Production & Operations · Engineering example
Machine Utilization Analysis
Turn machine signals into an honest view of scheduled time, setup, productive cycles, waiting, faults, and accepted output. Define the denominator and state meanings with operators so utilization leads to a better decision, not pressure to keep every machine running.
The operating challenge
“Running” does not mean making useful product.
Machine dashboards often show a single uptime or utilization percentage. That number can hide the difference between making accepted parts, cycling on scrap, waiting for an operator, preparing a new job, and sitting blocked by downstream work. It can also change dramatically when planned maintenance or unscheduled hours are included in the denominator. Without agreed definitions, two teams can calculate different values from the same machine.
States do not explain reasons
A controller may report idle or stopped while an operator performs a necessary setup, waits for material, or responds to a fault. The digital signal and the business reason are different facts.
Counts can mislead
A cycle counter may include test runs, rejects, or rework. It must be reconciled with the job and quality record before it represents good output.
High utilization is not the goal
Running a nonconstraint machine to improve a score can build excess inventory. Capacity, delivery, quality, and cost need to be considered together.
Why modernize: Give operators, maintenance, and planners a shared explanation for lost time and available capacity. The purpose is to improve scheduling and work preparation, not to rank people by a number that lacks context.
Illustrative operating example
A CNC cell appears idle while orders are late.
This machining-cell situation is illustrative. It is not a claim about a completed customer installation, measured throughput gain, or employee performance.
Imagine two CNC machines making several short-run components. One screen reports that Machine M-2 is active for only part of its scheduled shift, while a planner sees late orders. It would be easy to request another shift or blame machine downtime. A closer review shows that a fixture change and first-article approval consume much of the day, some jobs wait for released material, and a small number of completed cycles are rejected at inspection. A utilization percentage cannot by itself tell whether the next useful action is better kitting, setup preparation, maintenance, or more capacity.
- Agree on the schedule boundary.Define scheduled production, planned maintenance, staffing, and agreed exclusions. Show both elapsed and eligible time rather than quietly changing the denominator.
- Interpret controller states.Map power, mode, execution, program, alarm, and part-count signals to a provisional timeline. Mark unavailable or contradictory data explicitly.
- Add the human reason.Operators and planners record setup, first-article wait, material hold, tool issue, blocked output, and planned stop using concise, reviewable codes.
- Reconcile accepted output.Join job, cycle, inspection, scrap, and rework records so a completed program is not automatically counted as a good part.
- Choose one test.The team may stage fixtures or material before the shift, revise release timing, or investigate a repeat fault. Compare the next similar job before changing staffing or buying equipment.
What the reviewer would see
Machine M-2: scheduled-time review
The record explains the state mix without evaluating an individual operator.
- Scheduled window
- Defined shift and planned breaks or maintenance, with the denominator shown.
- Observed states
- Active cycles, setup, waiting, faults, and unknown time with source and timestamp quality.
- Output context
- Job family, accepted units, rejects, rework, and first-article approval.
- Decision
- Operations chooses a bounded preparation or process trial and records the result.
Other high-priority situations
Different time losses need different owners.
Repeated microstops accumulate
Short pauses may be ignored because each is small. A time-ordered view can group them by product, tool, alarm, and shift. Maintenance and operators verify whether the cause is equipment, material, or an intentional step.
Useful report: frequency, total time, context, verified reason, and corrective action.
Setup time varies by product sequence
Compare changeovers for similar tool and fixture transitions, not all setups as if identical. A scheduling or preparation change may free capacity without pushing the equipment harder.
Useful report: previous and next job, setup phases, first-good-part time, and approved sequence trial.
What a modernization would connect
Build a state model that operators can correct.
The machine can supply objective events, but the business meaning of an event depends on job, schedule, and quality context. A good first build maps signals into a versioned state taxonomy, exposes unknown time, and lets an authorized reviewer add a reason with an audit trail.
Required operating evidence
- Machine and controller identity, mode, execution, program, alarms, and power
- Job, product, route, and schedule with effective versions
- Cycle starts and finishes, part count, accepted output, scrap, and rework
- Setup, planned maintenance, material, quality, and staffing context
- Clock source, data freshness, manual reason, and correction history
- Documented definitions for run, setup, idle, blocked, starved, fault, and unknown
Controller execution is not a universal definition of productive machining; validate it against the actual process.
Connected technical layers
- Approved read-only machine interface or standard adapter where appropriate
- Production schedule or MES for jobs and planned windows
- Quality and maintenance systems for accepted output and findings
- Operator review screen and versioned reporting logic
Machine safety and control remain local. The analysis does not send commands to improve a percentage.
Publish the denominator beside every percentage. If unknown or unclassified time is large, show it as a data-quality problem rather than assigning it to idle or downtime.
Where automation and AI fit
Automate the timeline; let people explain the operation.
Deterministic automation
Normalize events, calculate durations under the approved taxonomy, mark gaps and overlaps, and reconcile cycle counts with accepted output. Route a sustained unknown state for review instead of labeling it as a fault.
Analytical methods
Compare state patterns across similar jobs, shifts, and machine configurations. A model may highlight an unusual microstop cluster or setup variation, but it should be tested against transparent summaries and operator explanations.
Evidence-linked AI assistance
An approved assistant can summarize why utilization changed, cite schedule and state records, and ask whether material, quality, or maintenance context is missing. It must not infer worker performance from equipment time.
Human operating authority
Operations, maintenance, and quality approve process changes, schedule adjustments, maintenance work, and any use of the metric in management decisions.
How we examine a proposed solution
AI advisors challenge definitions before dashboards scale.
These Adaptiv Stratum AI advisory roles can support bounded research and design review. They are not human shift supervisors or a claim that agents manage the machine.
Argus
Tests machine identity, state timestamps, and signal continuity.
Hephaestus
Checks what controller modes mean for this equipment and process.
Pythia
Questions whether anomaly methods add value beyond an honest state timeline.
Aegis
Examines gaps, contradictory states, and misleading conclusions.
Clio
Checks comparable-job trials and claimed improvement evidence.
Themis
Reviews metric definitions, decision rights, and inappropriate use of worker data.
What becomes visible
Show the time account before the percentage.
A shift report should make every scheduled minute explainable. A longer-term report should test whether better preparation or maintenance improved accepted output and delivery, not only machine activity.
Per-machine shift view
- Schedule and denominator, with approved exclusions
- Time in active cycle, setup, planned stop, blocked, starved, fault, and unknown
- Job and product context for each interval
- Cycle count, accepted output, reject and rework counts
- Manual reasons, data gaps, and reviewer corrections
Capacity and improvement view
- Utilization under the same definition and comparable job mix
- Accepted output, delivery, and work in process
- Setup and first-good-part time by transition type
- Microstop frequency and verified cause
- Unknown-time share and reason-code completeness
- Maintenance interventions and post-work performance
A higher utilization rate can coexist with more scrap, late orders, or excess inventory. Decisions should weigh the whole operation, not reward running for its own sake.
Operating safeguards
Keep definitions stable and operators in the loop.
- Published denominator and state definitions with version history
- Clock alignment, freshness, overlap, and missing-signal checks
- Operator-reviewed reasons without overwriting original events
- Accepted-output reconciliation instead of raw cycle counts alone
- Separate planned maintenance and unscheduled fault time
- Compare similar jobs and product mix before conclusions
- No machine command or employee ranking from an AI-generated score
- Review the broader effects on quality, inventory, and delivery
A practical first phase
Start with two machines and one scheduling question.
- Agree on terms.Operators, planning, and maintenance define scheduled time, state categories, accepted output, and the decisions a report should support.
- Observe real work.Walk setup, cutting, inspection, waiting, and planned stops while mapping controller signals to the physical sequence.
- Build a read-only timeline.Collect events with job context, expose unknown intervals, and let operators verify reason codes.
- Test one preparation change.Stage a fixture or material for a comparable job and document the effect on first-good-part time, output, and downstream flow.
- Decide whether to expand.Keep the method only if it gives trusted, actionable explanations without distorting safety, quality, or worker treatment.
Begin with a machine question
Which “idle” hours are not actually idle?
A pair of machines, recent schedules, and operator-verified state reasons can reveal a useful first improvement.
Evidence and further reading
Define machine states before measuring improvement.
These primary sources inform the method. The CNC scenario is illustrative, not an Adaptiv Stratum customer case study.
- NIST: Automating asset knowledge with MTConnect discusses machine data and the relationship to equipment performance measures.
- NIST MEP: Hydraulic press OEE system provides a real example of combining availability, process performance, and quality rather than a single run signal.
- MTConnect describes an open approach to machine-data access; a specific machine interface and semantic mapping must still be verified.