Enabling Automatic Operator Compliance Monitoring with Vision AI

Background

Safety and transparency in food manufacturing are essential today as factories are required to operate under strict food safety regulations. This particular customer has been using a traditional CCTV system which only provides passive information. In this case, the factory had to invest a lot of resources into operator compliance in order to meet regulations, making the process more complex and jeopardizing productivity. Therefore, the facility manager was looking for a solution that can co-exist with their original system, while still meeting compliance and sending automatic notifications when violations are discovered.

Requirements & Challenges

A cake factory wanted to build a system that could provide real-time notifications when illegal violations occur. This kind of monitoring is vital in the food industry because substandard quality issues will eventually jeopardize a brand’s value. However, their factory could detect abnormalities during processing so they decided to upgrade their existing system by integrating an AI application. The CCTV system can now receive real-time information and it can detect all violations. Other manufactures in their supply chain can also be assured that compliance is being strictly adhered to and all food production is being properly regulated in a safe and transparent way. A suitable system should provide real-time information of all violations and be saved to a video log in the server at the back end.

Challenges

• Unexpected abnormal events and behaviors.
• Unqualified products and high failure rates.
• Limited remedial action without real-time notifications.
• Excessive time and money gets spent maintaining violation records.
  

Advantech Offerings

After deploying a trained model to the IFS-RTM-730AXL5A edge AI system, the system will connect to the existing CCTV via the RTSP protocol and automatically detect all violations. The IFS-RTM-SKY72WX suite and the server will pre-install the I.App event center to send notifications to make records of any abnormal events that occur on the shop floor. 

Benefits

• Improved quality and production performance with AI inspection.
• Real-time monitoring and notifications.
• Saved violation video log.
• Optimized operation workflow and production process.
  

Project Implementation

• IFS-RTM-SKY72WX: High computing server, SKY-7221, with Xeon® Silver 4210 10C x 1/ 32G*2 / 256G*4 / 4TB*5 / RTM / 10,000tags/5s to be a event center in the control room.
• IFS-RTM-730AXL5A: Edge AI inference system, MIC-730AI, based on NVIDIA® Jetson AGX Xavier™ with iFactory/ activity recognition software.
• EKI-7710E-2CPI-AE: Industrial PoE switches provide power for IEEE802.3af/at-compliant devices via Ethernet cable, eliminating the need for additional wiring.

System Diargram

Conclusion

The factory has been able to save 40% of the time previously spent checking operator compliance. In addition, the system also helped to optimize processes and delivers real-time violation notifications to facility managers. The first step is to collect 5-10 minutes of video which is used to train the system according to the customer needs, then deploy the AI model to the IFS-RTM-730AXL5A edge AI system. After that, the edge system delivers the video stream from all the cameras on the shop floor via RTSP protocol. The edge AI system performs inspection functions and labels all violations captured by cameras. In addition, by deploying the IFS-RTM-SKY72WX suite, a server with the data center I.App installed can record all details of violations so now the whole system in the factory will go from passive to active management of violations.

The winning strategy was to provide an AI solution in a short amount of time. We offered a POC application in 4 weeks for the customer to evaluate to see if it was a worthwhile return on their investment. We then deployed the solution within 3 weeks and made sure each customer benefitted from a low total cost-of-ownership. The system was flexible enough so that different models can be built for different scenarios. For example, a hand-washing station, gadget checking station, and a packaging station could each have their own models. An IT manager can also upgrade the models to achieve better accuracy rates for different scenarios using the low-code/no-code platform.

By actively managing violations of regulations, about 40% of time usually spent on monitoring operator compliance and quality assurance was saved while maintaining the same high quality standards.

Why Advantech:

• Low total cost of ownership.
• Provides a POC for ROI evaluation.
• Advanced design strategy to allow space for upgrades and additions.
• Fast development time.
• Clear system architecture.
• Co-existence with existing conventional CCTV systems.

Explore More