A fabricator with three welding robots spends 40% of each robot's shift on programming and setup — the robot is stopped, a technician in a welding helmet touches the torch to each weld start and end point, and the teach pendant records positions one by one. For a complex part with 80 weld seams at varied angles, teaching takes two full shifts — 16 hours of robot downtime per part number. When the part comes back in a month, the program is recalled from memory, and the robot welds it in 22 minutes. An offline programming (OLP) workstation in the engineering office — using a 3D CAD model of the part and a digital twin of the robot cell — generates the same 80-weld program in 4 hours while the robot continues welding the previous production batch. The program uploads, the robot runs a dry cycle to verify, and production starts. The OLP software license costs $15,000 to $50,000 per seat per year. The teach pendant costs nothing except the educated hands of a $35/hour welding technician and the robot sitting idle. This article compares OLP and teach pendant methods on programming time, robot utilization, weld path quality, and break-even production volume — so fabrication managers can calculate when the software pays for the idle robot hours it saves.
Teach Pendant Programming: Tangible, Accurate, and Consumes the Robot
Teach pendant programming is physical: the technician jogs the robot torch to each weld seam, adjusts torch angle and stick-out, records the position, and moves to the next seam. The result is verified by eye — the technician sees the torch contact the actual part, not a CAD model of a nominally perfect part that was fabricated with tolerances the CAD model does not capture. For low-volume job-shop work with high part variability — custom structural steel, repair welding, one-off heavy equipment — the teach pendant's ability to adapt to the actual part in the actual fixture is an asset, not a limitation.
The cost of teach pendant programming is robot downtime. A welding robot with a $150,000 purchase price amortized over 5 years running a single shift costs roughly $25 per hour to own. Add a $35/hour technician, and an hour of teach pendant programming costs the shop $60 in direct costs plus the $80 to $200 per hour in lost welding revenue that the robot would have generated if it had been welding. At 500 hours of programming per robot per year — typical for a high-mix fabricator — the annual cost of programming downtime is $40,000 to $130,000 per robot. This is the cost pool OLP targets.
How offline programming works — and what it cannot do
OLP software imports the part CAD, models the robot cell (robot, positioner, fixtures, torch), and lets the programmer define weld seams, torch angles, travel speeds, and weave patterns in a simulation environment. The software runs reachability analysis on the full robot arm kinematic model — flagging seams that exceed joint limits, cause collisions, or require reorientation of the positioner mid-seam — before any code leaves the engineering office. The output is a robot program in the native controller language (ABB RAPID, FANUC KAREL, KUKA KRL, Yaskawa INFORM) ready to load onto the physical controller.
What OLP cannot do: account for the difference between the CAD model and the actual welded assembly. Weldments distort. Tacked assemblies shift by 2 to 5 mm. Fixture pins wear. The OLP-generated program, run on a real part with real tolerances, will produce a torch position that is correct for the CAD model but 2 to 3 mm off from the actual seam. The dry-cycle verification step — running the program on the real part without striking an arc, verifying torch position at each seam — takes 30 to 60 minutes and catches the offsets. The technician adjusts the program — typically by shifting reference frames rather than re-teaching individual points — and production begins. If the dry-cycle step is skipped, the first production weld is a gamble on fixture precision.
Where does OLP produce the biggest ROI?
OLP's ROI is driven by three factors: part complexity (number of weld seams), production repeat frequency (how often the same part returns), and robot utilization rate (how close the robot already is to full capacity). A fabricator running 50 unique part numbers, each produced quarterly, with each part requiring 8 hours of teach pendant programming, spends 1,600 hours per year — 200 shifts — programming. OLP cuts that to roughly 400 hours (4:1 typical OLP time saving for complex parts), freeing 1,200 hours of robot capacity — the equivalent of adding 0.6 robots to the fleet without buying a single machine. If the fabricator is already at capacity, that recovered productive time converts directly to additional revenue.
The break-even is surprisingly low. At a $20,000 annual OLP license cost and a burdened shop rate of $100/hour, OLP needs to recover 200 hours of robot productive time per year — about 4 hours per week — to break even. For a fabricator with two or more robots and 20-plus unique part numbers, the break-even is typically reached within the first quarter. The limiting factor for OLP adoption is not cost — it is the skill barrier: the programmer must be proficient in both CAD and robot programming, a combination not common in the welding trade.
When should I stay with teach pendant programming?
Teach pendant remains the right choice for shops with fewer than 10 unique part numbers, very high part variability where every assembly fits up differently (repair and rework operations), or where CAD data does not exist for the parts being welded. If the shop's volume mix is over 70% one-off work — each part welded once and never again — the OLP time to import CAD, set up the simulation, and verify the program is comparable to simply teaching the part on the floor, and OLP offers no net time savings. The teach pendant also serves as the fallback for the 10 to 20% of seams that OLP positions suboptimally — the technician teaches those few seams manually on the floor after the OLP program runs, and the hybrid workflow captures most of the OLP time savings while maintaining weld quality.
For shops with parts too large or heavy to conveniently fixture in a consistent location — and where CAD models do not exist — teach pendant programming remains the only viable method.
