The team considered and evaluated a number of solution proposals, including soap-bubble dunk tank, robotic helium sniffers, and sound-based technologies, but none of these met the requirements for leak point determination, Carlson says.
When computed tomography scans of leaking welds with surface cracks did not conclusively reveal a leak path, the team had to figure out how to investigate the issue in a lab setting. Because current helium detection methodology could not pinpoint a leak location, the team brainstormed whether the gas leak path could be imaged, Carlson says.
“The technical discovery was that weld cracks and defects are 3D and are not capable of being imaged in a high-volume production environment,” Carlson notes. “The breakthrough for us came from combining specialized equipment to image trace gas that could also meet the leak point size requirements with an algorithm capable of distinguishing gas plumes from the surrounding atmosphere.”
They decided on a CO2 gas imaging system for several reasons, he says.
“CO2 was selected because it met our core leak detection requirements for BVSeal,” Carlson said. “It is relatively low-cost, readily available, and enables imaging even for small leaks. Helium is another gas that is commonly used in gas leak detection; however its cost, combined with the fact that it cannot provide leak point detection, led us to CO2 as a better solution for BVSeal.”
Components of the BVSeal Automated Inspection System
The BVSeal system consists of the following:
- 2 Gas imaging camera (vendor undisclosed)
- 2 Fanuc CRX cobot
- 1 gas excitation unit (internally developed)
- 1 Advantec GPU based IoT edge computer
- AI software (internally developed)
- No specialized lighting, other than the overhead lighting already installed in the manufacturing facility, is needed, Carlson says.
How the System Works
The battery pack is positioned in a workstation on the production line by an autonomous mobile robot. On either side of the station are cobots equipped with gas imaging cameras as end-of-arm-tooling. A sensor is placed that detects and activates any trace gas leaking from the battery pack, while an AI-based model traces the gas movement to detect the leak point in real-time.
“Cobots have integrated safety systems, which allow for their deployment on the factory floor without fixed fencing as [is] required with conventional robots,” explains Carlson. “This allows for easier mobility of battery packs into and out of factory stations. Having a sensor on the end of the robotic arm allows for optimal positioning of the battery pack and ensuring that all surfaces are evaluated.”
The camera captures images of the gas plume flow. These images are transmitted to the computer loaded with AI software that analyzes the image data. Once the image data is analyzed, the algorithm pinpoints the location of the leak. Once the leak is located, human operator inspects the leak to determine whether it's reparable or if the battery pack must be scrapped.
The system, while installed on the line and managed via an edge IoT computer, stores data long-term via the cloud; that data can be used for reports and long-term analytics, Carlson says.
What's Next for BVSeal?
BVSeal is working, Carlson says. By finding defects in real time before the build is completed, the system is significantly reducing the number of rebuilds, tear-downs, and rejected products, saving time, labor, material, and money.
“The system improves productivity and quality by reducing battery tray and cover scrap, significantly lowering defect-related material losses,” Carlson says.
The goal was to come up with a scalable foundation that can be easily adapted to other applications rather than a single use solution. With that in mind, Carlson says the team continues to explore future applications for the system, especially those in which seal and joint integrity are critical, such as in engine and powertrain assembly.
