Resistance spot welding controllers: a benchmark of the models most used in the automotive industry

A body leaves the body shop with around 4,000 spot welds, and more than 90% of everything joined there goes through resistance spot welding (16). Multiply that by the 92 million vehicles the world produces each year and you arrive at something close to 370 billion joints. What decides how each one of them is made, in a few tens of milliseconds, is the welding controller.

But choosing a controller is not choosing a brand. You are choosing a power source topology, a control strategy and a data model, and those three decisions will determine, for the next ten or fifteen years, how much of its own process your line is able to see.

Before we begin, a word about what this text is. It was assembled from the technical documentation the manufacturers themselves publish (18) and from the literature listed at the end. We ran no comparative trial and took no independent measurements, so what is described here is what each manufacturer states about its own product. We are not naming the best controller on the market, nor recommending a purchase, not least because that depends on the installed base, on the material being welded and on the plant’s data strategy. The aim is simpler: to lay out the axes that genuinely differentiate this class of equipment, so that anyone working with spot welding has a map.

What the controller governs

By Joule’s law (1), the heat generated depends on the square of the current, on the total resistance and on time. The controller commands directly: current, time and force. Resistance is influenced by electrode force, inversely (17). Broadly, as force rises, resistance falls.

And that resistance does not sit still. It is the sum of everything opposing the current in the secondary: the electrode-to-sheet contact on each side, the sheet material itself, the sheet-to-sheet interfaces (2). The contact resistances are the sensitive ones, because they respond to oxide, coating, pressure distribution and worn caps, and they are the ones that change from one weld to the next. That is why the dynamic resistance curve became the industry standard instrument: it shows in real time what is happening in there. The idea is not new, it was established in 1980 (5), and today it underpins practically every commercial monitoring system (6).

Axis 1 · Power source topology

A thyristor controller, connected to the AC mains, can only act once every half cycle, which gives a reaction time of 10 to 20 ms. A medium-frequency inverter at 1,000 Hz responds in around 1 ms.

That looks like a catalogue detail, but it is worth doing the arithmetic. The nugget forms over 100 to 400 ms. With 20 ms of reaction time you get a few dozen chances to intervene across the whole weld; with 1 ms, you get hundreds. Manufacturers of adaptive systems are blunt about thyristors being too slow to run their algorithms (18). In practice, whoever operates an AC installed base monitors, and whoever wants to adapt needs medium frequency.

Axis 2 · The control strategy

The oldest mode holds some programmed electrical quantity constant, usually current (CCC, constant current control), sometimes voltage, power or firing angle (18). It works well as long as nothing moves. The problem is that something always does: shunt, gap, structural adhesive at the interface, thickness varying within stamping tolerance, caps wearing down. All of it delivers a different result under the same programmed current (4).

Adaptive strategies attack that problem by reproducing a key variable from a previously validated weld, the master weld, or reference weld. What changes from one to another is which variable the controllers choose to chase.

  • Matching the energy. The controller tracks instantaneous power and adjusts current until the energy accumulated by the end of the cycle matches that of the master weld. If resistance falls, it raises the current. It is a scalar target, it tolerates noise and it is simple to implement. The limitation lies in the nature of the quantity: energy is an integral, and an integral does not remember the path. Two welds can receive exactly the same total energy, distributed differently across the cycle, and come out with different nuggets and different microstructures. That is why constant-heat approaches cannot optimise the relationship between nugget diameter and energy spent: the process is far too non-linear for that (18).
  • Chasing the master curve. Here you do not match a total, you chase a path. The dynamic resistance profile of the master weld is recorded and current is adjusted continuously so that instantaneous resistance follows that trajectory (9). The thermal history is preserved, not just the sum. Some commercial implementations simplify this and chase only landmarks of the curve: the instant at which resistance reaches its maximum, for example, serves as a trigger to stretch or shorten the weld time (18).
  • Chasing the mechanical signal. Same logic, applied to the electrode displacement curve, which is not raw stroke but the distance between the two tips, with the elastic deflection of the arm already subtracted. When the material melts it expands and pushes the electrodes apart, so this curve is a far more direct observation of nugget growth than the electrical one (10).

Before choosing between the second and the third, it is worth knowing an experiment that compared the two side by side, published in the Journal of Manufacturing Processes by Shanghai Jiao Tong University in partnership with the General Motors research centre (10). They welded 0.8 + 0.8 mm DP590 on a bench-mounted servo C gun, instrumented with a displacement encoder and a force sensor, adjusting current period by period until the measured signal met the reference. It is not a commercial closed-loop controller, it is the controlled trial that isolates the two strategies. And they deliberately provoked two conditions any body shop knows: an initial gap between the sheets, and a weld close to the edge.

Both conditions reduce both signals relative to the master weld: dynamic resistance falls and electrode displacement falls. But current pushes those two signals in opposite directions, and that is where the difference between the two strategies comes from.

Current and displacement move together, because displacement is thermal expansion: more current, more heat, the more the material expands and pushes the electrodes apart. Resistance moves the other way. More current softens the asperities at the interfaces, the real contact area grows and the measured resistance falls; with less current the asperities soften less, the contact area grows less and the resistance read stays higher. The authors record this behaviour as an experimental observation, and from it comes the rule for each control scheme: to bring resistance back to the reference the controller must reduce current; to bring displacement back, it must increase it.

The result was the opposite of what you would expect. Chasing resistance, the curve met the reference and heat input collapsed, from 869 J to 584 J at a 2 mm gap and 8 kA: diameter and failure load came out worse than with no adaptation at all. Chasing displacement, heat input rose and the weld improved, in one case even surpassing the reference itself, but under severe conditions the increase in current brought internal expulsion forward and the nugget shrank again. And the finding that stings most: even with the measured signal adjusted until it coincided with that of the good weld, significant differences in diameter and in strength remained. For everything else the authors give a closed causal chain, current to heat input to nugget size; this last point they leave on record as a phenomenon that still requires study. Making the signal match does not guarantee the weld matches.

There is also a fourth approach, which does not compete with the other three: separating monitoring from control, letting a supervisory layer run on top of any controller, including one from another manufacturer. It adapts nothing, but it solves the mixed installed base problem that almost every plant has.

Axis 3 · What is measured, where, and at what rate

Something that tends to go unnoticed: dynamic resistance is not measured, it is calculated, by dividing voltage by current. So where the voltage tap is connected changes what goes into the calculation. If it sits far from the joint, the value read carries with it the entire resistance of the secondary loop, meaning busbar, cable, articulation, electrode holder. That share varies with the temperature of the equipment and with the mechanical state of the gun and has nothing to do with the weld, and the sheet-to-sheet interface, the only one that matters, becomes a small fraction of a large number (2). Sampling rate has a similar effect: it determines whether the system sees the local detail of the curve and events lasting less than a millisecond, such as expulsion (6).

But the serious limitation is not in the electrical domain. It is in how much information is extracted from it. Spot welding couples four domains: electrical, thermal, mechanical and metallurgical (2). Force sets the real contact area, area sets resistance, resistance sets heat, heat softens the material and changes the area again. And it is that same heat, together with the rate at which it is removed, that sets the resulting microstructure, which in turn accounts for the strength of the joint (15). All of this leaves its mark on the electrical signal, and the dynamic resistance curve is precisely where that coupling becomes observable. It is the route to a deeper reading of the process. The limit is not in the curve, but in the use made of it when it is treated merely as a trajectory to be matched to that of a master weld.

Reduced to that, it becomes ambiguous. Different physical situations produce similar curves. A worn gun delivering less force and an excess of oxide at the interface raise contact resistance in the same way, yet the correction for each case is the opposite. The gap is more treacherous still: once contact is established, it may simply not appear in the resistance curve, even with the effective force at the interface below design, while the displacement curve flags it immediately (10).

There are two ways out of that ambiguity. The first is to add sensors: measuring force and stroke alongside current and voltage, synchronously, is what consolidated multisensor fusion in the monitoring literature (7), (8). The second is to extract more from the signal already at hand. Dynamic resistance is only one of the quantities that can be extracted from the electrical signal: there are also the inflection instants, the rates of change, the expulsion signature, the phase-by-phase behaviour of the cycle and the relationship between current and voltage over time. Working with a rich set of features rather than a single trajectory, the electrical reading becomes sufficient again to say what happened at the joint. The difference is between monitoring a curve and reading the signal.

The comparison

First the quick read, then each manufacturer’s card with the detail. A note on the last column: by data layer we mean here the export of what was measured at each spot, that is, the dynamic resistance curve, current, voltage, force and stroke. It is not the same thing as the ability to store weld schedules, which is internal parameterisation, nor a fieldbus, which serves to command the equipment and exchange status signals.

Manufacturer Topology How it adapts Joint integrity Data layer
WTC MF + AC Against a reference, adjusting current and time Nugget Integrity View-R · MiT
Bosch Rexroth MF Against a master curve, plus force feedback PSQ 6000 · analytics MQTT · OPC UA
Matuschek MF 1,000 Hz Against a recorded master weld, with stepper NUGGET_Index_ MULTI04 · proprietary
Harms & Wende MF · AC · HF · CD Three routes: iQR, iQFlex and HSC IQ-Inspector PQS / XPQS
ARO MF Its own adaptive mode SQA no public information
Obara MF Constant modes plus adaptive against a reference SQA local logging · USB
Nadex MF + AC no public information no public information no public information

WTC

United States

Line
WT6000 MFDC and WT6000 AC; 2400 series
Topology
Medium-frequency inverter, programmable from 400 to 2,000 Hz; and alternating current
Adaptation
Adapts weld current and time in real time against a reference, to compensate for disturbance and minimise expulsion (RAFT)
Integrity
Nugget Integrity: the ratio between the weld performed and the nugget size of the reference weld. Accompanied by tooling integrity, process integrity, dressing verification and expulsion detection
Data
View-R and ViewNet. The MiT module instruments a legacy controller with a Rogowski coil; up to 48 modules per concentrator

Bosch Rexroth

Germany

Line
PSI 6000 (previous generation, still supported and upgradable); PSQ 6000; PRC7000 (current generation)
Topology
Medium-frequency inverter
Adaptation
Real-time comparison against a master resistance curve, on a millisecond basis. Force feedback that reads the thermal expansion of the stack and regulates pressure and current, with its own patent and an emphasis on aluminium. Waveform customisable in 100 or more heat blocks per program
Integrity
Online quality verification in the PSQ 6000. An analytics layer correlates quality with part fit-up and secondary circuit degradation. Per-spot evaluation criterion not detailed publicly.
Data
IoT Connector with MQTT and OPC UA, more than 200 values per spot; native in the PRC7000 and available as an expansion for the PSI 6000. Two-processor architecture. Web interface. Cloud analytics

Matuschek

Germany

Line
SPATZ+ and ServoSPATZ+ (M400, M600, M900), integrated with its own servo heads
Topology
Medium-frequency inverter at 1,000 Hz. Output of 450 to 950 A on the primary of the welding transformer; the current at the joint is that of the secondary, after the transformation ratio. Master-slave configuration for higher current
Adaptation
Adaptation against a recorded master weld, with current adjustment and extension or reduction of time, and an automatic stepper function. Versions for steel (MASTER) and aluminium (AluMASTER)
Integrity
NUGGET_Index_: condenses the electrical and mechanical curves of each weld into a single value, with a declared correlation to the diameter of the spot mark
Data
MULTI04 weld monitor and proprietary software. The public documentation does not detail the export model or the protocols.

Harms & Wende

Germany

Line
Genius and GeniusHWI
Topology
Medium frequency, alternating current, high frequency and capacitor discharge. Family from 250 to 3,500 A of inverter output, air or water cooled
Adaptation
Three distinct routes. iQR regulates current by the resistance or power curve and adjusts time as a function of the instant of maximum resistance, with dedicated upslope and downslope. iQFlex recalculates process resistance every millisecond and adapts from the very first spot, dispensing with a master weld, with contact preconditioning. HSC is intended for projection welding in high-strength steels
Integrity
IQ-Inspector monitors the quality of each weld. A set of inspectors per quantity: current, voltage, resistance, regulation limit, force, stroke, process stability, component presence and spatter occurrence
Data
PQS and XPQS with a measurement module that operates over any controller. Current, voltage, resistance, power, force and stroke synchronised at up to 36 kHz. Networked version

ARO

France

Line
MFDC cabinets SW 560, SW 800, SW 1200 and SW 2400, in robotic, manual and machine versions. Stores up to 256 weld programs
Topology
Medium-frequency inverter
Adaptation
Adaptive mode that compensates for material and thickness, cap wear, adhesive presence, poor fit-up and shunt effect. Algorithm not detailed publicly
Integrity
SQA: assigns a quality percentage to each spot and detects process disturbance, including mechanical and electrical failure, shunt, different coating, adhesive and electrode slippage
Data
No public information on the export of the data measured per spot. What the documentation does provide are the fieldbuses, EtherNet/IP, ProfiNet IO, DeviceNet, Interbus-S and Profibus DP, which are a command interface and not a process data interface

Obara

Japan

Line
MFDC timers, including the adaptive SIV32; earlier SIV21 and STN21 series
Topology
Medium frequency
Adaptation
Constant current, power, voltage and firing angle modes, plus adaptive control against a reference. Compensates for material and thickness, cap wear, adhesive, poor fit-up and shunt. Waveform control for spatter reduction. Algorithm not detailed publicly
Integrity
SQA: assigns a quality percentage to each spot and detects process disturbance
Data
Extended process data logging and online quality monitoring, with local extraction over a USB interface. The public documentation does not detail which quantities are exported, nor whether there is network concentration. The industrial buses supported are a command interface

Nadex

Japan

Line
Resistance welding timers; controller, transformer and gun as a set. Installed base at Japanese-owned assembly plants
Topology
Medium frequency and alternating current
Adaptation
Insufficient public technical information
Integrity
Insufficient public technical information
Data
Insufficient public technical information

Compiled from the technical documentation published by the manufacturers themselves (18), consulted in August 2026, without independent verification. It constitutes neither a ranking nor a recommendation. Where the public information proved insufficient to describe how something works, that was flagged rather than assumed.

Two layers consolidating

The data layer and IIoT connectivity

For a long time the controller was an island. It generated extremely rich data and that data died inside the cabinet, reachable only over a proprietary cable. In a plant with hundreds of controllers that does not scale, and the result was that the information existed and nobody used it.

The way out came in three stages (18). First the network interface became standard equipment. Then came the concentrators, which gather dozens of controllers behind a single interface, and the retrofit modules, which instrument older equipment with a Rogowski coil and voltage measurement points. That second point matters more than it appears: a body shop line runs fifteen years or more, and nobody replaces an entire working installed base just to gain data. The third stage gave the data meaning, bringing together MQTT, which solves transport over limited bandwidth, and OPC UA, which describes the quantity in a way a machine understands without needing an interpreter per brand. One manufacturer documents the collection of more than 200 real values per weld spot (18), “real” here meaning measured, as opposed to programmed.

It is worth opening up what sits inside that volume, because a bare number says nothing. A weld record has two natures. On one side the curves sampled across the cycle: current and resistance in the secondary, heat and accumulated energy. On the other the scalars that summarise the cycle, and that is where much of the count lives. WTC documents, for the record produced by its measurement module, weld time in milliseconds, maximum, average and minimum secondary current, average, peak and final resistance and the drop between them, total heat, total energy, the instant of expulsion and a fault or alert code (18). Add to that the metadata tying the spot to production, such as joint identification, program used and count since the last dressing, and it becomes clear why the record runs past two hundred fields.

The gain is concrete: knowing how many welds each cap has made and how many dressings it has undergone, catching a degrading transformer diode or insufficient cooling, noticing an expulsion rate creeping up over the shift before it becomes scrap, and cross-referencing quality with part fit-up or secondary circuit degradation to find root cause (12). And, for what this article discusses further on, it is dynamic resistance and the instant of expulsion that carry the information about the joint itself.

Joint integrity assessment

This is the most recent layer, and today practically every relevant manufacturer has its own, under its own name: a function that gives each spot a score, an index or a quality percentage, in real time, for 100% of production (18).

The names change, the principle does not. It is statistics. A reference condition is defined, which may be the master weld, a band around a curve or the historical distribution of that program, and how far each new weld departs from it becomes the index. Whatever leaves the normal condition is classified as a problem.

And that is precisely where the two limits appear. The first is the false positive: leaving normality does not mean a bad weld. A new batch of sheet, a freshly dressed cap still in break-in, a different stack-up, all of it shifts the signature without producing any defect at all. The second is more serious because it is silent: there are welds that stay inside the normal distribution and still produce a small or weak nugget, and no alarm sounds. A criterion built on deviation only sees what deviates. That is why, even with this layer installed and working, ultrasound and hammer and chisel remain on the line (13).

There is a further aggravating factor, and it comes from adaptive control itself. If the controller closes the loop on the signal, what it does in the face of a disturbance is precisely to bring the curve back onto the reference. In the Axis 2 experiment this was measured: the chase worked as control, the curve met the reference, and the weld came out worse. The signature returned to normal, the joint did not.

For an integrity layer that judges by departure, this is the worst possible scenario. It flags nothing, because there is no departure left to flag. The deviation was erased from the signal by the corrective action and remained in the joint, and the compromised weld enters production with a good score. The study did not test integrity indices, so this is a logical consequence and not a measured result, but the consequence is direct: validating quality against a reference that the controller itself is chasing is measuring the controller’s work, not the weld’s.

Closing thoughts

The blind spot everyone shares

Looking at the two central columns of the table, you can see that all the systems are fundamentally doing the same thing: inferring joint quality from the process signature, compared against a reference. None of them measures nugget diameter, which is the criterion the carmaker’s standard demands.

This is not a manufacturer failing. The nugget hides between the sheets and only lets itself be measured if the joint is destroyed. But the problem runs deeper than sample coverage, and it is worth unpacking.

Diameter is already itself a proxy. What engineering needs to know about a weld spot is how much it can take, that is, mechanical strength in newtons and energy absorbed to failure in joules. What is measured in the plant is millimetres. The relationship between the two is indirect and holds as a first approximation: bigger nugget, stronger joint (13). Diameter was adopted for a practical reason, not a theoretical one. Measuring mechanical strength properly requires a test on a universal machine, with controlled loading on a specimen, and that takes minutes per sample. In production it cannot be done, neither for the time nor for the geometry, since the body does not fit into a testing machine.

The experiment cited in Axis 2 gives a direct example of this. In the edge proximity condition, nugget diameter barely changed relative to the standard weld, and even so the failure load dropped significantly (10). The cause was not nugget size, but the position of the spot: off centre, the loading stops being symmetrical and an additional torque appears that raises the shear stress and brings failure forward. A dimensional criterion approves that weld without hesitation.

That approximation has been stretched by advanced high-strength steels and by new coatings. There is evidence of nuggets with diameters within the standard that fail the destructive test without much effort. These materials brought failure modes that did not exist before and that, depending on process conditions, weaken the joint even at the “correct” diameter (15). The dimensional criterion does not see this. Inferring nugget size helps, but it does not solve it. What needs to be measured is the mechanical strength of the joint.

Closing the gap without replacing the controller

The dynamic resistance curve your controller already produces carries a great deal of information about the weld. But on its own it is not enough to say how the joint will behave mechanically. One curve is not the whole signal.

That is the space in which Spot Fusion works. It starts from the signals the process already generates and extracts from them a far broader set of features than dynamic resistance alone, and that is what makes it possible to calculate joint strength, delivering the same thing you would get by separating the sheets to check the spot. Without separating them, and for every spot. The basis for this is the training of an artificial intelligence model on more than 30,000 weld spots, generated under the most varied material and process conditions, and tested manually by destructive testing.

Spot Fusion runs as an artificial intelligence layer that moves across controllers and across the data available, taking advantage of exactly the data layer the manufacturers built over the past decade. That is why it is independent of the architecture and topology of what is installed, and nothing needs to be replaced.

Want to know how much of your welding process you can actually see today? Talk to Strokmatic engineering.

References

  1. ASM INTERNATIONAL. ASM Handbook, Volume 6: Welding, Brazing and Soldering. 1993.
  2. ZHANG, H.; SENKARA, J. Resistance Welding: Fundamentals and Applications.
  3. AMERICAN WELDING SOCIETY. AWS C1.1M/C1.1: Recommended Practices for Resistance Welding. 2019.
  4. WILLIAMS, N. T.; PARKER, J. D. Review of resistance spot welding of steel sheets. Part 1: modelling and control of weld nugget formation. 2004.
  5. DICKINSON, D. W. et al. Characterization of spot welding behavior by dynamic electrical parameter monitoring. 1980.
  6. MA, N.; WU, H. Review on techniques for on-line monitoring of resistance spot welding process. 2013.
  7. CULLEN, J. D. et al. Multisensor fusion for on-line monitoring of the quality of spot welding in the automotive industry. 2007.
  8. KONG, X. et al. Multi-sensor measurement and data fusion technology for manufacturing process monitoring: a literature review. 2020.
  9. KAS, Z.; DAS, M. Adaptive control of resistance spot welding based on a dynamic resistance model. Mathematical and Computational Applications, 2019.
  10. ZHOU, L.; XIA, Y.-J.; SHEN, Y.; HASELHUHN, A. S.; WEGNER, D. M.; LI, Y.-B.; CARLSON, B. E. Comparative study on resistance and displacement based adaptive output tracking control strategies for resistance spot welding. Journal of Manufacturing Processes, v. 63, p. 98-108, 2021. DOI 10.1016/j.jmapro.2020.03.061.
  11. SAWANISHI, C.; OKITA, Y.; MATSUDA, H.; IKEDA, R. Development of resistance spot welding technology applying multi-stage adaptive control for narrow pitch spot welding. Welding International, v. 33, n. 1-3, p. 17-29, 2019.
  12. ZHOU, K.; YAO, P. Overview of recent advances of process analysis and quality control in resistance spot welding. 2019.
  13. SUMMERVILLE, C.; COMPSTON, P.; DOOLAN, M. A comparison of resistance spot weld quality assessment techniques. Procedia Manufacturing, v. 29, p. 305-312, 2019.
  14. MATHISZIK, C.; NOPPER, B.; KOAL, J.; FÜSSEL, U.; SCHMALE, H. C. Influence of experience level on determining weld diameter in resistance spot welding. Welding in the World, v. 69, p. 483-497, 2025.
  15. POURANVARI, M.; MARASHI, S. P. H. Critical review of automotive steels spot welding: process, structure and properties. 2013.
  16. CUNHA, C. F. A.; GOMES, J. O.; CARVALHO, H. M. B. A new approach to reduce the carbon footprint in resistance spot welding by energy efficiency evaluation. The International Journal of Advanced Manufacturing Technology, 2022.
  17. SENA, F. Deep learning aplicado à inspeção de solda a ponto. Doctoral thesis, Instituto Tecnológico de Aeronáutica, 2023.
  18. Public technical documentation of the manufacturers Bosch Rexroth, Harms & Wende, Matuschek, WTC, ARO and Obara, consulted in August 2026.

Back to all articles