Seam Tracking in Welding: How Robots Stay on the Joint

Picture a fabricator who spends six figures on a welding robot, only to watch it lay a perfect bead two millimeters off the joint because the last part in the stack was tacked slightly crooked. The robot did exactly what it was taught — the steel simply didn’t cooperate. That gap between programmed path and real-world joint is precisely what seam tracking exists to close.

What Is Seam Tracking in Welding? The Simple Answer Behind a Big Productivity Number

A robot that lays a flawless bead two millimeters off the joint is not broken. It is obedient. The fabricator in our opening story programmed a sensible path, clamped the part the way the work instruction said, and watched the robot follow orders perfectly. The steel simply arrived at the fixture with a slightly different idea about where the joint should be.

That is the problem seam tracking solves. In plain terms, seam tracking in welding is a closed-loop system that senses the actual joint during the weld and adjusts the torch path in real time. A sensor watches the seam, software calculates where the joint center sits relative to the torch, and the robot applies a small correction while the arc keeps running. This is genuine weld joint tracking: continuous measurement and correction, not a one-time guess before the weld starts.

What Seam Tracking Is, and What It Is Not

Seam tracking gets confused with several neighboring technologies. They solve different problems, and mixing them up leads to bad equipment decisions.

• Seam finding happens before the arc starts. The robot searches for the joint, locates the start point, and sometimes maps a few reference points along the seam. It answers, “Where does this weld begin?”
• CAD-to-path offline programming creates a nominal path from a 3D model. It is efficient, but it has no idea what the real part looks like after cutting, forming, and tacking.
• Fixture-based repeatability assumes every part lands in the same place. That works until fixture wear, spatter buildup, or a crooked tack weld enters the picture.
• Weld parameter control manages voltage, current, wire feed speed, and travel speed. It changes how the weld is made, not where the torch points.

Seam finding and seam tracking work best as a pair. Finding establishes a correct starting point; tracking holds the torch on the joint for the rest of the pass. A system that only finds can still drift across a long seam. A tracker that starts in the wrong place may spend its first few centimeters correcting an error that a simple search would have avoided. The trade-offs between robotic seam finding and seam tracking are worth understanding before specifying either one.

The Four Functional Blocks

Every seam tracking system, from a budget arc-sensing package to a high-end laser vision setup, performs four jobs in sequence.

1. Sensing

The system needs a signal that changes as the torch moves relative to the joint. That signal can come from the arc itself, a laser stripe and camera, a tactile wire touch, or another contact sensor. Through-arc tracking reads electrical behavior; laser vision seam tracking reads geometry.

2. Feature Detection

Raw signals are messy. The controller must extract the joint position from them, whether by analyzing current or voltage asymmetry during weaving, triangulating a deformed laser stripe, processing an image, or detecting the voltage spike of a wire touch.

3. Coordinate Transformation

The sensor rarely sits exactly at the torch tip. The controller converts sensor coordinates into robot and torch coordinates using calibration data, tool center point (TCP) values, sensor offsets, and look-ahead distance. Get this wrong and the robot corrects confidently in the wrong direction.

4. Motion Correction

Finally, the robot applies the offset. That may mean a lateral shift, a blended path adjustment, or a change in weave width. The correction has to be smooth enough to avoid a kink in the bead and fast enough to keep up with travel speed.

Where Seam Tracking Is Used

Seam tracking is common in robotic gas metal arc welding, and it also appears in laser welding, submerged arc welding, and cladding. Laser welding brings its own challenges, which is why TWI’s guidance on seam tracking for laser welding focuses on tight focal tolerances and clean optics. For a deeper look at laser-based systems, see our guide to Alpha laser welding systems, uses, and safety.

The productivity payoff comes from what tracking prevents: scrapped parts, manual touch-up, and long stretches of arc-off time spent reworking a weld that missed the joint. It buys tolerance. It does not make a badly fitted part disappear, which is why the next question worth asking is why seams move in the first place.

Why Seams Move: The Real-World Causes of Joint Variation

A welding robot does not miss a joint because it is lazy or badly programmed. It misses because the joint was never where the program said it would be. Between the teach pendant and the arc, a surprising number of forces push the seam sideways, up, or down. Understanding those forces is the difference between blaming the robot and fixing the process.

Repeatability Is Not Accuracy: The Misunderstanding That Sinks Projects

Here is the root of most confusion. Industrial robots are extraordinarily repeatable. Send one to the same coordinate a thousand times and it will return to within a fraction of a millimeter, every time. That is repeatability, and it is what robot brochures love to quote.

Absolute accuracy is a different number, and it is much worse. It describes how close the robot actually gets to a coordinate in real space, and it is affected by kinematics, gear backlash, thermal growth in the arm, and manufacturing tolerances in the castings. In practice, a robot can be excellent at repeating a point that is slightly wrong. As robotic seam-finding specialists point out, this is exactly why a taught path can be perfect on part one and off by two millimeters on part fifty. The robot did its job. The world around it shifted.

Thermal Distortion: The Seam Moves While You Weld

Heat is the least forgiving variable because it acts during the weld, not before it. As the arc deposits metal, the surrounding steel expands, then contracts as it cools. On a long part, that contraction pulls the joint out of alignment progressively along its length. A three-meter beam welded from one end will often show the seam creeping away from the torch midway through the pass, then settling somewhere else by the time the run finishes.

The effect is serious enough that researchers treat it as a dedicated engineering problem. A 2022 study in Scientific Reports designed a seam tracking device specifically to counter thermal deformation during tandem welding, where two arcs running close together concentrate even more heat into a narrow zone. If a lab rig needs a tracking system to hold the joint, a production cell welding eight hours a day certainly does.

Fit-Up, Fixtures, and the Human Factor

Not every deviation comes from heat. A lot of it is baked in before the arc ever strikes:

• Tolerance stack-up. Every part in an assembly carries its own permitted deviation. Stack four or five of them and the joint can legally land millimeters from nominal.
• Fixture wear. Clamps, stops, and locating pins wear, bend, and get replaced with subtly different parts. A fixture that was tight at commissioning may allow real movement a year later.
• Tack weld placement. Tacks pull the joint as they cool and can sit proud of the surface, giving a sensor a false feature to chase.
• Cutting and forming variation. Plasma kerf, laser edge quality, press-brake springback, and roll variation all change where the two edges actually meet.
• Clamping inconsistency. Different operators, different torque, different sequence. The part goes in slightly differently every shift.

Thin-gauge work shows this vividly. A sheet-metal enclosure that measures perfectly on the bench can bow half a millimeter once clamped, and aluminum makes it worse because it moves more with heat and is harder to hold flat. Shops working that material quickly learn that aluminum MIG welding demands respect for fit-up before anyone talks about tracking.

TCP Drift: When the Torch Itself Moves

The tool center point is the reference the robot uses for every motion. Bump the torch against a fixture, swap a torch body, change a contact tip and nozzle, or straighten a bent neck by eye, and that reference shifts. A small angular error at the torch multiplies over the length of the torch, and the whole path can land several millimeters off the joint even though the robot is moving exactly as programmed. Any cell that has had a crash and did not re-verify TCP afterward is a cell with a tracking problem waiting to be misdiagnosed as a sensor problem.

The practical takeaway: before buying sensing hardware, measure what is actually moving. Fixture wear and TCP drift are free to fix. Thermal distortion and stack-up are not, and those are the reasons seam tracking exists.

The Anatomy of a Seam Tracking System: Sensing, Detection, Transformation, Correction

Every seam tracking system runs the same four-step rhythm: sense the joint, detect its position, transform that data to robot coordinates, and correct the torch’s path. Miss any one of these blocks, and the system will drift, no matter how advanced its sensor may be.

Block 1: Sensing — Where the Information Comes From

Sensors fall into two broad categories: contact and non-contact. Contact sensors physically touch the joint, while non-contact sensors read it from a distance using electricity, light, or a laser. A more practical split comes down to energy source: arc-based, optical, and tactile.

Through-arc tracking uses the welding arc itself as a probe, watching changes in current and voltage as the torch weaves. Laser vision projects a structured line ahead of the arc and images its bent profile. Tactile sensing uses the wire-touch method, which is cheap and robust, but can only locate the joint before welding begins. Each of these carries its own ceiling for accuracy based on the physics behind it.

Block 2: Detection — Turning a Signal Into a Joint Position

Raw sensor data isn’t a coordinate. Detection is the math that extracts one. For through-arc tracking, the controller watches current or voltage oscillate during a weave. As the arc shifts towards a sidewall, arc length shortens and current rises. The asymmetry between left and right swings reveals the joint center, and averaging many cycles helps filter spatter noise. This is why a steady weave is required for arc tracking to work.

Laser vision takes a different approach. The camera sees a stripe deformed by the groove’s geometry and extracts details such as root position, groove edges, gap width, and mismatch. To do this reliably, laser sensors must combat arc light and spatter, which is often the greatest hurdle to overcome in development. Touch sensing is the simplest: the robot feeds wire down until contact, catches a voltage or current spike, and records that point as the joint location.

Block 3: Transformation — From Sensor Language to Robot Language

A sensor reports in its own frame, not the robot’s. Transformation bridges the two, and it relies on three things being correct: hand-eye calibration (sensor position relative to the torch), tool center point or TCP (where the wire tip actually is), and the sensor offset with its look-ahead distance.

These calibrations cap tracking accuracy more than many realize. For example, a 0.1 mm class laser cannot outperform a torch with a 1 mm TCP error. The sensor will be blamed, while the calibration is the issue. Look-ahead is equally important, as the sensor must measure ahead of the arc to allow time for a correction. Too much look-ahead can cause over-correction on tight curves before the arc reaches them.

Block 4: Correction — How the Robot Actually Moves

Most systems don’t rewrite an entire program to track. Instead, they apply a blended offset to the existing path, adjusting laterally, in height, or both. Some go a step further by changing weave width to center the arc within a groove. How smoothly this blend is applied depends on the controller’s real-time trajectory generation.

Timing is the limiting factor here. Control loop latency restricts performance at high travel speeds and small-radius corners. At 60 inches per minute, a 50-millisecond delay means the torch travels roughly 1.3 mm before a correction lands. On long straights this is negligible, but on a tight corner, it can be the difference between tracking and chasing the joint.

Comparing the Sensing Principles

Sensing principle	Typical resolution	Best joint types	Main weakness
Through-arc (current/voltage)	Roughly 0.5–1 mm in good conditions	V-groove, fillet, butt with gap, long curved seams	Requires weaving; struggles on narrow gaps and thin sheet
Laser vision (triangulation)	Sub-millimeter, often around 0.1 mm class	Thin material, tight butt joints, gap and mismatch measurement	Smoke, spatter on the lens, arc glare, reflective aluminum
Touch / tactile (wire touch)	Dependent on robot accuracy, typically a few tenths of a mm	Start-point finding, fixture error correction, simple fillets	Slow, cannot track during the weld, needs wire trimming

All of these blocks are only as good as the part presented to them. For reflective metals like aluminum with oxidized or oily surfaces, detection quality can drop before any correction is calculated. In those cases, proper aluminum prep before welding is just as critical to seam tracking as it is to arc stability.

Get all four right, and tracking feels almost boring: the torch simply stays in the groove. Get just one wrong, and the system spends its time chasing it instead.

Through-Arc Seam Tracking: How the Welding Arc Becomes the Sensor

Of all the ways to keep a torch on a joint, the most elegant uses the welding arc itself. Through-arc seam tracking, sometimes called through-the-arc sensing, infers joint position from changes in current or voltage as the torch oscillates across the seam. No camera, no laser stripe, no extra hardware bolted to the torch. Just the arc, a weave pattern, and a controller fast enough to read the signal and act on it.

That simplicity is why arc sensing remains the workhorse of heavy fabrication. Lincoln Electric’s technical material on intelligent robotic arc sensing describes the same core idea: the arc behaves differently depending on where it sits relative to the joint, and that difference can be measured.

The Principle: Reading the Joint Through Current and Voltage

The robot weaves the torch left and right across the joint in a controlled oscillation. As the arc swings toward a sidewall, arc length shortens and the electrical characteristics shift. In constant-voltage GMAW, current typically climbs as the arc nears the plate; on the opposite side of the weave, it drops. Submerged arc systems often read voltage instead, depending on power source mode.

The controller samples that signal continuously, filters noise from metal transfer, and compares the left half of the weave against the right. If the arc spent more time close to the left wall, the joint center is left of where the torch thinks it is. It calculates the offset and shifts the path laterally. Next weave, the signal is more symmetric. The loop repeats dozens of times per second, which is why the correction looks continuous rather than stepped.

Weaving Is Not Optional, and It Costs You

Arc sensing needs asymmetry to measure, and asymmetry requires the torch to cross the joint. A straight stringer bead gives the controller almost nothing. Every through-arc system therefore depends on a weave, and that weave adds heat input and cycle time. Weave width and frequency are the tuning knobs: a wider weave gives a stronger signal but more heat and slower travel; a narrow weave is faster and cooler but noisier.

Wire diameter and stickout matter too. Stiffer arcs from larger wire produce cleaner signals, while long stickout makes current readings noisier and more sensitive to torch height variation. Change wire size or stickout without updating the tracking recipe and the system starts chasing phantom offsets.

Joint Types and Process Compatibility

Arc tracking is happiest where there are sidewalls to sense:

• Works well: V-groove butt joints, fillet welds, U-grooves, butt joints with an open gap, and long or curved seams where the joint wanders gradually.
• Struggles: tight square butt joints with no gap or sidewall, thin sheet where weaving risks burn-through and distortion, and joints where the arc never gets close enough to a feature to register a difference.

GMAW and SAW are the sweet spot. Both produce strong, measurable electrical variation and are commonly run with weaving. TIG and plasma are harder because their arcs are stable and low-current by comparison, so signal-to-noise is poor. Optical sensing usually makes more sense there.

Arc sensing is largely immune to the smoke, spatter, and glare that degrade optical sensors, but not to material behavior. Aluminum is the classic example, where oxide and reflectivity change arc behavior in ways that unsettle the tracking loop, which is why shops working those alloys often lean on process control first, as covered in our aluminum MIG welding guide.

A Shipyard Scenario

Picture a panel line building deck stiffeners: twelve-meter fillet welds, dozens per shift, with fit-up that varies because the plates were cut on a different machine and tacked by hand. A laser sensor would need a clean lens in a cloud of fume and spatter. An arc sensor just needs the arc it is already running. The robot weaves, reads the imbalance, and walks the seam end to end, absorbing part-to-part variation that would otherwise force an operator to babysit the cell. Long runs and imperfect fit-up are the norm in marine work, and our AJ Marine welding guide covers the broader process picture.

Advantages and Honest Limitations

On the plus side: no extra sensor to foul, no lens to clean, no air curtain to maintain. It works in dirty, dark, fume-heavy environments, costs far less than laser vision, and tolerates the part-to-part variation heavy fabrication lives with.

The other side of the ledger matters just as much. The weave adds heat and time, so cycle times rise. Narrow gaps give weak signals. The system is sensitive to parameter changes, so a tweak to voltage, wire feed speed, or stickout can quietly degrade tracking. And unlike a laser, it cannot report gap width or mismatch before the arc starts. It corrects position, not fit-up.

For a wider look at how arc sensing stacks up against optical alternatives in production cells, Yaskawa Motoman’s discussion of the pros and cons of robotic seam finding and seam tracking is a useful reality check before you specify a system.

Laser Vision and Touch Sensing: Optical Precision vs Tactile Simplicity

Laser vision and touch sensing tackle seam tracking from opposite ends: one maps the joint in 3D before the arc, the other pokes it with the wire. Both work; neither is universally right.

Choosing well separates a system that works quietly for a decade from one switched off after three frustrating months.

Laser Vision: Reading the Joint as a 3D Profile

Laser vision projects a laser line or structured-light pattern onto the workpiece just ahead of the torch. An offset camera views that stripe from a known angle, and because emitter and camera are calibrated, any bend becomes height and position data. The controller fits a profile and extracts centerline, width, gap, mismatch, and sometimes the joint area available to fill.

That detail is the differentiator. The sensor reports not only that the joint sits 1.2 mm to the left, but that the gap has opened to 1.8 mm and the plates are mismatched by 0.6 mm, which lets a controller adjust wire feed or travel speed before the puddle burns through. TWI’s technical guidance on seam tracking for laser welding notes optical tracking is commonly specified for material under 1 mm thick or where variation stays below 1 mm, because there a millimeter of error separates a weld from a hole.

The strengths add up: sub-millimeter accuracy, often quoted in the 0.1 mm class on good surfaces; no weaving, so no extra heat input; fast updates; and solid performance on thin sheet where arc sensing has nothing to push against. On thin gauge it is often the only practical option.

Where Laser Vision Struggles

Optics live in a hostile place. Arc light washes out the camera image, fume scatters the beam, and spatter lands on the window within hours. Systems fight back with bandpass filters, shutter timing between arc pulses, and a compressed-air curtain or air knife across the lens. Skip the air supply and you will clean windows daily.

Reflective materials cause their own trouble. Aluminum throws the stripe back at odd angles, and heavy oxide or a scratched surface can scatter it entirely. If you weld much aluminum, the surface prep advice in this aluminum MIG welding guide applies doubly to optical sensing: a clean, consistent finish makes the measurement repeatable. Add a higher price than arc sensing and a calibration routine that demands care, and you have a technology that rewards disciplined shops and punishes casual ones.

Touch Sensing: The Wire as a Probe

Touch sensing, also called wire touch, finds the joint by contacting the workpiece with the wire and sensing electrical contact or mechanical compliance. The robot moves to a taught search point and feeds slowly toward the expected feature. When the wire touches metal, the controller records a voltage drop or deflection, then repeats at other search positions. Three touches in a V-groove, or a pair across a fillet, give enough data to calculate the true joint position and shift the program.

The advantages are hard to argue with: a fraction of a laser’s cost, no optics to foul, reliable in darkness and heavy fume, and very good at correcting fixture error and locating start points on thick plate. The trade-offs are clear. Every search is a slow, deliberate move; the wire must be trimmed to a known stickout first; it cannot follow distortion that develops mid-weld; and it struggles on joints without a clean measurable feature, such as a tight square butt.

Laser Vision vs Touch Sensing at a Glance

Factor	Laser Vision	Touch Sensing
Principle	Structured light triangulation, 3D profile	Wire contact via voltage or compliance
Accuracy	Sub-millimeter, roughly 0.1 mm class	Limited by wire stiffness and search resolution
When it works	Before and during the weld	Before the weld only
Best joints	Thin sheet, butt, lap, tight fit-ups	V-groove, fillet, thick plate start points
Measures gap/mismatch	Yes	No
Main weakness	Fume, spatter, reflectivity, cost	Slow, no in-process correction
Relative cost	Highest	Lowest

A Thin Stainless Scenario

Picture a shop running 1.2 mm stainless enclosures on a tired fixture. Fit-up varies by half a millimeter, tacks sit proud, and the arc wanders. Touch sensing finds each seam’s start, but by the time thermal distortion pulls the joint during a 400 mm run, that correction is stale. A laser watching the stripe ahead of the puddle sees the gap widen and lets the controller trim wire feed before burn-through. Thin stainless is laser vision territory, which is why shops invested in photonic processes standardize on it, as this Alpha laser welding systems guide shows.

Hybrid Approaches: Find with Light, Track with Arc

You do not have to choose. A common, cost-effective architecture uses a laser or touch sensor for seam finding at the start of each pass, then hands off to through-arc tracking for the weld itself. As Motoman’s comparison of robotic seam finding and seam tracking points out, the two functions solve different problems, and pairing them often beats over-specifying either one. Find the start accurately, follow the joint cheaply.

The decision comes down to thickness, joint type, surface condition, and your appetite for calibration discipline. Laser vision buys accuracy and information; touch sensing buys simplicity and robustness. Most successful cells buy one and design the fixture around it.

Seam Finding vs Seam Tracking: A Step-by-Step Setup and Commissioning Workflow

When it comes to seam tracking welding, the two actions that make the entire process possible are often mistaken for one another. Seam finding locates the joint before the arc strikes, whereas seam tracking holds the torch on the joint during the weld. Simply put, you must find the joint first, then track it. Without that locate step, any real-time correction is just guessing. With it, your robot can compensate for the small fit-up variations that would otherwise send your bead off course.

With that in mind, a reliable commissioning process isn’t rushed. The most common path is TCP validation, sensor calibration, nominal path teach, seam finding setup, tracking tuning, and finally, first-article verification. Follow these steps in order to build a stable baseline for your cell.

The Commissioning Sequence

1. Validate fixture and part consistency: Before touching the robot program, inspect the jig for wear or clamp shift. Even the best sensor cannot compensate for parts that are wildly inconsistent from one load to the next.
2. Establish accurate TCP: Verify your tool center point with a proper TCP calibration. A small TCP error will be interpreted by your sensor as a joint error, throwing every correction off.
3. Mount and calibrate the sensor: Physically secure your sensor to its bracket and complete its hand-eye calibration. This is the step that translates sensor measurements into usable robot coordinates.
4. Teach a nominal path with correct work and travel angles: Teach an “ideal” path off of a perfectly fit sample. Keep your work and travel angles consistent, as changing them will impact how your sensor reads the joint. This is especially important when welding more sensitive alloys, such as those covered in our guide to aluminum MIG welding.
5. Set seam-finding search parameters and offsets: Program the search area for seam finding. For most setups, a search distance of 5-10 mm is a safe starting point to locate the joint center before the arc strikes. Apply any needed pre-weld offsets based off of your calibrated sensor.
6. Set tracking gains, deadband, and look-ahead: This is where tuning begins. As a starting point for many vision or arc sensing systems, use a tracking gain between 0.3 and 0.6, a deadband of 0.1 to 0.3 mm, and a conservative look-ahead distance. Keep in mind that these are only starting points to tune, not universal values.
7. Run first-article trials with gap/mismatch coupons: Run your program on sacrificial parts with intentional, measured variations to see how the corrections react in real time.
8. Verify with destructive or visual inspection: Inspect for proper fusion, leg size, and penetration. Adjust parameters gradually from your baselines, changing only one variable at a time.
9. Document parameter sets per part number: Save every final recipe to its respective part number with sensor calibration dates attached.
10. Build a maintenance schedule for lens cleaning and calibration checks: For optical systems, schedule routine air purge checks and lens cleaning. For all systems, log calibration checks at a fixed interval to catch drift before it becomes scrap.

Acceptance Testing With a Deliberate Offset Coupon

The most conclusive way to prove your system is correcting and not just compensating blindly, is to lie to it on purpose. Take a clean, representative coupon for the joint you’ll be running, and deliberately offset the nominal taught path by 2-3 mm laterally from the true joint center. Run the full cycle with seam finding enabled, allowing the system to locate and track the actual joint.

Once complete, inspect the bead. A passing result will have the torch-corrected bead sitting directly over the true joint, with acceptable fusion and leg size, despite your program being taught off-center. If the bead still follows your offset, or wildly oscillates trying to find it, revisit your seam-finding offsets, TCP, or tracking gain before moving onto production. This simple test takes minutes, but saves hours of rework.

The Business Case: Where Seam Tracking Pays Off and Where It Doesn’t

Seam tracking isn’t about having the flashiest automation on the floor. It’s about dollars and cents. The question isn’t “can this system track?” but “will it pay for itself?” The honest answer: sometimes yes, sometimes not at all.

The Payoff: What Actually Saves Money

Two wins drive everything else: higher arc-on time and reduced rework.

• Less rework and scrap: Correcting joint drift in real time cuts beads that land off-joint, under-penetrate, or get overbuilt. That matters on complex work like aluminum boat welding, where inconsistent fit-up is expensive to put right.
• More arc-on time: Fewer manual adjustments and less rework mean the cell spends more hours laying weld metal instead of stopping to tweak.
• Less fixturing: Tracking absorbs the small part-to-part variation that would otherwise demand a new jig or a hyper-precise fixture. Fixtures don’t disappear, they just stop being the bottleneck.
• Unattended capability: On the right jobs, this opens the door to lights-out runs, freeing skilled welders for setups, inspection, or intricate work like aerospace welding.
• Consistent quality: Steady penetration and leg size across a whole batch, with less operator fatigue.
• Welds previously out of reach: Some crooked, inconsistent parts finally become worth automating instead of being handed to a manual welder.

A Worked Example

Take a difficult part in a robotic cell. Before tracking, it runs 60% arc-on time, an 8% rework rate, and a 15% scrap rate — common numbers for jobs with inconsistent fit-up.

A well-commissioned system won’t fix everything. A modest, realistic improvement to 4% rework, 7% scrap, and 72% arc-on time compounds quickly. The return comes from more usable welds plus less touch-up labor. It isn’t one big number; it’s dozens of small savings stacking over thousands of parts.

The Costs to Consider

A tracking system is more than a sensor purchase. Budget for the sensor itself, integration into the controller, programming and commissioning, operator training, and ongoing maintenance.

Costs vary widely by cell, so get quotes from reputable sensor and torch vendors and judge them on fit for the application, not commission alone.

Where It Doesn’t Pay Off

• Very low volume or one-off work: a handful of runs will never recoup programming and integration.
• Joints with no measurable feature: if the joint is too open or lacks a distinct edge, the sensor has nothing to correct to.
• Tolerances already fine manually: if a manual welder hits spec with no rework, automation isn’t the answer.
• Poor fit-up with excessive gap or mismatch: the big one. Tracking can’t create metal where there isn’t any, so fit-up still has to be controlled. For process hazards and controls, see OSHA’s welding, cutting, and brazing hazard guidance.

Decision Checklist

1. Consistency: Is part-to-part variation too much for your current fixture but still inside the tracker’s correction range?
2. Volume: Will you run enough parts to amortize integration, training, and maintenance?
3. Cost of failure: Is rework, scrap, or touch-up expensive enough on this part to justify the spend?
4. Fit-up control: Can fit-up be brought into a trackable range first? Fix that before buying any sensor.
5. Application fit: Does the joint, material (say aluminum, covered in this aluminum MIG welding guide), and process suit arc sensing or laser vision?

Seam tracking buys tolerance, not neglect. Control fit-up first and have the volume to back it up, and the compounding savings from arc-on time and rework are very real. If you can’t, your money is better spent on fixturing or part presentation.

Failure Modes and Troubleshooting: When Tracking Makes Things Worse

Here is the frustrating part of seam tracking: when it fails, it usually fails loudly. The torch darts side to side like it is searching for something it lost. The bead starts dead center and then wanders off the joint halfway down the part. Or the robot locks onto a tack weld and happily welds six millimeters away from where it should be. To an operator, all three look like “the sensor is broken.” In practice, most tracking faults trace back to setup, calibration, or contamination rather than failed electronics, which is exactly why Motorman-style guidance from integrators stresses that seam finding and seam tracking solve different problems and need different tuning Yaskawa Motoman.

The good news is that seam tracking failures are highly diagnosable if you read them as symptoms instead of errors. Below are the five patterns that show up most often in real weld cells.

Torch Hunts or Oscillates Side to Side

This is the classic complaint. Excessive tracking gain and a long look-ahead distance are the two most common causes of torch hunting and oscillation. The controller sees a small deviation, overcorrects, sees the overshoot, corrects again, and the whole system starts chasing itself down the seam. Turn the gain down before you touch anything else. Then shorten the look-ahead so the correction applies closer to where the sensor actually measured. Noisy signals and a sloppy tool center point (TCP) produce the same behavior, because the controller is reacting to bad data as if it were real joint movement.

It Tracks Fine at the Start, Then Drifts

Early success followed by drift usually means the joint moved outside the correction range, or the sensor stopped seeing clearly. Thermal distortion is the usual suspect on long parts, and a fouled window or a loose ground connection will mimic it. Check the lens and the work clamp lead before you rewrite the program.

It Tracks the Wrong Feature

Tack welds, spatter blobs, and the edge of a wide root gap all look like joint geometry to a sensor with the detection window set too wide or the threshold set too loose. On optical systems, the culprit is often poor laser stripe contrast. Arc light, spatter, and fume are documented causes of false features and signal loss in optical seam tracking Sensors (MDPI), so narrow the window, tighten the threshold, and confirm the stripe is crisp on the actual surface.

Mild Steel Works, Aluminum Does Not

Aluminum is the standard test of any optical tracker: high reflectivity, a tough oxide layer, and harsh arc glare scatter the stripe. Cleaning and surface prep matter more than most shops expect, and our guide to prepping aluminum for welding covers the practical side. If you are tuning the process around those parts, the complete aluminum MIG welding guide is worth reading alongside this section.

Intermittent Faults After a Crash

A bumped torch or a bent sensor bracket shifts the TCP and the sensor offset in one go. The weld still runs, so nobody notices until the bead starts missing. Re-check TCP and bracket straightness after every collision, not just the big ones.

Symptom	Likely Cause	First Action
Torch oscillates or hunts	Gain too high, long look-ahead, noisy signal, poor TCP	Reduce gain, then shorten look-ahead
Tracks at start, drifts later	Thermal distortion, fouled window, ground/cable issue	Inspect lens and work lead
Locks onto wrong feature	Detection window too wide, wrong threshold, weak stripe contrast	Narrow window and re-teach feature
Fails on aluminum only	Reflectivity, oxide, arc glare	Clean surface, adjust exposure/filtering
Faults after a crash	TCP shift, bent sensor bracket	Re-calibrate TCP and bracket
Arc tracking drifts on gap variation	Weave or arc signal disrupted by parameter changes	Verify weld parameters and weave width

Preventive Maintenance That Actually Prevents Faults

• Lens air purge: keep positive air flow across the optic during every arc-on second; a dry, dirty lens is the fastest route to signal loss.
• Spatter guard: inspect and replace on a schedule, not after a failure.
• Weekly calibration coupon: run a known-offset test part and confirm the system still finds it.
• Spare sensor bracket: keep one on the shelf so a crash costs hours, not days.
• Log tracking corrections: record correction magnitude per part and per shift.

That last item deserves emphasis. Logging the magnitude of tracking corrections over time reveals fixture wear and TCP drift before scrap appears. A correction that creeps from 0.2 mm to 0.8 mm over three weeks is a warning, not a nuisance.

None of this replaces good process discipline. Tracking is a compensation layer, and compensation layers amplify whatever you feed them, good or bad. For a deeper look at how the arc itself supplies that signal, see the through-arc tracking explanation earlier in this guide, and remember that OSHA’s welding hazards guidance still applies while you are leaning into a cell cleaning a lens.

Common Mistakes, Safety, and a Practical Buying Checklist

Most seam-tracking projects that disappoint do not fail because the sensor was bad. They fail because the cell around it was not ready. The same mistakes show up in job shops and heavy-fab plants alike, and nearly all of them are cheaper to fix before the purchase order than after.

Seven Mistakes That Undermine Good Hardware

• Buying a sensor before fixing fit-up. Tracking compensates for small, gradual variation. It cannot rescue a joint that moves 6 mm from part to part. Fix cutting, forming, and tacking first.
• Skipping calibration. A sensor that is mounted but not properly calibrated will report confident, precise, wrong numbers. Hand-eye calibration and TCP verification are not optional extras.
• Treating tracking as a substitute for fixturing. Good clamps reduce how hard the system has to work. Loose fixtures force the sensor to chase movement instead of following a stable seam.
• Ignoring the cycle-time cost of weaving. Through-arc tracking needs an oscillation, which adds travel time and heat. On a high-volume cell, that can erase the productivity gain.
• No spare parts or training. A cracked lens, bent bracket, or failed cable can idle a cell for days. Budget for consumables and structured welding training before commissioning, not after the first breakdown.
• Failing to document recipes. Every part number and joint type needs its own saved parameters. Tribal knowledge walks out the door with the operator.
• Expecting one configuration to handle every joint. A recipe that works on a thick fillet will not automatically suit a thin square butt. A single tracking setup rarely fits an entire shop.

Aluminum adds its own wrinkles: reflectivity, oxide, and fit-up all affect what the sensor sees. Our aluminum MIG welding guide is a useful companion when tracking thin, reflective parts.

Safety and Compliance Around an Automated Cell

A tracking system makes the robot more independent, which also makes guarding and procedure more important. OSHA identifies metal fumes and ultraviolet radiation as primary health hazards in welding, cutting, and brazing. Robotic cells add motion, pinch points, and automatic restart hazards, so they fall under machine-guarding requirements. OSHA’s robotics guidance is a sensible starting point for risk assessment.

Practical controls include perimeter guarding with interlocked gates, light curtains or area scanners where appropriate, and a lockout procedure for maintenance. Operators need arc-rated clothing, eye and face protection, and screening against UV exposure. Fume extraction should be designed for the actual process and position, not added as an afterthought. Consensus standards such as ISO 10218 and ANSI/RIA R15.06 address industrial robot safety generally; your integrator should confirm the exact edition and local requirements that apply. Never let anyone reach into the cell during automatic motion, even for a quick wire trim.

A 12-Point Buying Checklist

1. List every joint type and the worst realistic gap and mismatch for each.
2. Measure part variation across at least 30 real pieces, not one perfect sample.
3. Audit fixture repeatability and verify TCP accuracy before shopping.
4. Require a demo using your material, your joint, and your worst coupon.
5. Confirm the sensor’s field of view and look-ahead fit the torch and part geometry.
6. Check compatibility with your robot controller and fieldbus.
7. Price integration, calibration, and programming, not just the sensor box.
8. Make a witnessed acceptance test part of the purchase agreement.
9. Stock spare lenses, air-purge filters, brackets, and cables.
10. Budget training for operators, maintenance techs, and programmers.
11. Require recipe backup, documentation, and change-control procedures.
12. Agree on warranty terms and technical support response times.

Cobot or Traditional Cell?

A cobot with integrated tracking can be the better fit for low-volume work, small parts, frequent changeovers, and shops without a dedicated robot engineer. It is easier to redeploy and often simpler to program. A traditional cell still wins on large fabrications, high duty cycles, fast travel speeds, and harsh environments where payload and reach matter. Collaborative operation is not automatically safe operation; the same risk assessment applies.

Affiliate disclosure: if this article links to a sensor, torch, or software vendor and you buy through that link, we may earn a commission at no extra cost to you. That never changes our recommendations.

The Bottom Line: Seam Tracking Is a Fit-Up Strategy, Not a Magic Wand

Seam tracking is best understood as a compensation tool. It helps a robot follow a joint that has moved, but it cannot turn a poorly fitted part into a good weld. The moment a shop treats tracking as a substitute for fixture discipline, the technology starts covering up problems instead of solving them. Seam tracking compensates for variation; it does not eliminate the need for fit-up and fixture discipline.

That reframe matters because the market makes tracking sound automatic. In practice, the order of operations is process control first, then fixturing, then sensing, then automation. Get the process stable, make the fixture repeatable, and only then add a sensor to handle the variation that remains. A robot with perfect tracking and poor grounding, dirty wire, or inconsistent clamping will still produce inconsistent welds.

Choose the Sensing Method That Fits the Job

• Through-arc sensing: the budget-friendly choice for long, curved, thicker joints. It needs a weave and works best on V-groove, fillet, and gapped butt joints. No extra torch hardware, but it is slower and less precise on narrow gaps.
• Laser vision: the precision choice for thin material, tight tolerances, and jobs where gap and mismatch must be measured. It can see ahead of the arc and adapt fill, but smoke, spatter, and reflective aluminum demand care. If aluminum is in the mix, our aluminum MIG welding guide covers the surface-prep issues that affect both the weld and the sensor.
• Touch sensing: the cheap, robust choice for finding start points and correcting fixture error before the arc starts. It cannot track during the weld.

Research is pushing the field forward. AI-based seam extraction, multisensor fusion, and tighter integration with weld data monitoring are all active areas, and sensor fusion that combines arc and optical data is showing promise for more robust seam extraction (Nature). The direction is clear, but the shop-floor rule has not changed: sensing is the last layer, not the first.

Back to the Fabricator

Return to the fabricator from the opening. The robot was never the problem; the stack was. A realistic fix is unglamorous. The team checks the fixture stops, verifies the TCP after the last torch change, and runs a first-article coupon with a deliberate offset. Then they add through-arc tracking to the long, curved seams and keep touch sensing for start-point finding. The robot still follows the joint, but the team now spends its time improving fit-up instead of chasing the torch.

Five Things to Do This Month

1. Audit fixture repeatability on your worst part and record the actual gap and mismatch.
2. Verify TCP and sensor calibration before changing any tracking parameter.
3. Trial one sensing method on one problem seam, not the whole product line.
4. Write a recipe per part number and document the tuning values.
5. Log tracking corrections weekly to spot drift before it becomes scrap.

Seam tracking is a powerful way to buy tolerance, not a cure for loose fit-up. Control the process, tighten the fixture, then let sensing handle the rest. For a broader look at where welding automation is heading, see the AWS overview of automation and inspection. If you want to go deeper, browse our related welding guides on robot setup, sensing, and material behavior.

Frequently Asked Questions

What is the difference between seam tracking and seam finding?

Seam finding locates the joint before the arc starts, usually by touching the part with the wire or scanning it with a laser. Seam tracking corrects the torch position continuously while the weld is being deposited. Most production cells use both: find first, then track.

Do I need a laser sensor to do seam tracking?

No. Through-arc tracking uses the welding arc itself as the sensor by weaving and measuring current or voltage changes, and it costs far less than laser vision. Laser systems are worth the extra cost when you need high accuracy on thin material or want to measure gap and mismatch.

Can seam tracking fix poor fit-up?

Only within limits. Tracking corrects position, and some systems adjust for gap, but a joint with excessive gap, severe mismatch, or missing material will still produce defects. Fit-up and fixturing must be controlled before tracking can help.

Does through-arc tracking slow down the weld?

Yes, somewhat. Arc sensing requires a weave or oscillation, which adds travel time and heat input compared with a straight bead. The trade-off is usually acceptable on thick sections and long seams where the alternative is rework.

Why does my robot track perfectly on the first part and drift on later ones?

The usual causes are thermal distortion accumulating as the part heats up, fixture wear, or a slowly fouling sensor window. Check whether the correction magnitude is climbing over the run, and verify TCP and sensor calibration.

What accuracy can seam tracking realistically achieve?

Laser vision systems are commonly specified in the sub-millimeter range, with vendors quoting figures around 0.1 mm under good conditions. Through-arc tracking is typically coarser, often in the 0.3-0.5 mm range, and depends heavily on weave parameters and joint geometry.

Is seam tracking only for large companies?

No. Entry-level arc sensing and touch sensing are affordable for small shops, and some cobot packages include seam finding as standard. The real cost is usually integration and programming time, not the sensor itself.

What maintenance does a laser seam tracking system need?

Expect daily checks of the lens and air purge, spatter guard replacement as needed, periodic calibration verification with a test coupon, and inspection of the mounting bracket after any crash. Dirty optics are the most common cause of intermittent tracking faults.