Our paper titled “WeldMon: A Cost-effective Ultrasonic Welding Machine Condition Monitoring System” has been published in the IEEE Ubiquitous Computing, Electronics & Mobile Communication Conference (UEMCON). The paper was co-authored by Beitong Tian, Ahmadreza Eslaminia, Kuan-Chieh Lu, Yaohui Wang, Prof. Klara Nahrstedt, and Prof. Chenhui Shao.
WeldMon: A Cost-effective Ultrasonic Welding Machine Condition Monitoring System
Ultrasonic welding machines are essential in lithium battery manufacturing, where ensuring high-quality welds is vital for operational safety and efficiency. However, existing tool condition monitoring systems are often costly, require machine downtime, and lack adaptability across different setups.
Our paper introduces WeldMon, a cost-effective ultrasonic welding machine condition monitoring system that runs on an edge device for real-time monitoring. WeldMon utilizes off-the-shelf sensors, a low-cost data acquisition system, and a neural network-based classification pipeline to accurately assess tool conditions without requiring expensive equipment or extensive machine downtime.
🔑 Key Highlights:
- Affordable solution: WeldMon costs less than $200, over 30 times cheaper than commercial systems (> $3600).
- High accuracy: Achieves an average cross-validation accuracy of 95.8%, outperforming existing methods.
- Robust performance: Improves tool condition classification accuracy by 8.3% through a novel data augmentation strategy.
- Fast processing: Processes each welding cycle within 385 milliseconds on an edge device.
- Real-world validation: Successfully deployed and compared with a commercial system on an actual ultrasonic welding machine.
Read the full paper here.
Read the news here.