Adaptive Vision Turns Robotic Welding Variability into Productivity

TAS (https://tas.com), Houston, Tex., a modular design, engineering, and manufacturing company specializing in the fabrication of structural skids for modular data center and energy units, had been facing upstream variability. This limited the efficiency of fixed-program automation. Find out how AI-powered, machine-vision control provided a solution. 

Behind the Scenes 

The urgency behind TAS’s implementation of adaptive welding intelligence stemmed from both the need for welders and the demand for infrastructure. The American Welding Society (https://weldingworkforcedata.com/) estimates the United States will need roughly 320,500 new welding professionals by 2029. At the same time, demand for structural skids is climbing. Fabricators serving this market are challenged to hire enough welders to meet the required volume without losing out to the competition. Automation offers a path forward, but TAS faced a problem common in structural fabrication: inconsistent fitup conditions prevented conventional robotic systems from delivering repeatable results. 

The Fabrication Challenge  

TAS builds a structural framework for the data center market, serving as the backbone of every modular block. Box tubing construction and unistrut form an 8 x 24 ft frame with over 300 welds. These welds are 2F fillets, 1G flare-bevel welds, and a reinforced 2G flare-bevel weld on the unistruts. Multiple manual fitup stations and bulk tube cutting on large bandsaws introduced significant variation in joint tolerances. With a 1/32-in. weld flatness requirement, implementing a fixed parameter set produces blow through or excessive reinforcement, requiring hours of postprocessing. As a result, no two assemblies arrived at the robotic cell with the same geometry. Those fit-up variations exposed a fundamental limitation of traditional robotic welding systems. 

Fixed Robotic Programs

A preprogrammed robot path assumes the joint sits where the CAD model says it does. When fit-up drifts, fixed travel speed, fixed wire feed, and voltage schedule produce burn-through in wide gaps, incomplete fusion in tight ones, and inconsistent profiles in between. 

Moving welding to Yaskawa Motoman (https://www.motoman.com/en-us) arc-welding cells cut the nominal total processing time from about 16 hours (1 hour welding/15 hours postprocessing) to 9 hours (2 hours on the robot/7 hours postprocessing). 

Closing the Loop with Adaptive Vision 

To address the problem, TAS added a real-time, machine-vision layer on top of the existing robots, converting an open-loop path follower into a closed-loop adaptive welding unit. The vision system images the joint immediately ahead of and within the weld pool, tracks the joint, and adjusts welding parameters on the fly — travel speed, wire feed, voltage and current, weave width and dwell, and contact-tip-to-work distance — to match the gap and geometry it observes rather than the geometry that was programmed. 

Path generation is handled in NovPlan, Novarc Technologies’ (www.novarctech.com) offline robotic programming software, before the cell runs. Deployment proceeds in phases, with each phase requiring a site acceptance test and a structured validation run.  

On the two trial cells, adding adaptive control increased arc-on time by about 10 minutes per frame and reduced postprocessing by approximately three hours.  

During phase-two validation, twin robots welding at the demanding −180- and 0-deg positions produced zero burn-throughs on the autonomy-controlled welds, with the remaining engineering effort focused on tuning weld height into tolerance.  

The team initially validated the system using a single frame, followed by a ten-frame run. 

Measured outcomes on the trial cells were as follows: 

• Metric Robot only Robot and adaptive vision  

• Arc-on weld time per frame ~ 1.5 hours +~ 45 minutes 

• Rework per frame ~ 11 hours ~ 3 hours 

• Burn-throughs (phase two validation) — 0 

 Two practical lessons are worth noting for fabricators considering the same path. First, adaptive control reduced rework significantly but did not eliminate it; closing the loop manages variability, yet it does not erase the upstream causes. Second, the cell still depended on housekeeping: Frames had to be cleaned of saw shavings before entering the robot area so that spatter and an obscured joint line did not defeat the vision system. 

Sensing-based welding raises the value of disciplined part preparation. 

Implications for High-Mix Structural Work 

Robots reliably compress arc-on time, but in fabrication shops fed by variable upstream processes, the limiting factor is the repair loop, not the weld itself. Adaptive, vision-guided parameter-control attacks that loop directly, letting the system respond to the part in front of it rather than to an idealized model. 

TAS’s experience suggests that adaptive welding can help fabricators capture the productivity benefits of robotics even when upstream fitup conditions remain inconsistent. As labor shortages persist and project demands increase, the ability to respond to real-world part variation may become as important as welding speed itself. 

The objective is not to replace welders. It’s to enable a smaller, skilled team to produce sound, code-compliant welds across a changing product mix without incurring rework costs. As more fabricators adopt sensing-based welding, the front-end disciplines — cut quality, tack consistency, and cleanliness — will increasingly determine how much of that promise is realized. 

Novarc’s solution brought AI-powered, real-time vision, and adaptive control to the Yaskawa robots implemented at TAS to build structural frames for the data center market, automatically adjusting weld parameters for gaps, misalignments, and tacks. Although still in its early days, the autonomy solution is set to increase ROI on the fabrication floor at TAS by reducing manual rework and compensating for part variations upfront. 

This article was written by Soroush Karimzadeh, CEO, Novarc Technologies, Vancouver, British Columbia, Canada, for the American Welding Society.