Measure the Whole Cycle: Assessing Output from Shipbuilding Welding Robots

Define the finished work before timing it

The most useful productivity question for ship small-assembly welding is how much accepted work leaves the manufacturing route in an agreed period. Travel speed and arc-on time remain useful process measures, but neither tells the buyer how long fitting checks, sensing, handling, inspection and ordinary recovery will take. A complete-cycle assessment connects those activities to a clearly defined finished assembly.

Start and finish events must be agreed before a comparison. One trial might begin with a fitted assembly already clamped in the cell, while another starts when the assembly arrives at the loading position. One may end when the robot stops, while another includes examination and removal. The measured numbers can be accurate and still be incomparable because they describe different boundaries.

TitanWeld's marine project uses two arms as the basic coordinated unit, with a current ceiling of 2 + 2, or four units and eight arms. It is a highly non-standard project without an off-the-shelf single machine. These facts establish the proposed equipment scale. They do not establish a speed multiple, a daily output figure or the fraction of the complete production cycle that can be performed concurrently.

Choose a denominator that represents useful output

Use an identified assembly family and a defined acceptance state. Accepted assemblies per shift may suit a repetitive family; accepted weld scope per elapsed hour may help compare varied assemblies, provided the underlying joint mix remains visible. Raw weld length can mislead when short returns, starts and stops require very different work from long straight fillets.

Define how partial assemblies, unfinished exceptions and repair are counted. If the automated cycle completes the long sides but leaves required returns for another station, the output is partially completed assembly scope until that work is done and accepted. Preserve that distinction even if a robot utilization report shows a successful cycle. The physical requirement is determined by the drawing and quality plan.

TWI distinguishes destructive and non-destructive weld testing methods as part of quality evaluation. [1] The project should identify the examinations required for its output definition rather than assuming that every motion-complete assembly is acceptable. This article proposes a measurement method; it does not prescribe a particular inspection extent or replace the applicable acceptance criteria.

Map the route and the cell boundary

Draw the actual route from incoming fitted assembly to released output. Identify loading, locating, fit-up confirmation, registration, local sensing, welding, repositioning, servicing, examination and unloading where applicable. Assign activities to the cell, an upstream station or a downstream station. This prevents labor from disappearing from the calculation simply because it occurs outside the robot enclosure.

State which activities can overlap and which cannot. An operator may prepare the next part while welding proceeds, but only if the layout, staffing and approved operating method permit it. Inspection may run in parallel at another station or may occupy the same fixture needed for the next assembly. Do not count an overlap based solely on two boxes being drawn on separate rows.

Keep the route model simple enough to validate on the shop floor. Each important event should have an observable start and finish. If an activity is difficult to time reliably, explain the estimation method and uncertainty. The resulting model should be understandable to production staff, who can often identify a missing transfer or recurring wait that a robot-only schedule overlooks.

Separate clock time from accumulated resource time

Elapsed cycle time is the time between the agreed start and finish events. Arc-on time describes how long a welding arc is active. With multiple arms, the sum of all arc-on durations can exceed elapsed time because arcs operate concurrently. Both measures are useful when labelled, but their units and meaning must remain clear.

Consider a purely illustrative calculation: two arms each weld for six minutes during a ten-minute cell interval. Combined arc-on time is twelve arm-minutes, while the interval remains ten minutes. This example is arithmetic only, not a TitanWeld result. It shows why dividing total weld length by combined arc time cannot by itself predict how many accepted assemblies will leave the cell.

Record resource occupancy where it explains a constraint. A fixture may remain occupied during cooling or examination even while no robot is welding. A shared service position may create a queue. An operator may need to attend several units. These resource times help explain the observed cycle and identify which change could improve output without assuming that faster robot travel is the dominant opportunity.

Collect events that can be reconciled to the part

Use a common assembly and weld identity across the timing record, machine events and inspection record. Fronius describes component-level welding documentation that can associate welds with a component and welding system. [2] That is an external example of a data capability, not a statement that WeldCube or an equivalent function is installed in the TitanWeld marine project.

For an acceptance trial, an appropriately designed manual event sheet may be more useful than an unconfigured dashboard. Record the assembly, configuration, start and finish events, interventions, exceptions and accepted result. If machine timestamps are used, verify what each event means and whether the clocks are aligned sufficiently for the analysis. A job-finished signal may exclude unloading or examination.

Retain the reason for waiting where practical. Categories should be few enough to apply consistently and specific enough to support action: unavailable incoming part, sensing review, shared access, service, inspection or recovery. Avoid a large miscellaneous category that absorbs the very losses the project is intended to understand. Record uncertainty instead of assigning a convenient cause retrospectively.

Use a representative mix rather than a showcase part

Select assemblies that represent the expected distribution of seam lengths, end details, access restrictions and changeovers. State how the sample was chosen and which families remain untested. A successful run of the easiest part can establish feasibility for that part, but it cannot establish a production rate for a different mix.

Include setup and changeover at the frequency relevant to the intended operation. A long batch of identical panels can spread preparation time over many pieces, while a varied daily schedule cannot. Record model preparation, fixture adjustment and job verification separately so the buyer can see whether a proposed batch pattern is realistic. Do not assume every future order arrives with a ready and validated program.

ABB's RobotStudio platform provides simulation and offline programming tools. [3] Simulation can support an estimated schedule, but its output should remain labelled as an estimate until representative measured events establish the complete-cycle result. The model's value improves when observed service and handling times are fed back into the assumptions.

Include routine interruptions and recovery

Decide which ordinary events the trial must represent: torch servicing, consumable replenishment, a fit-up rejection, a sensor review or a controlled restart. The exact list depends on the agreed configuration and production method. A short uninterrupted demonstration can omit events that occur regularly over a longer operating period, giving an incomplete view of usable output.

Keep abnormal test events separate from normal production observations while reporting both. A deliberately simulated failure tests recoverability; it should not be inserted into a production rate as if it occurred at a measured frequency. Conversely, excluding every naturally occurring interruption as an abnormal event can make the reported rate unrepresentative. Define the treatment before the trial rather than after seeing the result.

Record the work needed to restore acceptance, not just the time until motion resumes. An interrupted weld may need examination, preparation, completion and another release. If those activities move to another station, retain them in the route assessment. The buyer needs to know whether recovery consumes the bottleneck resource and whether partially completed assemblies accumulate outside the cell.

Compare alternatives using equal scope and explicit assumptions

A fair baseline should use comparable incoming condition, weld requirements and final acceptance. If the proposed method requires more precise fitting, record the added preparation. If it reduces handling or makes inspection easier, record that benefit too. The comparison should describe the complete changed route rather than attribute every difference to the number of welding arms.

Present measured results with the observed spread and the sample description. Avoid extrapolating the shortest cycle into a guaranteed shift output. State assumptions about available operating time, staffing, work availability and downstream capacity. A calculation can then be reviewed as a scenario instead of being mistaken for a production result.

When comparing two and four units, identify which activities shorten and which remain shared. Extra arms may reduce independent welding work while leaving loading or inspection unchanged. A capacity model should preserve these constraints. There is no defensible automatic rule that doubling units doubles accepted output; the result depends on the observed route and the work that can actually proceed in parallel.

Turn the result into an acceptance statement

A useful output report includes the assembly family, drawing and procedure revisions, equipment and fixture configuration, timing boundary, trial sample and acceptance status. It records unfinished scope and intervention labor, then states the measured result and any separate estimate. These details make the number reviewable without burying the reader in every controller event.

Agree how the result will be used commercially. A feasibility trial, a factory acceptance test and a production performance commitment have different conditions and responsibilities. The project documents should identify which evidence supports each statement. This article offers a measurement framework and makes no claim that a TitanWeld production commitment or independent acceptance has already been completed.

The decision becomes clearer when the report follows the actual assembly from arrival to release. Arc time then explains part of the cycle, event records explain the waits and quality records establish what output counts. A buyer can see whether an additional unit, a fixture change, better fitting or improved recovery addresses the real constraint, and can negotiate a project-specific requirement on an evidence-based boundary.

A worksheet for the complete cycle

Enter observed or estimated durations for one defined assembly. Record overlapping tasks in the notes; do not add concurrent durations as if they were sequential. Blank fields are intentional inputs, not claimed equipment performance.

Project review checklist

• Define the assembly family and the start and end of the measured cycle.
• Record loading, positioning, scanning, welding, servicing, inspection and unloading.
• Identify overlapping tasks and shared resources without double-counting time.
• Record interventions, unfinished scope and accepted output separately.
• State trial conditions, sample size and the distinction between a measurement and an estimate.