Digital & AI

Smart Factory & IoT

Digital shop-floor, MES, sensors and digital twin.

A mature smart-factory programme connects machine-level sensors, MES transaction data and manual line-balancing records into a single OEE (overall equipment effectiveness) view, so planners can see availability, performance and quality losses per line, per shift, in near real time rather than reconstructing them from end-of-day paper reports. The technical core is data plumbing: normalising signals from heterogeneous machine ages (a 15-year-old machine bolted with a retrofit IoT sensor sits next to a new networked machine with native connectivity) into a common schema the MES and dashboards can consume, and separating true machine downtime from operator or material starvation.

The harder, ongoing work is organisational: OEE dashboards only change outcomes if supervisors act on them within the shift, and predictive maintenance models only earn trust once they have caught a handful of real failures without excessive false alarms. Digital twin initiatives for line layout or capacity simulation are typically still pilot-stage for most apparel factories, useful for testing 'what-if' line balance changes before physical changeover, but not yet a substitute for physical trial runs when introducing genuinely new styles or machine classes.

How the work is done

1
Machine and workstation instrumentation
Fit IoT sensors or retrofit counters to sewing machines, cutters and pressing equipment to capture cycle counts, stitch counts, stoppage events and idle time.
2
Data integration into MES
Feed machine-level data alongside manual bundle-tracking or RFID scans into the MES so work-in-progress location and stage timing are visible line-by-line.
3
OEE and downtime categorisation
Classify downtime causes (machine fault, material shortage, operator absence, changeover) so the loss driver is identifiable, not just the aggregate OEE number.
4
Dashboard rollout to supervisors
Push shift-level OEE, defect rate and line-balance dashboards to floor supervisors with alert thresholds for immediate intervention, not just end-of-day review.
5
Predictive maintenance model build-up
Accumulate sensor history (vibration, temperature, cycle time drift) on critical machines and train alert thresholds, starting with rule-based alerts before statistical/ML models once enough failure history exists.
6
Digital twin / simulation pilots
Where mature enough, model line layout and capacity changes virtually before physical changeover, validating against a small physical pilot before full rollout.

Decisions you have to make

Retrofit sensors on existing machines versus phased replacement with networked machines?: Retrofitting is cheaper and faster to deploy across a large existing fleet, but native machine connectivity gives richer, more reliable data; blend both based on machine age and remaining useful life.
How much OEE granularity to expose to line supervisors versus management only?: Supervisors need real-time, actionable line-level data to intervene within the shift; aggregated cross-factory dashboards are more useful to management for trend and investment decisions.
Rule-based alerting versus ML-based predictive maintenance?: Start with rule-based thresholds (temperature, vibration limits) since they work with limited history; move to ML models only once enough labelled failure events exist to train and validate them credibly.
Build MES/IoT capability in-house versus buy a vendor platform?: Vendor platforms speed deployment and reduce integration risk but can lock in data formats; in-house builds give control but require sustained software engineering capacity most factories underestimate.
How aggressively to chase 100% sensor coverage versus targeting critical bottleneck machines first?: Prioritise bottleneck and high-value machines for instrumentation first to prove ROI before scaling coverage factory-wide; blanket instrumentation without a use case wastes capex.

Key metrics (indicative)

Overall Equipment Effectiveness (OEE)

track against baseline, indicative improvement trajectory

Combines availability, performance and quality loss into one trend indicator for line health.

Unplanned downtime hours per machine per month

track against baseline, trend downward

Rising unplanned downtime signals maintenance model or sensor coverage gaps.

Predictive maintenance alert precision

indicative working range, track against baseline

High false-alarm rate erodes supervisor trust and causes alerts to be ignored.

Time from downtime event to supervisor action

track against baseline, minutes not hours

Dashboards only add value if they shorten reaction time on the floor.

Line re-balance cycle time using digital twin/simulation vs physical trial

track against baseline

Confirms the simulation tool is actually saving physical trial time, not just adding a parallel step.

Metric targets are indicative working ranges, not standards or legal limits.

Common pitfalls

Instrumenting every machine at once without a clear use case, generating data nobody analyses and burning budget on low-value sensors.
Deploying OEE dashboards to management only, so floor supervisors never see actionable data in time to intervene within the shift.
Building predictive maintenance models before enough failure history exists, producing unreliable or noisy predictions that get ignored.
Treating digital twin simulation output as production-ready fact, skipping physical pilot validation and hitting real-world line-balance surprises.
Mixing manual and automated downtime logging inconsistently, so OEE figures are not comparable across lines or shifts.

Advanced notes and limits

Digital twin modelling for full-factory capacity planning is still largely pilot-stage in apparel manufacturing; it is more mature for line-segment simulation than whole-factory optimisation.
Retrofitted IoT sensors on old machines can degrade data quality over time (drift, connectivity loss) faster than native sensors, requiring an ongoing calibration and maintenance budget, not a one-off install.
OEE benchmarks vary enormously by product complexity and line type, so cross-factory or cross-style OEE comparisons without normalising for style mix are misleading.
Predictive maintenance ROI is concentrated on high-cost, high-downtime-impact machines (e.g. cutters, automated spreaders); applying it uniformly to low-cost manual machines rarely pays back the sensor and analysis cost.

Worked example

Calculating OEE for a sewing line shift and isolating the biggest loss

Planned shift time: 480 minutes
Unplanned downtime (machine faults + material shortage): 60 minutes
Ideal cycle time per garment: 1.2 minutes
Total units produced: 280 units
Good units (passed quality): 252 units

1Availability = (Planned time - Downtime) / Planned time = (480 - 60) / 480 = 420/480 = 0.875 (87.5%).
2Performance = (Ideal cycle time x Units produced) / Run time = (1.2 x 280) / 420 = 336/420 = 0.80 (80%).
3Quality = Good units / Total units = 252/280 = 0.90 (90%).
4OEE = Availability x Performance x Quality = 0.875 x 0.80 x 0.90 = 0.63 (63%).
5Loss breakdown: Availability loss = 12.5%, Performance loss = 20%, Quality loss = 10% — performance is the largest single loss driver.
6Direct supervisor attention to the performance gap (likely line imbalance or minor stoppages not logged as downtime) rather than machine reliability first.

The line's OEE for the shift is 63%, and because performance loss (20%) is larger than availability loss (12.5%) or quality loss (10%), the technologist should investigate line balance and micro-stoppages before spending on further machine reliability fixes.

Case study

Context

A knitwear factory retrofitted IoT stitch counters onto 30 sewing machines across two lines and built an OEE dashboard, but after three months the dashboard was only reviewed by the factory manager once a week during a planning meeting.

Problem

Line supervisors kept managing by walking the floor and eyeballing bundle backlogs, so downtime events sat unaddressed for hours before anyone acted, and the OEE trend line showed no improvement despite the sensor investment.

Action

The technologist reconfigured the dashboard to push shift-level alerts directly to supervisor tablets when a machine idled beyond a set threshold, and retrained supervisors to treat the alert as an action trigger rather than a report to read later.

Outcome

Median time from downtime event to supervisor intervention fell from over two hours to under 20 minutes within six weeks, and OEE on the instrumented lines rose measurably as availability loss shrank, confirming that the data plumbing alone had not been the bottleneck — the organisational response loop was.

Audit checklist

Are OEE figures broken into availability, performance and quality loss, not reported only as one aggregate number?
Do line supervisors receive shift-level alerts in near real time, not just end-of-day or weekly management reports?
Is downtime categorised by cause (machine fault, material shortage, operator absence, changeover) so root cause is identifiable?
Are retrofitted sensors on older machines being recalibrated periodically, not treated as a one-off install?
Is predictive maintenance limited to high-cost, high-downtime-impact machines rather than applied uniformly to low-value manual equipment?
Are rule-based alert thresholds in place before any ML-based predictive model is trusted for action?
Is digital twin/simulation output validated against a physical pilot before full-scale line rebalancing?
Are OEE comparisons across lines or factories normalised for product/style complexity before being used for benchmarking decisions?

Glossary

OEE (Overall Equipment Effectiveness): A composite metric = Availability x Performance x Quality, used to quantify how effectively a machine or line is used against its theoretical maximum output.
Availability loss: Production time lost to unplanned or planned stoppages (machine faults, material shortage, changeovers) as a share of planned production time.
Performance loss: Output lost when a line or machine runs slower than its ideal cycle time even while running, often caused by line imbalance or minor uncounted stoppages.
MES (Manufacturing Execution System): Software that tracks work-in-progress, machine and labour transactions on the shop floor in real time, sitting between ERP planning systems and the physical production line.
Retrofit IoT sensor: A sensor added to an existing (often older) machine to capture cycle counts, vibration or temperature data without replacing the machine itself.
Predictive maintenance: Using sensor history (vibration, temperature, cycle-time drift) to forecast likely machine failure before it happens, rather than servicing on a fixed schedule or after breakdown.
Digital twin (factory): A virtual model of a production line or factory used to simulate layout or capacity changes before physical implementation; in apparel this is more mature at line-segment scale than whole-factory scale.
Alert precision: The proportion of predictive maintenance or downtime alerts that correspond to a real, actionable issue, as opposed to false alarms that erode operator trust in the system.
Line balancing: Distributing operations across workstations/operators so cycle times are as even as possible, minimizing idle time and bottlenecks along a sewing line.
Bottleneck machine: The workstation or machine with the longest cycle time on a line, which sets the maximum output rate for the whole line regardless of other stations' speed.

Practice questions

1. A line has 450 minutes planned time, 50 minutes downtime, ideal cycle time 0.9 min/unit, 350 units produced, 315 good units. Calculate OEE and identify the largest loss category.
2. Why is rule-based alerting recommended before ML-based predictive maintenance on a newly instrumented machine fleet?
3. A factory instruments all 200 machines simultaneously with no prioritisation. What is the likely risk, and what should have been done instead?
4. Why can cross-factory OEE comparisons be misleading without adjustment?
5. A predictive maintenance system on a manual button-sewing machine generates frequent alerts that turn out to be false positives. What should the technologist do?
6. What organisational condition must be true for an OEE dashboard rollout to actually improve production outcomes?

Sub-topics in this chapter

MES: Manufacturing Execution Systems that direct, track and report production in real time.
IoT machine monitoring: Retrofit or built-in sensors that stream machine state, cycle time and stop reasons.
Digital shop-floor: Screens and tablets replacing paper travellers with live data at every workstation.
OEE dashboards: Overall Equipment Effectiveness = availability × performance × quality, tracked per line.
Predictive maintenance: Sensor + ML models that predict machine failures before they happen to reduce downtime.
Digital twin: Live virtual model of a line or factory used for what-if analysis and optimisation.

Lessons that teach this chapter

Where this chapter is applied

The value chain stages that use this chapter's skills — chapter to stage to skill.

Study tools for this chapter

Saved on this device only

Study without an account, or sign in to save progress across devices. See the full chapter syllabus.

All chapters and the 42-stage course