Which laser-cell metrics do the manager, operator and technologist need? ART-145

Define the decision first, then the metric

A metric is useful not because it is easy to read from a controller. It is useful when a change in its value clearly indicates who should review which decision. For every metric, record:

1. the business or production question; 2. the decision owner; 3. the data source; 4. the period and aggregation level; 5. the formula and unit; 6. permitted exclusions; 7. the action to take when it deviates; 8. its reliability limitations.

ISO 22400-2 describes manufacturing-operation KPIs through formulas, elements, time behavior, units and user groups. The important principle is that a label such as “productivity” is not a manageable KPI without a formula and an intended user. ISA-95, also known as IEC 62264, separates business and manufacturing-operation levels and provides a shared exchange model. In practice, ERP orders, shop-floor execution and machine signals should be linked, but must not be treated as the same thing.

Three management horizons

The operational horizon covers minutes and hours. The immediate need is to see that the machine is waiting for a sheet, the program has stopped or completed nests are accumulating for unloading. Data must be fresh, although it does not need to contain every financial detail.

The technological horizon covers an order, material batch, program, recurring part or week. This is where estimated and actual time, quality, cutting-condition stability, restart causes and repeated trials are compared.

The management horizon covers a week, month or order portfolio. The manager evaluates plan attainment, utilization, the structure of losses, queues, on-time completion, costs and whether the process or capacity needs to change.

The same event has different meanings. Ten minutes waiting for metal is an operational signal for the operator or logistics; if it happens every shift, it is a process problem for the manager; if it occurs only with a certain sheet format, it is a subject for the technologist to analyze.

Minimum set for the operator

The operator screen should answer “what is happening now, and what should happen next?” A useful minimum is:

| Metric or signal | Purpose | Condition for correctness | |---|---|---| | job state | show active, completed and paused work | states have unambiguous definitions | | waiting reason | distinguish material, program, gas, unloading and faults | a short reason directory with an option to clarify | | completed parts or sheets versus the job | show the remaining quantity | completion is confirmed, not merely program start | | actual cycle versus expected cycle | reveal deviations | identical cycle boundaries are compared | | repeated starts and interruptions | reveal instability | automatic and manual events are distinguished | | nonconforming parts awaiting a decision | prevent mixing them with acceptable parts | there is a simple status and physical identification | | next preparation action | reduce the gap between jobs | the queue is aligned with the plan |

MTConnect defines distinct execution states including ACTIVE, READY, INTERRUPTED, WAIT and STOPPED, as well as separate states for waiting for material or unloading. This illustrates why a single “machine on” signal is insufficient. A local dashboard does not need to copy the whole standard, but it needs an unambiguous mapping between a machine state and a clear operational reason.

An operator should not have to enter a long report when a stoppage occurs. If the list contains fifty reasons, “other” will usually be selected. Manufacturing-record examples show that free-form descriptions, inconsistent abbreviations and excessive code lists make analysis worse. A practical starting point is 8–12 top-level categories, a short clarification and later reclassification by a technologist or supervisor when necessary. The range of 8–12 is an initial L-SEL editorial recommendation, not a requirement of ISO, ISA, MTConnect or NIST: the actual number of categories should be tested against local events.

Minimum set for the technologist

The technologist needs the context of the result rather than a general total of machine hours:

• program, nest and cutting-condition version IDs;
• machine and configuration;
• material, thickness, batch and surface condition;
• estimated and actual cycle measured over the same boundaries;
• pierce count, contour length and structure, or another relevant driver;
• interruptions by category;
• first acceptable part and repeated trials;
• quality-control result;
• manual cutting-condition corrections;
• actual number of parts obtained from the sheet;
• a note about unusual geometry or requirements.

The technologist’s main question is: “why did two apparently similar jobs produce different results?” It cannot be answered without versions. If the record contains only a file name and the CAM condition was changed without a log, a cycle-time chart cannot establish a cause.

CAM forecast accuracy should also be tracked as a median and as a distribution of deviations by job family, rather than as one average percentage. ART-142 examines the plan-versus-actual method in more detail. In ART-145, this metric is only one signal for the technologist.

Minimum set for the cell manager

The manager needs a compact panel:

1. plan attainment in acceptable parts or completed orders; 2. on-time completion; 3. actual throughput for a relevant job group; 4. planned time, productive time and losses by major category; 5. the queue before the laser and the queue after it; 6. the share of recutting and nonconformities; 7. actual-time deviation from the estimate; 8. material yield and scrap cost when the data is confirmed; 9. work in progress; 10. a data-reliability indicator.

The final item is often omitted. If the reason is unknown for 35% of stoppages, an attractive distribution of the remaining 65% must not be presented as the whole picture. The dashboard should show the share of unclassified time, missing quality confirmations and orders that are not correctly linked to a program.

A manager should not evaluate an operator solely by laser “cutting time.” A high share of beam-on time may coexist with excess work in progress, delayed sorting, recutting or production of the wrong priority. A KPI should support the overall flow rather than local maximization of one machine.

Metrics shared by everyone

There is a small shared layer:

• job completion fact;
• current state and main reason for deviation;
• number of acceptable and nonconforming parts;
• program and cutting-condition version;
• time of the latest update;
• data quality.

The level of detail differs. The operator sees the current event, the technologist sees the history of a particular program, and the manager sees aggregated losses and a trend. A shared definition prevents “downtime” from meaning three different things on three screens.

Example of a role-based metric card

Suppose the cell wants to control the time from program completion until the pallet is free. Do not immediately display the same chart to every role. First create a metric card:

| Field | Working definition | |---|---| | Question | does unloading delay the next production cycle? | | Start event | confirmed `PROGRAM_COMPLETED` or a locally validated equivalent | | End event | the pallet is confirmed ready to receive the next sheet | | Unit | minutes per nest | | Definition owner | cell manager together with the technologist or analyst | | Data quality | share of nests for which both events have a valid time and ID | | Exclusions | test programs, training and other agreed categories shown separately |

For the operator, this is not a monthly KPI but a current signal: the pallet is still occupied, the reason is selected and the next action is clear. For the technologist, it is a distribution of time by nest, part count, thickness, format and program version. For the manager, it is the share of planned time lost to blocking, the queue after cutting and the effect on timely completion of sets.

If a machine program-completion event exists for every program but pallet-release confirmation exists for only some nests, the dashboard must not fill the gaps with zero. It should show the value only for covered cases and show coverage beside it. Otherwise an “improved average time” may merely mean that difficult cases are no longer being recorded.

After a two- or three-week pilot, the team checks three things: whether people understand the start and end in the same way, whether the metric leads to a specific action, and whether it causes unwanted behavior. Faster pallet release, for example, must not encourage mixing unidentified parts. If there is no decision or the metric duplicates another, revise the card instead of adding another widget.

Time is the most frequent source of errors

Before launching analytics, define the time boundaries:

• calendar time;
• shift time;
• planned production time;
• time when the job is available;
• program execution time;
• actual cutting time;
• auxiliary machine time;
• waiting for the operator, material, program or unloading;
• planned maintenance;
• unplanned stoppage.

Companies do not all need to use the same classification, but it must remain stable within one report. If lunch or lack of orders is excluded from planned time in one month and included in another, the trend loses meaning.

For a laser cell, “program active” must not be equated with producing acceptable output. The program may execute non-cutting movements, wait, restart or finish a sheet whose parts have not yet been separated and inspected.

Quality means more than counting scrap

A simple defect rate is needed, but it is of little use without categories. It is sensible to separate:

• recutting caused by the technological result;
• loss caused by a geometry or revision error;
• damage during removal or sorting;
• wrong material;
• a defect found at the next operation;
• a part held pending a decision.

The number of defects must be related to a base: parts, sheets, orders or value. “Three defects” on three large one-off parts and on ten thousand serial blanks describe very different situations.

Not every cosmetic mark is a defect, and not every cut part has been accepted. Criteria come from the drawing, technical requirement, inspection route or an agreed company standard.

Flow after cutting

A high-speed laser can move the constraint to unloading and sorting. The manager should therefore see:

• time from sheet completion to pallet release;
• number of sheets or nests waiting to be unloaded;
• time until an order set is complete;
• share of parts without unambiguous identification;
• damage or loss after cutting;
• readiness of the next operation to accept the set.

ART-147 examines manual sorting as a constraint, while ART-148 covers automatic unloading capabilities. The management point here is simple: the cell boundary is not the machine enclosure. It extends from material availability to the transfer of an acceptable set to the next operation.

Building the dashboard step by step

Stage 1 — vocabulary. Define 10–15 events and fields. Test them on one shift: do two people classify the same event in the same way?

Stage 2 — entity links. Order, program, sheet, material, machine, operator shift and result need IDs. Without them, analytics will be assembled from file names.

Stage 3 — basic data quality. Show unknown intervals and omissions. Do not replace them with assumptions.

Stage 4 — role-based screens. Operator: today and now. Technologist: comparison of contexts. Manager: flow, losses and trends.

Stage 5 — regular review. Each week, examine not only the value but whether the metric led to a decision. Remove or change a metric that nobody uses.

Antipatterns

The first is ranking operators by productive hours without considering the product mix. It encourages people to avoid difficult jobs.

The second is one OEE figure without its components or data quality. The cause disappears inside the product.

The third is comparing different machines without normalizing for the work. A machine cutting thin serial parts and a machine processing thick one-off parts perform different tasks.

The fourth is asking people to enter manually what the controller already provides reliably, while automatically “guessing” what only a person knows. It is better to combine machine time with a short human reason.

The fifth is drawing financial conclusions from machine time alone. Cost requires material, gas, labor, preparation, scrap, defects and overhead rules.

Checklist before launching KPIs

• [ ] Every metric has a question and a decision owner.
• [ ] Its formula, unit, time boundaries and exclusions are described.
• [ ] A machine state is not presented as a business result.
• [ ] The share of unknown or incomplete data is visible.
• [ ] The operator screen contains only actionable shift signals.
• [ ] The technologist can find the program, cutting-condition and material versions.
• [ ] The manager sees the queue before and after the laser.
• [ ] Quality is calculated from a confirmed result.
• [ ] People or machines are not compared without context.
• [ ] Targets are established from the company’s own baseline.
• [ ] KPIs are reviewed after a process change.
• [ ] Data is not used to punish correct reporting of a problem.

Conclusion

Begin with three different answers, not one large screen. The operator needs to know what to do now. The technologist needs to know why the result differs and which version to inspect. The manager needs to know where flow is being lost, whether the plan is being met and how reliable the data is. Once these questions are clear, 8–12 well-defined metrics will provide more value than a hundred automatically collected tags.

Safe boundaries

• Target percentages are not set without a local baseline.
• Metrics do not automatically constitute an employee assessment.
• The article does not claim that integration with ERP, CAM or a controller is ready.
• OEE is mentioned for context; its full method belongs to ART-146.

Need help selecting equipment?