From Scrap Material to Error Correction: Performance Revolution in Manufacturing with Edge AI

[Ahmet Ö.]

Globalization, labor shortages, supply chain issues, and sustainability pressures create a complex picture for manufacturers. Under these conditions, manufacturers must improve quality, reduce scrap and emissions, and maintain efficiency, but they often lack the necessary personnel and expertise to achieve these goals.

Digital solutions come into play here. Artificial intelligence (AI) and machine learning (ML) technologies offer significant advantages in meeting the requirements for operational excellence. However, most visible AI applications rely on cloud technology due to high processing power requirements and scalability needs. In industrial environments, latency, poor connectivity, security risks, and costs make cloud usage impractical. Cloud infrastructure cannot provide the necessary performance for production lines requiring real-time feedback.

### Differences Between Cloud AI and Edge AI in Industry

• Cloud AI applications are not suitable for many industrial applications due to latency, connectivity issues, and security.
• Edge AI minimizes latency by processing data locally and eliminates dependence on cloud connectivity.

Leveraging AI and ML in manufacturing requires moving with the right local platform and partner rather than having expert personnel. Therefore, local AI solutions on industrial computers (IPCs) are gaining prominence for real-time and autonomous AI systems in factories.

### On-Site AI Usage Has Become Essential
Real-time industrial AI cycles capture data, analyze it, make decisions, and transmit the results to controllers, enabling immediate intervention. For example, for a camera capturing 30-100 images per second on a production line, milliseconds are crucial. Delays in cloud connectivity can lead to scrap, production downtime, or safety risks.

Site selection also exacerbates connectivity issues; in remote locations like mining and oil, cloud access is limited, and connections can be unreliable. If critical AI tools on the production line are connected to the cloud, production stops when the connection is lost.

Security and cost are other factors limiting cloud AI usage. Sending production data to the cloud is unacceptable for many companies, and data transfer requiring high bandwidth incurs significant costs.

### On-Site Processing Advantages

• Latency is reduced as data does not leave the facility.
• Operational costs decrease.
• Data privacy and security requirements are met.

Consequently, many companies are bypassing cloud AI and opting for industrial AI solutions on purpose-built IPCs with integrated AI accelerators.

### The Right Processors for Industrial AI
Processors designed for Edge AI are optimized for industry. While traditional GPUs perform well in training, edge AI processors provide high throughput in real-time inference. The SiMa.ai MLSoC platform, used in Emerson's next-generation industrial PCs, is an example of this processor generation.

### SiMa.ai MLSoC Features

• Provides real-time closed-loop control.
• Offers immediate variation detection and intervention in production.
• Operates without the need for cloud connectivity.
• Suitable for high image processing speeds and low power consumption.

### Suitable Platform for Industrial AI
A suitable platform is essential for industrial AI. Emerson's PACSystems generation IPCs are durable for harsh conditions, capable of fanless cooling, long-lasting, and offer stable performance. These devices are resistant to factors like dust, vibration, and humidity in the production environment.

### PACSystems IPC Features

• Designed for continuous operation.
• Multi-core x86 processor with integrated AI accelerator.
• Reliable, predictable performance.
• Low heat and power consumption.
• Long hardware lifespan.

### Importance of the Right Processor and Platform
This hardware combination enhances efficiency by correcting production processes before errors occur with real-time closed-loop intelligence. The AI processor performs high-performance inference, while the IPC manages data pre-processing, protocol compliance, and visualization.

### Field Application: Pipe Wrapping Inspection System
A thermoplastic pipe manufacturer installed an industrial AI visual inspection system to monitor the tape wrapping of pipes. Errors in tape placement weakened the structure, while human inspection was slow and inconsistent. The new system controls wrapping with real-time AI analysis and prevents errors, reducing the scrap rate to almost zero.

This solution can inspect pipes of different diameters and adds minimal space to the production line. Additionally, AI records provide video footage for quality and compliance documentation.

### Widespread Use of AI Applications Across Sectors

• Laser welding alignment and quality control
• Automotive adhesive applications
• Life sciences packaging and infusion bag inspection
• Oil and gas combustion and leak detection
• Mining ore analysis
• Recycling belts and separation optimization

Furthermore, applications such as AI motion control and predictive maintenance are also developing.

In conclusion, on-site AI solutions improve quality in production, reduce costs, and ensure business continuity.