Ultrasonic phased array measurement & compression for in-process weld bevel estimation.

Area of Science:

• Robotics and Automation
• Non-Destructive Testing
• Materials Science

Background:

• Industrial robotic welding faces challenges in real-time weld geometry estimation due to environmental limitations and high data volumes from ultrasonic testing.
• Phased array ultrasonics offers volumetric imaging but is hindered by data scale and latency in robotic applications.

Purpose of the Study:

• To develop a real-time ultrasonic inspection pipeline for accurate weld bevel orientation measurement in robotic welding.
• To address barriers of reliable geometry estimation and efficient data management in high-throughput ultrasonic data.

Main Methods:

• Implemented a GPU-accelerated orientation estimator and a lightweight AUC compression strategy for ultrasonic data.
• Validated the pipeline on conventional V-groove and narrow-gap geometries using a robotically deployed phased array platform.
• Integrated Full Matrix Capture (FMC) processing with data sparsification for real-time inference.

Main Results:

• Achieved ±0.4° angular precision for conventional V-grooves and geometric errors below 3 mm.
• Demonstrated sub-250 ms execution for V-groove scanning and sub-1500 ms for narrow-gap configurations.
• Maintained over 88% data reduction while ensuring geometric accuracy.

Conclusions:

• The developed pipeline provides a modular, interpretable solution for real-time weld geometry inference and data reduction in robotic welding.
• This proof-of-concept demonstrates a viable pathway for embedded, closed-loop ultrasonic inspection systems in automated welding.
• The system successfully simulates realistic industrial scenarios by dynamically scanning stepped bevel geometries.