Parametric modelling of saddle-shaped welds for autonomous manufacturing & inspection

The preparation, welding, and inspection of complex pipe-in-pipe saddle or nozzle welds present significant challenges in manufacturing, particularly for high-integrity industrial applications in the nuclear, aerospace, and naval sectors. Existing manual and semi-autonomous methods are labour-intensive and introduce uncertainty due to the complexity of the geometry. This often results in weld preparations and depositions that require substantial rework, as well as inspection strategies that may fail to detect critical defects. These challenges stem from the intricate saddle-shaped geometry formed at the intersection of cylindrical pipes, producing a volumetric weld interface with dynamically varying bevels and surface curvature.

This research addresses these limitations through an automated path-planning framework that integrates a comprehensive parametric model with an adaptive, sensor-driven robotic system. The mathematical model is underpinned by the sample geometry, its dimensions, and the required weld preparations, enabling the optimization of tool placement and deployment paths for fabrication, welding, and inspection applications.

The integrity of the weld is fundamentally dependent on the ability to inspect and validate the saddle joint upon completion. To enhance confidence in its capability, the parametric model has been extended within the domain of Non-Destructive Evaluation (NDE). A developed ray-tracing algorithm, built upon the parametric model, simulates phased-array ultrasonic wave interactions within the structure. This enables precise sound path mapping and supports an adaptive scanning strategy that dynamically adjusts to local curvature variations in real time.

Validation across a full-scale industrial sample demonstrates the effectiveness of this model- driven framework, enabling preparation, joining, and in-process force-torque-controlled dry- coupled ultrasonic inspection. By bridging the gap between computational modelling and robotic automation, this research advances intelligent manufacturing, offering an autonomous manufacturing solution.