Gogoro, a global technology leader known for its electric Smartscooters and battery-swapping ecosystem, is using AI vision to optimise the production process.
The company has implemented PowerArena’s Human Operation Platform (HOP), an advanced smart manufacturing solution driven by AI, and Advantech’s SKY-6400 GPU server for AI deep learning inference and training, as well as real-time data analysis.
The background
Traditional production line management relies on industrial engineers stationed next to workstations, making visual observations, and manually recording work hours with stopwatches. They later analyse the data, identify influencing factors, and propose hypotheses to enhance line production. However, this approach is complex, time-consuming, and incomplete. The data obtained may contain inaccuracies due to subjective errors in judgment. Implementing improvements lacks timeliness, and making immediate adjustments can be difficult.
AI vision fills the information gap in production operations. It replaces manual monitoring with objective, 24/7 and complete cycle time collection through real-time video analysis. This eliminates the possibility of human operator measurement errors.
Over the years, Gogoro has continuously implemented cutting-edge technologies such as MES, IoT, and more recently AGVs to optimise its industry-leading EV production line. The result is a highly flexible operating station with an excellent balance between human and machine involvement. This setup heavily relies on human operators who play a crucial role in both the output volume and quality control of the assembly line.
Inspection issues
However, a specific case had been causing resource drain for Gogoro in managing and operating their production. A lack of supervision had led to some of the procedures being susceptible to inspection issues. To address this problem and close the data gap related to “human movements” and “unconnected tools”, Gogoro actively sought solutions to identify the root causes of inefficiency and quality issues in its operations. PowerArena’s AI Vision platform, the Human Operation Platform (HOP), offered the necessary tools and technology to pinpoint underlying issues and provide valuable insights into production processes.
This comprehensive approach ensures the highest level of quality control for every single vehicle throughout the production process.
“AI vision is the last piece of the smart manufacturing puzzle,” says Charlie Huang, Gogoro’s manager of smart factory platform and solutions. “… PowerArena’s AI vision allows the concept [IPQC] to be brought to life.”
How the HOP platform overcomes key challenges
In-process quality control for layer-by-layer assembly
At the parallel stations dedicated to the front wheel bracket, two operators work together on opposite sides of the electric scooter, using a wide variety of tools. A critical step in the process involves applying oil before closing the enclosure. However, once the enclosure is sealed, the areas where the oil is applied become hidden, making it virtually impossible to verify whether the oil application has been completed.
Thanks to HOP’s motion analysis, the process of oil brushing can now be verified in real time. AI vision is used to safeguard critical worksteps. By combining this with video footage, management gains the ability to trace back to any specific point in the production process.
Photo courtesy of Advantech
Production line footage tracking and tracing
Another unexpected abnormality stems from workers’ attempts to speed up production by taking shortcuts. Gogoro’s internal management system captured a cycle time that differed by a few seconds compared with the data detected by AI vision. Upon analysing the footage recorded by AI vision, it became apparent that the standard operating procedure (SOP was not followed correctly. The system’s completion button was supposed to be pressed after fastening two screws, but the worker, for ease of movement, pressed the button after the first screw was fastened, before moving on to the second one. This discrepancy led to the numbers on the MES system differing from the accurate numbers captured by AI vision.
Before HOP’s AI vision, it was impossible to detect the error by looking at the numbers and finished products alone because the process was complete, just not in the correct order. The only way to make this small discrepancy visible was to trace back the AI footage, which enhanced the reliability of captured data.
Zero defect rate
Unlike electronic products, faulty vehicles can remain undetected until much later. Noises or operating irregularities may only start to appear after they have been on the road for a considerable period of time. Typically, the issues are identified during inspections, which often require disassembly to uncover the problem. Beyond the risk to user safety, manufacturers also have to deal with the substantial costs of remediation and negative effects on their reputation.
In the automotive assembly industry, achieving high quality and precision while eliminating all defects in the manufacturing process is crucial. When AI vision detects an abnormality in Gogoro’s production line, the AGV carrying the electric scooter will promptly come to a stop and notify the manager. The work in process will remain at the station until the manager confirms that all the identified mistakes have been rectified. Implementing these in-station corrections incurs minimal costs compared to the potentially significant expenses associated with remedial actions in the future.
After HOP was introduced into Gogoro’s production line, product yield successfully reached 100%. HOP’s AI vision prevented all defective products from leaving the station.
Capture 24/7 production data
HOP replaces the conventional manual monitoring approach with an objective and precise system that continuously captures cycle time through real-time video analysis, offering a reliable record of operations through AI vision. It excels at identifying unusual cycles and equipping managers with comprehensive data to optimise the performance of production lines.
Video track and trace, production visualisation
With HOP, managers can examine the state of any workstation at any point of production to uncover hidden problems and locate the root causes of inefficiencies. This allows managers to concentrate on important matters and tackle the issues that have the greatest impact. HOP also supports data analysis for situations like machine failures, material shortages, or employee changes.
Real-time alert digital Poka-Yoke
HOP sends out alerts immediately when abnormalities occur. Managers will be notified whether the problems stem from equipment malfunction or operators deviating from specific SOPs. At key workstations, the procedures can be checked in real time, ensuring seamless integration with MES to swiftly identify deviations. These proactive measures prioritise personnel safety and product quality, quickly resolving issues and significantly reducing the costs associated with investigations, speculations, rework and waste disposal.
Implementing Advantech’s SKY-6400 GPU server
The Gogoro production line transmits approximately 15 MB of image data to the server every minute. This results in a daily computational data volume of nearly 50 GB and a storage requirement of about 150 GB. In total, up to 1.5 TB of data is processed each month. Due to the need for continuous image data, it is essential to have highly stable and high-performance hardware equipment.
To meet these demands, Gogoro utilises the Advantech SKY-6400 GPU server for AI deep learning inference and training, as well as real-time data analysis. Advantech servers are employed for on-site local computing at the factory, ensuring the utmost stability and efficiency in safeguarding customer data.
The Advantech SKY-6 series GPU servers offer a wide range of flexible and efficient solutions, available in 1U to 4U configurations. These servers are equipped with dual Intel® Xeon® Scalable processors and can accommodate up to 10 PCI-express expansion slots, providing excellent flexibility for integrating GPU cards required for AI training and inference. The SKY-6 series boasts an industrial-grade design, making it suitable for various harsh environments, ensuring the most stable performance in AI smart factory applications.
It features:
- 4U Rackmount Intel® Xeon® Scalable GPU Server
- Processor: Dual 2nd Gen Intel® Xeon® scalable processors
- Memory: 12 x DDR4 2666/2933 MHz ECC RDIMM/LRDIMM / Intel Optane DC Persistent Memory, up to 2TB
- Remote Management: IPMI function support
- Expansion: Supporting 4 * PCIe x16 double-deck card + 1* PCIe x8 single-deck FH/HL card + 1*PCIe x4 card
AI vision aims to simplify managers’ work, enabling them to unleash their full potential. It serves as a game-changer when it comes to upgrading and effectively managing various aspects of manufacturing operations. Tasks that once required a significant amount of time from industrial engineers, such as measurement operations, can now be streamlined effortlessly with through AI vision implementation. AI vision has the capability to uncover hidden problems that are otherwise difficult to monitor. This is especially beneficial for assembly stations, which are known for their quality control challenges. By leveraging AI vision, managing the intricate variability of human factors becomes an efficient and effective practice.
ESIS is the leading supplier of outstanding industrial electronic equipment in the Australian market since 1971. Check out our exclusive range of Fanless PCs, Industrial PCs (IPC Solution), Rugged Notebooks, HMIs, Panel PCs and other electronic components. Our products are certified and made with high-quality custom materials for accurate results. Contact us for a detailed product catalogue or to share your queries.
Article courtesy of Advantech.
