How to prepare a laser-cutting department for an unattended second shift ART-149

What “unattended” actually means

Different companies use this term for very different modes. One machine may complete two proven sheets after the evening shift ends. Another department may run overnight with a tower store, automatic loading and unloading, and a queue of dozens of jobs. A third option is a low-manned shift in which one responsible employee monitors several systems and intervenes only after an alert.

The operating boundary must therefore be defined in writing first:

• how many hours the department must run, and without which exact personnel being present;
• which materials, thicknesses, formats and parts are permitted;
• whether automatic transitions between jobs are allowed;
• which deviations the system may handle on its own;
• which events must stop production;
• who responds to a notification and within what time;
• what counts as a successful shift.

Without this description, a controlled scenario cannot be distinguished from a risky experiment. The words “autonomous machine” in a supplier’s offer do not prove that a particular department is ready with its own materials, network, extraction system and logistics.

Five readiness areas

It is more useful to assess five independent areas than to count installed options.

| Area | Main question | Evidence of readiness | |---|---|---| | Jobs | Can the queue run without manual reprogramming? | Proven programs, priorities, prohibited combinations and a stop rule | | Material | Will the machine receive the correct and usable sheet? | Material identification and control of stock, orientation and sheet condition | | Process | Are the conditions normally noticed by an operator controlled? | Documented functions of the actual configuration and verified responses | | Deviations | Does the system enter a safe state? | Event–action–notification–responsible-person matrix | | Acceptance | Can the actual result be reconstructed in the morning? | Log of jobs, alarms, unfinished sheets and accepted parts |

A strong result in one area does not compensate for failure in another. Reliable nozzle monitoring, for example, cannot correct incorrectly loaded material, and a remote notification is of little value if nobody has an agreed response.

Start with a stable daytime process

The night shift should not be the first place where production tests a new material, difficult geometry or a changed cutting mode. A history of repeatable daytime runs is required first. For each job family, the team should know the frequency and causes of stoppages, the causes of rejects, the need for manual intervention and the actual cycle time.

A suitable candidate is a job set in which:

• material and thickness are identified unambiguously;
• the program has a controlled version;
• the process has already been reproduced on this machine;
• parts do not create an untested risk of tipping, catching or accumulating;
• the sheet quantity matches the available stock and automation capacity;
• completion does not require an immediate manual operation to preserve quality.

Complex or unstable jobs remain on a staffed shift until they have their own evidence of repeatability. This is not excessive caution. It prevents process development from being mixed with an autonomy trial.

The job queue must be executable, not merely long

Night plans are often filled with as many sheets as possible. A long queue is useful only when the order of materials, cassette availability, capacity for finished parts and skeletons, and prohibited transitions have all been considered.

Before release, the queue is modelled as a route. For every job, the material, sheet, program, expected duration, required consumables, output location and completion condition are known. If the unloading area becomes full after the third sheet, or the required material type runs out, the next ten programs do not provide a production reserve.

It is sensible to define the “last guaranteed safe job.” If the system cannot confirm the next sheet or an available output location, it must not improvise. It should complete the current controlled cycle and stop according to the manufacturer’s rules.

Material needs its own admission check

For a person, the difference between two stacks may be obvious from markings or context. An automatic system can use only the attributes that are actually available to its sensors and data. The physical sheet must therefore be reconciled with the digital job before the night shift.

Check:

• grade, thickness, format and batch of the material;
• the quantity of usable sheets in the required positions;
• absence of mixed or unidentified remnants;
• flatness, contamination and other limits specified by the manufacturer;
• permissible sheet and stack weight for the actual equipment;
• correct cassette or storage position;
• the route for skeletons, remnants and finished parts.

Do not assume that the system will “recognise an error by itself.” The team must know which properties the machine checks and which ones it simply receives from the production job.

Autonomy is determined by the response to deviations

Manufacturers offer automatic nozzle changing and checking, sheet-position detection, process monitoring and alarm collection. These are important functions, but their presence and limits cannot be assumed across models. Verify the configuration, software version, licences and documented capability of the actual machine.

Every critical deviation needs one of three confirmed responses:

1. The system corrects the condition within a documented limit and continues. 2. The system skips or completes a defined step without spreading the risk. 3. The system stops safely, preserves the data and notifies the responsible person.

The dangerous fourth response is “unknown.” Without a verification protocol, an event cannot be treated as automatically controlled merely because a sensor is installed.

Event and escalation matrix

The matrix is not intended to produce a large instruction manual. Its purpose is to ensure that the response to a signal does not depend on whichever person happens to receive it.

| Event | Automatic action | Notification | Remote action | Site visit required | |---|---|---|---|---| | Job completed | Continue through the approved queue | If required | None | No | | Next sheet not confirmed | Stop the transition | Shift owner | Status viewing only | According to procedure | | Process deviation | Only the manufacturer-documented response | Immediately | No bypassing safeguards | Defined by instruction | | Output area full | Controlled stop | Responsible person | Do not restart without checking | Yes | | Connection lost | Local safe logic | After the channel is restored | None | Based on risk |

Exact actions come from the manufacturer’s documentation and the company’s risk assessment. Remote access must not become a way to clear alarms, command motion or bypass interlocks without inspecting the site.

The infrastructure must operate for the entire shift

The laser is not an isolated consumer. Cooling, compressed air, process-gas supply, extraction, fire and production safety systems, network communications and material automation must remain stable with it. During a staffed shift, an operator may notice an indirect symptom such as an unusual fan noise, a pressure drop or an overflowing container. In unattended operation, critical conditions need measurable control and a defined response.

The check is not limited to whether an auxiliary system is listed in the specification. It must be capable of operating for the full planned period with the necessary reserve. Total electrical capacity is discussed in ART-152, power quality in ART-153, and interruptions and backup in ART-154.

Remote observation is not remote control

A useful remote screen shows job status, time, alarms and stoppages. It helps a responsible person decide whether action is required. A camera or web panel, however, does not provide complete information about the physical space around the machine.

Permissions should therefore be separated into:

• viewing status and logs;
• receiving notifications;
• confirming an organisational decision;
• commands that affect motion or the cutting process.

The last category is permitted only within the limits explicitly allowed by the manufacturer, the local risk assessment and the access-control system. The convenience of a smartphone does not change the requirements for a safe equipment state.

Expand the pilot in steps

A dependable launch does not begin with a full night. A useful sequence is:

1. One known sheet under operator supervision, recording every intervention. 2. Several homogeneous sheets during a staffed shift, but without planned intervention. 3. A short period without an operator at the machine while the responsible team remains on site. 4. A limited after-shift window with one material type. 5. Expansion of the product range only after actual stoppages have been analysed.

Set a trial limit and a rollback criterion at every step. If the number of unexplained stoppages increases, the answer is not another attempt with the same queue. The causes must be investigated first.

### Criterion for moving to the next step

The decision to continue or stop trials is based on an agreed body of evidence, not one successful run. For each pilot window, retain the queue identifier, program versions, list of actual sheets, event log, all manual interventions and the morning acceptance result. Record separately any event the system could not classify unambiguously. Even if it caused no reject, it blocks expansion until the team determines its cause and a safe response.

Progress is allowed only when completed jobs have been accepted, critical stoppages can be reconstructed from the log and actual interventions remain within the set limit. Incomplete evidence is not a conditional pass: the previous permitted boundary remains in force, and the trial is repeated after corrective action. The protocol is signed by those responsible for production, process engineering and safety. A change of material, program, automation or escalation rules triggers a new assessment of the affected area.

How to measure success

Hours of operation alone are not enough. A machine can spend a long time waiting or produce parts that later fail acceptance. The pilot should report:

• planned and actually completed jobs;
• productive time and controlled waiting time;
• number of events handled automatically;
• number of remote and physical interventions;
• accepted parts, rejects and unfinished production;
• cause of every stoppage;
• time to detection and response;
• morning time spent on analysis and recovery.

“The night ran” has value only when accompanied by quality, traceability and the absence of hidden morning work.

Positive list and exception register

Controlled autonomy is built with a positive list: only explicitly permitted work enters the night queue. A record should define not only material and thickness, but also part family, program revision, permissible sheet format, output-handling method and the validity period of the approval. If one attribute does not match, the job returns to a staffed shift.

Maintain a separate exception register. It explains why a particular geometry, material batch or combination of operations is temporarily excluded. The reason may be technological, logistical or informational. Once the cause is removed, the record is not simply deleted. A controlled daytime run is completed and the decision to expand the positive list is documented.

This approach is better than a long list of prohibitions. The night team should not have to interpret whether a new part resembles an old exception. Approval must be unambiguous and tied to a data version.

Check the length of the complete continuous chain

Potential cutting hours are limited by the shortest link. A tower may hold eight hours of sheets, while the skeleton container fills in three hours, nozzle life has only been confirmed for four, and finished parts block the destination after the fifth job. Planning eight hours for such a system is incorrect.

For every resource, define:

• capacity available at the start of the shift;
• predicted consumption or filling by job;
• a warning that the limit is approaching;
• safe action when the limit is reached;
• a method of morning confirmation.

The chain includes more than metal and storage space. Check gas, compressed air, cooling, filters, waste containers, marking supplies, network data storage and every consumable the particular configuration needs to run without intervention.

Managing changes in data

After the evening check, the queue must not change unnoticed because of a new file, automatic import or manual edit. Record a fixed state: the job list, program versions, materials, priorities and a checksum for the package. Any late change returns the job for repeat verification by the responsible person.

Define behaviour for loss of access to ERP, MES, a network directory or a cloud service as well. An autonomous machine must not receive incomplete data or continue with an unknown version. The permissible local buffer, synchronisation rules and limits of operation without an external system are confirmed before the pilot.

Morning acceptance is part of the process

The first shift should not simply collect the parts and press Continue. It reconciles the actual job list, machine condition, alarms, material balance, part quantity and quality, remnants and unfinished operations. An unexplained event receives its own record rather than a generic note such as “night error.”

If the result cannot be reconstructed from the log, the autonomous shift is not controlled. Morning analysis provides the evidence used to expand the permitted operating boundary gradually.

Common mistakes

Buying a loader and declaring the department unattended. Material handling does not cover the cutting process, infrastructure, exceptions or acceptance.

Sending the most difficult parts to the night shift. The high value of daytime capacity does not make an unstable process suitable for autonomous operation.

Relying on notifications without escalation. A signal is useful only when it is clear who receives it and what that person may do.

Allowing remote restart by habit. Some events require physical inspection. The manufacturer and the risk assessment define the exact boundary.

Ignoring morning losses. Sorting, repeat inspection and recovery after a night stoppage can consume all the additional cutting hours.

Decision checklist

• Is the exact unattended mode and permitted time window defined?
• Is there a positive list of materials and jobs?
• Does every program have a verified version?
• Is the physical material linked to the digital queue?
• Do loading and output capacities match the plan?
• Do critical conditions have a documented automatic response?
• Is it known which functions are actually present in the configuration?
• Is there an event, notification and escalation matrix?
• Are auxiliary systems confirmed for the full period?
• Do remote permissions prevent unsafe bypasses?
• Has a step-by-step pilot and intervention review been completed?
• Are morning acceptance and expansion criteria defined?

Conclusion

A department ready for an unattended second shift is not a machine with the maximum number of automatic options. It is a controlled production system. It knows what to cut, from which material, in what order, how to recognise an acceptable deviation, when to stop and how to hand the result to the next shift.

Start with a narrow positive list and a short pilot. Expand autonomy only on the basis of the actual event log, not expectations created by an equipment label. This turns additional hours into accepted parts rather than an overnight accumulation of unknown risks.

Need to assess the readiness of a laser-cutting department?