The Greater Systems Engineering Challenge
Appointing AIoT systems in automotive manufacturing is, therefore, as much of a systems engineering problem as it is something else. There are various considerations regarding sensors, data availability and quality, data delivery and storage, operations, security, and so on. While an AI model could get significantly better results out of good data, it cannot help if that data is of poor quality, intermittent, or incorrectly labeled. Because of all these challenges, it is necessary to think about the data lifecycle as well as operational aspects, leading the effort much closer to something else that needs to be integrated into the systems engineering space.
Top comments (0)