Automated weld bevel detection and characterization via ultrasonic imaging with novel data compression

Intelligent manufacturing relies on feature extraction for process optimisation and defect detection for precise results. In the context of welding, feature extraction is a critical component of intelligent automated welding systems, enabling real-time geometry recognition and adaptive process control. Furthermore, accurate detection of lack-of-sidewall fusion remains a challenge for intelligent welding applications.
This study explores the use of advanced phased array ultrasonic imaging techniques as a viable alternative for lack-of-sidewall fusion detection and bevel geometry measurement in robotic welding applications. While conventional vision-based techniques, such as laser (active) and machine vision-based (passive), are widely used, their sensitivity to surface conditions and reflections limits their applicability for multi-pass welding. This study investigates advanced ultrasonic imaging techniques as a robust alternative to defect detection and bevel measurement in robotic welding applications.
A phased array probe was integrated with a robotic system and deployed on a single sided V-groove (90° bevel angle) bevel in a pulse-echo shear wave inspection configuration using a robotic manipulator to assess joint geometry through various ultrasonic imaging methods, including linear scans, sector scans and the Total Focusing Method (TFM). A key challenge in ultrasonic-based inspection is the high data volume generated during advanced imaging which necessitates efficient compression strategies for integration into deposition workflows. To address this, a novel signal processing approach was introduced, using the integral of the upper envelope of the signal, to reduce data size while preserving critical weld information. Additionally, a bevel angle estimation method, based on the histogram of oriented gradients, is proposed for real-time measurement at the point of data acquisition. Beyond geometric tracking, this study evaluates the capability of advanced ultrasonic imaging for detecting weld defects in real-time during welding. To assess this, welding experiments were conducted with highly controlled, intentional disturbances to simulate sidewall lack of fusion while the components were inspected using the PAUT method. Ultrasonic data analysis provided insight into the relationship between disturbance size and ultrasonic amplitude, demonstrating the potential for enhanced detection of lack-of-sidewall fusion in automated welding environments.
Through experimental validation under varying inspection conditions, the proposed approach is assessed for its accuracy, efficiency, and potential integration during deposition. With optimised robotic control and a selected ultrasonic modality, the error in bevel length measurement is below 2%. Compression rates of up to 90% are achieved for a Full Matrix Capture dataset composed of 16e6 samples in approximately 100ms, with error below 8 percent, irrespective of robotic parameters.
Furthermore, the bevel orientation is extracted from a noisy dataset with an error below 1%. Results indicate that the method effectively enhances weld monitoring capabilities, offering insights into quality control. These findings underscore the potential of ultrasonic-based sensing for enhancing bevel monitoring, improving defect detection, and laying the foundation for automated robotic welding control in intelligent manufacturing, through detection of changes in the bevel length. By integrating ultrasonic imaging into automated welding workflows, this approach enhances weld monitoring and defect detection capabilities, strengthening the role of NDT in intelligent manufacturing environments.