AI-based real-time visual inspection is needed now more than ever

In recent years, the use of AI has progressed rapidly worldwide, with an increasing number of companies introducing AI for quality control and production efficiency.

However, the rate of AI adoption in Japan's manufacturing industry remains at just 21.4% (※), with many companies still in the consideration stage.

However, 83.1% of companies that have actually introduced AI responded that they "feel like it has solved problems" (※), making it clear that AI is effective in improving work sites.

*Source: MMD Research Institute "21.4% of manufacturing companies have introduced AI, 83.1% of people feel that introducing AI has solved problems, and 63.5% of people in the manufacturing industry have problems in their work."

In the domestic manufacturing industry, the number of personnel required to maintain high quality is decreasing year by year due to the retirement of skilled workers and a shortage of young talent.

This makes it difficult to replicate the eyes and intuition of experienced workers who have supported the site for many years, and maintaining quality itself is becoming an issue.

Additionally, manufacturing lines can contain invisible abnormalities such as defective molded products, slight misalignment of conveyors, and deformed components.

Even if each abnormality is inconspicuous, if it is overlooked it could lead to line stoppages, mass defects, and expensive recalls.

To address these issues, real-time visual inspection using AI is gaining attention.

In this article, we will introduce specific use cases along with actual demo videos to show how AI can contribute to resolving issues on the manufacturing floor.

Case 1: Real-time detection of minute fluctuations in a high-speed packaging line

On packaging lines that operate at high speed, even the slightest deviation can lead directly to equipment trouble and packaging defects.

In particular, visual changes such as tilted conveyor belts or slight bends in structural components are difficult for the human eye to notice, and delay in responding can lead to production stoppages and major losses.

In this demo, a computer vision system powered by SiMa.ai 's MLSoC™ constantly monitors the packaging line via cameras, detecting subtle visual anomalies in real time that could impact machine performance.

Here is a demo of anomaly detection on a packaging line (video in English).

From the captured video, the SoC performs the following operations:
 

1. Detecting objects and structures on the line

② The AI model estimates slight differences compared to the normal state

3. Visualize minute misalignments and distortions in real time

④ The judgment results are instantly processed within the SoC and any abnormalities are displayed as flags

It is now possible to detect minute variations that were previously overlooked at an early stage, providing strong support for stable operation of packaging lines.

By adding constant monitoring using AI in this way, the health of equipment can be maintained with greater precision, helping to reduce the risk of defective products being released or unexpected downtime.

Case 2: Real-time detection of surface abnormalities in plastic molded products

On a production line, even the slightest abnormality can lead to equipment failure, line stoppage, and even the release of defective products.

In particular, for plastic products, surface defects such as minute scratches, chips, and dents are directly linked to quality, so early detection is essential.

In this demo, AI running on SiMa.ai 's MLSoC™ is used to evaluate the surface condition of injection-molded plastic lids in real time.

Here is a video showing real-time visualization of surface anomalies.

From the image captured by the camera, the SoC performs the following steps:
 

①Object detection using the CenterNet model

②Anomaly detection using Student-Teacher Feature Pyramid Matching Network​

3) Visualize the difference from the learning model's expected value as a heat map

④ Real-time judgment and line response by the Arm Cortex microprocessor installed in the SoC​

Abnormal areas are displayed as a heat map, so you can see at a glance which areas have problems and to what extent.

Furthermore, all processing is completed instantly within the SoC, preventing defects from reaching the manufacturing process.

Summary: How AI will change the future of defect prevention and quality control

In this way, as introduced this time,
　

1) Example of detecting abnormal signs on the production line on the spot to prevent problems before they occur

② Example of "visualizing" the judgment of experts to reduce variations in quality inspections

Establishing a system that ensures AI operates reliably on-site will lead to improved productivity, stabilized quality, reduced manpower, and standardized know-how, and will be a powerful tool for the manufacturing industry in the future.
　

​The SiMa.ai​ ​MLSoC™ used in this demonstration is a next-generation chip optimized for realizing AI that can be used in the field.

Despite its compact size and energy-saving design, it achieves highly efficient inference of up to 50 TOPS and supports high-speed production lines with real-time processing at 120 FPS. The flexible development environment provided by the built-in Arm Cortex-A65 is also a major benefit of its introduction.

If you are considering introducing AI, you can start small like in this case study to verify the improvement effects on your own production line.

We hope you will find this article useful as your first step in utilizing AI.

Inquiry

Please feel free to contact us with any questions about our products, technical inquiries, sample requests, or estimates.