Seam Tracking in Welding: Laser, Through-Arc, and Vision Systems Explained

Updated August 2026

Seam tracking welding is in-process correction of a programmed path during welding. Select a system by evidence: can the joint produce a usable signal, can the controller turn it into stable motion, and can the complete loop pass a representative production test?

What Seam Tracking Corrects, and What It Does Not

Welding seam tracking is in-process path correction. A sensing system estimates the location of the actual joint, a controller converts that estimate into the machine coordinate system, and controlled motion adjusts the torch or workpiece within a qualified correction envelope.

Programmed paths represent where joints were expected to be. Production parts may differ because of forming, fixturing, tack placement, thermal distortion, or accumulated dimensional variation. Tracking can respond to the part that’s present, but only when the joint creates a usable observation and the rest of the control loop is able to act on it.

This peer-reviewed review of robotic welding sensing describes a broader sensing landscape that includes seam location, tracking, weld-pool observation, and quality monitoring. That breadth makes one distinction essential: when the observation occurs and what the controller does with it are separate questions.

Some systems perform more than one function. That does not erase the boundaries. Quotations and acceptance plans should state which output commands motion, which output is informational, and which separate method accepts the finished weld.

The Correction Loop Behind Every Tracking System

The tracker is a measurement-and-control chain, not just a sensor. Joint variation must be observed, transformed into the correct coordinates, filtered, delivered to the motion controller, executed within allowed limits, and checked against the weld result.

The Sensor-to-Torch Offset Stack exposes the parts of that chain that a sensor brochure can hide. Start with the real joint variation. Each sensor converts an observable feature into a position estimate or an error signal. Calibration and registration relate the sensor frame to the torch, robot, carriage, or workpiece frame. Filtering handles noise; the controller accounts for timing and decides which axes may change. Its motion system executes that decision, while the weld process continues to create heat, arc light, spatter, and distortion.

NIST’s calibration and registration work is a useful vocabulary check. Accuracy concerns closeness to a reference; repeatability concerns the spread when a condition is repeated; registration concerns the relationship between coordinate systems. A repeatable measurement in the wrong transform can still drive the torch to the wrong place. A catalog accuracy number does not close registration, timing, or motion error.

One common commissioning mistake is to move the sensor until one coupon looks right, then preserve no transform or configuration record. Such an apparent fix can’t be reproduced after a torch change, sensor service, software update, or fixture move. This complete stack should have a baseline and a defined recheck trigger before it can automate a real-time response at the planned travel speed.

How Laser Profile and Vision Systems Observe Joint Geometry

Laser profile trackers project structured light onto the workpiece and use a camera to calculate joint geometry. Other camera systems may use passive or separate illumination. Their advantage is direct geometric observation; their limitation is that the required feature must remain visible and correctly related to the torch.

TWI’s laser-welding seam-tracking guidance explains both pre-process sensing and in-process approaches. A sensor mounted ahead of the process can observe a cleaner joint, but it creates a physical sensor-to-process offset. Tight curves, corners, and orientation changes can make that look-ahead relationship difficult. Sensing near the process reduces the look-ahead problem but places the optical system closer to intense process emissions.

Laser lines do not create magic contrast. An algorithm still has to isolate a usable joint feature from surface reflectivity, preparation marks, tack welds, variable edges, arc light, smoke, and spatter. Protective windows and optical paths can become maintenance variables. These constraints apply to optical sensing; they should not be copied wholesale into a through-arc limitation list.

Procurement should therefore ask a specific question: which joint feature is extracted, under which surface and process condition, at what sensing location, and how is loss of confidence reported? Request sample images or profiles from representative parts. Include the least convenient orientation, expected tack pattern, surface range, start/stop area, and an intentionally difficult but production-realistic condition.

One 2021 experimental study reported tracking-error reductions of up to 38.38% and 41.71% for two directions in its particular robotic GMAW method and test setup. Those figures show that a defined laser-vision signal-processing method can be measured; they aren’t performance promises for another sensor, joint, or production cell.

When a supplier says “vision,” ask for the sensing modality and the operational job. One camera family may support seam finding, in-process tracking, weld-pool observation, or inspection, yet the configuration, timing, protective hardware, and controller connection can be different.

How Through-Arc Seam Tracking Reads the Welding Process

Through-arc seam tracking uses changes in welding-process signals to infer the torch’s relationship to the joint. It can avoid a separate optical view of the seam, but it depends on a stable, informative process signal and a compatible controller, joint, and motion pattern.

That peer-reviewed sensing review describes arc-signal approaches alongside visual, acoustic, and other methods. In a common through-arc arrangement, the torch samples different sides of the joint during a controlled weave. Differences in current or voltage behavior are processed into lateral and, where supported, height corrections. Arc behavior and wire position become part of the measurement chain.

That is why “no external sensor” should not be read as “no sensing constraints.” Wire extension, contact-tip condition, arc length control, transfer behavior, joint geometry, weave pattern, travel condition, welding parameters, power-source communication, and electrical noise can all affect the available signal. A stable weld process is not merely the output; it is also part of the input.

Do I Need to Use Weaving When Running TAST Seam Tracking?

Often, a through-arc algorithm needs a repeatable comparison across the joint, and weaving is a common way to create it. Blanket answers are not always appropriate. An applicable controller manual and application test must show how its algorithm obtains a lateral or height signal. If a vendor supports a different sampling pattern or limited straight-travel function, record that system-specific condition rather than calling it universal TAST behavior.

What Joint Types and Thicknesses Are Suitable for TAST?

No responsible answer starts with a universal thickness range. First ask whether the joint presents enough process-signal difference across the planned motion, whether the required weave and heat input are compatible with the welding procedure, and whether the controller supports the joint and welding mode. Groove and fillet applications are common candidates, but “common” is not an acceptance result. Prove the exact material, joint preparation, pass, position, process window, travel direction, and expected fit-up on representative coupons.

Ask to see the raw or processed signal during an induced joint deviation, the algorithm’s confidence or alarm behavior, the allowed correction axes, and what happens when the arc becomes unstable. If the system keeps sending corrections when its input is no longer trustworthy, the missing feature is not a better accuracy claim—it is a controlled failure state.

Separate Sensing Modality From the Job It Performs

A sensing modality describes what creates the observation; an operational job describes what the system does with that observation. The peer-reviewed sensing review supports this separation: laser profile, passive camera, through-arc, and tactile sensing belong on one axis, while finding, tracking, monitoring, and inspection belong on another.

This two-axis model prevents a false comparison such as “vision versus tracking.” Vision can be used for tracking. It also prevents the opposite mistake: assuming every product called a vision system closes the loop and commands motion.

On a quotation, replace “vision included” with four fields: modality, observed feature, timing, and output. That small change tells engineering and procurement whether two bids are actually comparable.

Use the 9-Row Joint-Signal Fit Board to Shortlist a System

The Joint-Signal Fit Board starts from what the application makes observable. It turns nine recurring joint conditions into a technology shortlist, a disqualifying representative-coupon test, and evidence to retain. Every row remains conditional until that test is passed.

Fill the 9-Row Joint-Signal Fit Board with production evidence before a demonstration. Its source categories follow the peer-reviewed sensing review, while the board itself is a buyer framework rather than a published standard. Its candidate column is not a ranking. It identifies a first test based on a visible feature, process signal, or accessible edge. A supplier may propose another method, but should explain what it observes and how a failed observation is reported.

Keep the shortlist to two primary candidates and one documented reason each could fail. That focuses the demonstration and protects procurement from comparing a proven configuration with an undefined promise.

Translate Sales Language Into Acceptance Evidence

Vendor vocabulary is useful only when it maps to a test. Use this compact translation board to keep tracking and weld acceptance evidence separate.

Language in the brief	Evidence to request
Joint and workpiece	State the seam type, material thickness, workpiece, actual seam, and fit-up. Name butt joints, lap joints, fillet welds, or any complex seam geometry.
Process signal	Record the current welding process, welding techniques, welding power source and other power sources, welding torch, wire feed, arc characteristics, weld bead, and weld bead formation within the welder-approved window.
Automation path	For automated welding, welding robots, or production lines, document the taught path, seam path, tracking point, data to the robot, correct position, permitted torch movement, and high-speed condition. Do not equate fully automatic operation with repeatable results.
Sensing claim	Ask how the seam tracker or proposed seam tracking solutions track the seam, perform joint tracking or real-time tracking, measure the distance, and compensate. Test sensing solutions on highly reflective parts; reject unsupported advanced systems, high precision, micrometer, or precise seam claims.
Commercial claim	For fabrication or shipbuilding, translate welding technology, automation, easy integration, easy programming, pre-configured, industrially proven, and return on investment language into buyer-owned acceptance criteria.

Integrate Tracking With the Robot, Controller, or Manipulator

After a sensor locates the joint, the project still needs coordinate frames, a controller interface, defined motion ownership, correction limits, alarms, recovery states, and retained data. Controlled motion and seam tracking are separate configured layers.

A robot may carry the torch, a carriage may move it along a beam, or a positioner may move the workpiece. An integration brief should say which controller receives the observation, which device owns each axis, and how simultaneous motions are coordinated. For long or circumferential work, a Schweißmanipulator can provide controlled positioning, but that doesn’t mean every manipulator includes a tracker, vision package, robot controller, or through-arc function. Project configuration must name them.

OSHA’s industrial robot guidance warns that erroneous inputs from peripheral equipment can lead to unpredicted robot operation.

That supports an integration warning, not a claim that tracking is a safety function. Safety-rated limits, protective devices, stop functions, risk assessment, and authorized recovery remain separate from the quality correction loop.

Commission the Complete Loop With a Proof Pack

The Commissioning Proof Pack links a representative joint, welding condition, tracker configuration, calibration, commanded and actual motion, alarms, and finished-weld result. It turns a demonstration into evidence the buyer can reproduce after a controlled change.

1. Freeze the test identity — record the joint, material, preparation, fit-up, position, pass, surface, fixture, and welding procedure condition.
2. Baseline the complete stack — save sensor, software, recipe, coordinate frames, calibration, controller interface, limits, and motion configuration.
3. Challenge the observation — include buyer-selected production variation and an expected weak condition without leaving the qualified welding or safety boundary.
4. Capture correction and failure behavior — retain observations, commands, actual motion, confidence, alarms, and recovery decisions.
5. Assess the finished weld separately — apply the project’s defined inspection and acceptance method; do not infer quality from tracking data alone.
6. Approve a bounded configuration — identify the pass condition, decision owner, controlled range, and changes that require revalidation.

Keep four evidence scopes separate. ISO 17662:2025 addresses calibration, verification, and validation of equipment used for controlling process variables in welding and allied processes. Where applicable, ISO 15613:2025 covers qualification based on a pre-production welding test, while ISO 5817:2023 establishes quality levels for specified fusion-weld imperfections. ISO 10218-2:2025 covers industrial robot applications and cells across integration and their lifecycle.

These scope references do not select a laser, camera, tactile probe, or TAST package. Nor does one passing coupon prove equipment calibration, procedure qualification, every finished production weld, and robot-cell safety at once. Assign an owner and record to each lane.

Consider a hypothetical comparison. Test A uses a clean, straight coupon with a visible joint profile and current calibration. Test B uses the same sensor on a curved assembly with tacks, a changed bracket, and longer look-ahead. Passing Test A does not transfer automatically: observation, registration, geometry, and timing changed. Re-establish those records and assess the finished weld; do not invent a generic correction allowance.

The Trend Is Sensor Fusion, the Buying Test Is Still Evidence

Newer seam-tracking research combines richer images, process signals, models, or multiple sensors to improve observability. That can expand what a system detects, but it also adds coordinate, timing, training-data, configuration, and failure-mode questions.

Recent work in Scientific Reports und Welding in the World illustrates the active development of visual recognition, sensing, and adaptive tracking methods. For procurement, the lesson is modest: better algorithms may make a previously weak feature usable under a studied condition. They do not remove the need to test the buyer’s joint or to control the path from sensing to motion.

For a fused system, ask what happens when inputs disagree, one sensor loses confidence, a model sees an out-of-distribution surface, synchronization drifts, or software changes. Retain both the combined output and enough underlying evidence to diagnose the result. Sophisticated sensing should make the proof pack richer, not optional.

Häufig gestellte Fragen

Related Resize reading: compare a pipe welding manipulator, a 3-axis welding positioner, and foundational weld positioners. For drawing terminology rather than path correction, see the separate guide to the seam weld symbol.

Referenzen und Quellen