Human-Supervised CPS-Based Optimization of Insulation Material Production: An Industrial Case Study

Featured Application

The proposed human-supervised CPS-oriented framework is directly applicable to insulation-material plants and comparable process industries in which bottlenecks in loading, setup, internal logistics, reliability, and energy use can be addressed through cyber-supported monitoring, feedback-informed decision-making, and selective automation.

Abstract

Insulation-material manufacturers face increasing pressure to improve productivity, cost efficiency, energy performance and worker safety while maintaining stable quality in highly constrained production environments. Existing lean and smart-manufacturing studies often examine isolated tools, individual monitoring technologies or material-level sustainability, but fewer studies provide conservative plant-level validation of an integrated intervention in insulation-material production. This study therefore examines the optimization of insulation-material production in a human-supervised cyber–physical manufacturing system through an industrial before–after intervention. The framework combines bottleneck identification, value stream mapping, SMED, selective automation, preventive maintenance and KPI-based digital monitoring. The baseline system was constrained by manual crusher loading, long changeovers, inefficient pallet transport, repeated breakdowns, scrap and limited real-time visibility. After implementation, productivity increased from 7864 to 9000 kg/day (+14.5%), monthly production costs decreased from EUR 200,000 to EUR 180,000 (−10%), breakdown frequency fell from 5 to 3 events/month (−40%), scrap decreased from 5% to 3% (−40%), crusher loading time fell from 30 to 10 min/pallet (−66%), annual energy use dropped from 500 to 450 MWh (−10%) and reported safety incidents decreased to zero during the 12-month post-implementation observation period. An OEE-based surrogate model yielded pre- and post-state theoretical capacity estimates differing by less than 1%, supporting internal consistency. The results are interpreted as descriptive and practically meaningful before–after differences because the full raw monthly dataset is commercially sensitive and classical inferential testing was not performed. The study contributes by presenting a reproducible, conservative and human-supervised CPS-oriented plant-intervention protocol rather than by claiming a fully autonomous closed-loop CPS.

1. Introduction

Cyber–physical systems (CPSs) in manufacturing connect physical production processes with digital data acquisition, monitoring, analysis, and feedback mechanisms, enabling improved process visibility, faster response to deviations, and more informed operational decision-making [1,2,3,4,5]. In this study, the term CPS is used in the manufacturing sense of a cyber–physical manufacturing system (CPMS), in which shop-floor operations, material flow, maintenance activities, and performance indicators are linked to a cyber layer that supports monitoring and feedback-based optimization. Following the NIST view that CPSs involve interacting digital, physical, and human components engineered through integrated logic and physics [1,2], the present work deliberately positions the implemented system as a human-supervised CPS rather than as a fully autonomous closed-loop controller.

This distinction is important because insulation-material production is shaped by interacting with physical and organizational constraints. Operational performance depends not only on machine efficiency and throughput, but also on reliable material handling, changeover stability, internal logistics, energy consumption, defect reduction and worker safety. Manual loading, inefficient transport, frequent stoppages and delayed reaction to deviations can significantly reduce competitiveness and increase production costs. At the same time, insulation manufacturing is increasingly exposed to sustainability and resource-efficiency expectations [6,7,8,9,10,11,12,13,14,15,16,17,18], as well as to smart-manufacturing, digital-monitoring and Industry 4.0 expectations [19,20,21,22,23,24,25,26,27,28,29,30,31,32].

The baseline production system examined in this study was constrained by manual crusher loading, long setup and changeover activities, inefficient pallet transport, frequent breakdowns, material scrap and limited performance visibility. These constraints jointly reduced productivity, increased production costs, and weakened process reliability. The case therefore represented a suitable industrial setting for examining how a conservative CPS-oriented improvement framework could support practical optimization in a real insulation-material manufacturing environment.

Figure 1 summarizes the logic of the industrial case by linking baseline bottlenecks, physical interventions, cyber-layer support and validated plant-level outcomes before the detailed methodological description.

Four literature streams are relevant. First, lean manufacturing and sustainability studies show that lean, Six Sigma, and sustainability are increasingly treated as integrated systems rather than as independent managerial approaches [6,7,8,9,25,26,33]. Second, SMED and setup-reduction research confirms that changeover improvement is most effective when combined with standardization, organizational clarity, and structured implementation [34,35,36,37]. Third, digital monitoring, predictive maintenance, digital-twin and IIoT research position manufacturing as a data-rich environment in which sensing, feedback, diagnosis, scheduling, and intervention are increasingly coupled [19,20,21,22,23,24,27,28,29,30,31,32]. Fourth, insulation-material and mineral wool studies address circularity, reuse and recycling, but often focus on material pathways rather than factory-level operational optimization [10,11,12,13,14,15,16,17,18].

The artificial intelligence literature also indicates a pathway for future development. Recent work on super-resolution reconstruction, graph-based transfer learning for rotating machinery fault diagnosis, fire-door defect inspection, and semi-supervised AHU fault detection illustrates how AI can support advanced diagnosis, inspection, and predictive maintenance [38,39,40,41]. However, these AI approaches were not implemented in the present plant. They are cited to position future extensions and to clarify that the present contribution is an industrial CPS-oriented monitoring and improvement framework, not an AI diagnostic algorithm.

Despite the breadth of the literature, four gaps remain relevant. First, many lean-sustainability studies are conceptual, review-based, or broad framework integrations rather than tightly bounded industrial interventions with explicit before-and-after KPI validation. Second, many smart-manufacturing studies validate only one dominant mechanism at a time, such as SMED, bottleneck detection, energy monitoring, digital-twin monitoring, or predictive maintenance, rather than a single plant-level intervention that combines lean methods, selective automation, maintenance, and cyber-supported KPI monitoring. Third, the insulation-material literature is stronger in circularity and product-system sustainability than in factory-level production optimization. Fourth, few studies report a unified KPI set spanning productivity, cost, reliability, scrap, energy, and safety in insulation-material production within a CPS-oriented environment.

The novelty of this work is therefore not the proposal of a new autonomous CPS algorithm. Rather, it lies in the conservative industrial validation of a human-supervised CPS-oriented diagnosis–intervene–monitor–feedback framework in a real insulation-material production line, using a coherent KPI architecture and an OEE-based consistency check.

The study addresses the following research question: How, and to what extent, can a human-supervised CPS that combines bottleneck analysis, lean methods, selective automation, preventive maintenance, and KPI monitoring improve productivity, cost efficiency, reliability, quality, energy performance, and worker safety in insulation-material production?

The working hypotheses are:

H1: 

The framework reduces non-value-added time in critical operations;

H2: 

The framework improves production output and cost efficiency;

H3: 

The framework improves operational stability and quality;

H4: 

The framework improves sustainability-related performance.

In this study, the CPS layer considered an enabling architecture that supports all four hypotheses through enhanced visibility, improved feedback quality, and greater intervention effectiveness.

Table 1. The representative literature benchmark and positioning of the present human-supervised CPS-oriented study.

Study	Sector/Context	Main Method(s)	Primary KPI Focus/Reported Effect	Gap Relative to This Study
Hozdić et al. [3]	Manufacturing CPPS	CPPS conceptual and control model	Cyber, physical, and human layers	Conceptual/simulation-oriented CPPS; no insulation-production before-and-after KPI validation
Sousa et al. [37]	Automotive injection/setup workshops	Structured SMED workshops	38.8% setup time reduction	Setup-centered intervention; no integrated logistics, energy, and safety KPI package
Hanifi et al. [31]	Smart manufacturing/legacy equipment	IIoT energy and air monitoring	Machine-level utility savings	Energy-focused intervention rather than a bounded multi-KPI plant intervention
Yang et al. [30]	Manufacturing operations	Digital-twin-enabled monitoring and scheduling	Improved operational visibility	Digital representation emphasis; no insulation-manufacturing before-and-after KPI package
Wang [40]	Construction inspection	GNN-based defect text classification	Automated fire-door defect categorization	Relevant for future inspection extension, not a factory-level production intervention
Wang [41]	Building AHU systems	Semi-supervised fault detection and diagnosis	AHU AFDD support	Relevant for future predictive maintenance extension, not the present plant implementation
Present study	Insulation-material production	Bottleneck analysis + VSM + SMED + selective automation + maintenance + KPI monitoring	+14.5% output; −10% monthly cost; −40% breakdowns; −40% scrap; −10% energy; 0 safety incidents	Integrated human-supervised CPS case with a conservative before-and-after KPI architecture

positions the present manuscript against the representative CPPS, lean, digital-monitoring, AI and industrial case study literature.

2. Materials and Methods

The study was designed as a single-site industrial intervention case study with a before–after evaluation framework implemented in a human-supervised cyber–physical manufacturing system. The methodological logic combines process diagnostics, lean intervention, selective automation, and KPI-based validation in a real insulation-material production environment. Because the intervention was implemented on a single operating production system, a fully controlled experimental design with replicated treatment and control lines was not feasible. The analysis therefore focuses on absolute and relative changes in key performance indicators before and after the intervention, triangulated across multiple operational data sources.

The physical layer comprised material preparation, crusher loading, setup/changeover operations, pallet transport, machine operation, maintenance activities, and downstream production flow. The cyber layer comprised MES-supported production data collection, smart monitoring of machine and energy performance, KPI visualization, and feedback routines connecting measured plant performance to operational and managerial responses. The architecture did not represent a fully autonomous closed-loop control system; it functioned as a human-supervised CPS environment in which digital visibility supported adaptive decision-making, faster response to deviations and iterative process optimization.

Table 2. Case setting, production-line context, and confidentiality boundaries.

Item	Description
Production context	Insulation-material production involving material preparation, crushing/grinding, loading, setup/changeover operations, pallet transport, maintenance, and downstream production flow.
Main product family	Insulation-material products with repeated material handling and conversion steps; product-specific details are anonymized due to commercial sensitivity.
Dominant baseline constraints	Manual crusher loading, long setup/changeover, forklift-based pallet transport, repeated equipment stoppages, material scrap, and limited KPI visibility.
Operational constraints	Single operating plant; no parallel untreated control line; phased implementation required to avoid unacceptable production disruption.
Confidentiality boundary	Vendor names, exact sensor models, plant layout coordinates, and proprietary monthly raw data are anonymized, but functional architecture, KPI definitions, and evidence sources are reported.
Equipment/software disclosure	Manufacturer names, supplier locations, exact sensor models and software versions are anonymized under the industrial confidentiality boundary; the manuscript therefore reports functional equipment classes, data sources and aggregation levels rather than vendor-specific identifiers.

expands the production-line setting and explains the confidentiality boundary that limits disclosure of vendor-specific details.

The as-is data reported a crusher loading time of 30 min/pallet, monthly production costs of EUR 200,000, breakdown frequency of 5 breakdowns/month, downtime/time losses of 10 h/month, daily productivity of 7864 kg/day, scrap of 5%, cycle time of 45 min/unit, and an average waiting time of 10 min between operations. Worker movement of approximately 5 km/shift also indicated suboptimal layout and internal material flow. These baseline values were used to identify the dominant bottlenecks and to prioritize the intervention package.

The baseline diagnosis used a combination of time studies, direct process observation, spaghetti diagrams, value stream mapping (VSM), 5 Whys analysis, fishbone analysis, MES-supported operational records, and a review of existing plant KPIs. Spaghetti diagrams were used to visualize worker and material movement. Time studies and production records quantified critical operation durations. VSM represented material and information flows and identified waiting times, transport losses, and cycle inefficiencies. Root-cause techniques were then applied to distinguish dominant losses from downstream symptoms.

Figure 2 presents the PDCA-based improvement workflow used to move from baseline diagnosis to monitored post-implementation stabilization.

The intervention followed a PDCA-based human-supervised CPS-oriented improvement logic and can be summarized in seven stages: (1) baseline diagnosis of the production system, (2) identification and prioritization of bottlenecks, (3) selection of lean and technical interventions, (4) phased implementation, (5) post-implementation KPI monitoring, (6) before–after comparison, and (7) corrective action and continuous improvement. Within this framework, lean methods and selective automation acted primarily on the physical layer, whereas KPI monitoring, digital data acquisition, and supervisory feedback acted on the cyber layer. The interaction of these layers enabled repeated adjustment of interventions based on observed production system behavior.

Figure 3 summarizes the root-cause structure that guided intervention prioritization.

Table 3. Intervention sequence, implemented measures and primary KPI influence.

Phase	Intervention Block	Implemented Action	Primary Influence
1	Baseline diagnosis	Time studies, direct observation, VSM, spaghetti diagram, 5 Whys/fishbone, review of KPI records	Loss mechanisms quantified before capital actions
2	SMED/changeover package	Additional jaw sets, quick clamping, lift-assisted handling, standard work	Setup/changeover time; safety; downtime
3	Crusher loading automation	Automatic crusher/pre-crusher loading support	Crusher loading time; productivity; manual handling risk
4	Pallet transport automation	Roller conveyor/internal transport automation	Transport time; waiting; operator travel
5	Maintenance and monitoring	Preventive maintenance, machine-state monitoring, deviation response	Breakdowns; downtime; availability
6	Digital inventory/labeling support	Labeling support, inventory visibility and KPI dashboard reporting	Order-processing speed; traceability; response time
7	Training and PDCA stabilization	Operator/supervisor training, follow-up corrective actions, standard-work updates	Sustained KPI stabilization and CAPA closure

clarifies the implementation sequence and separates the main intervention blocks, responding to the concern that multiple measures were introduced together.

The measures were intentionally implemented as a phased package rather than as isolated experiments. This design improves industrial feasibility but limits the ability to attribute each final KPI change to a single measure. The strongest directly attributable local effects were crusher-loading automation for loading time, SMED-oriented measures for setup/changeover time and transport automation for pallet movement. Broader effects on output, cost, downtime, scrap, energy and safety are interpreted as system-level outcomes of the combined intervention.

Table 4. KPI operational definitions, primary evidence sources, and role in hypothesis testing.

KPI	Operational Definition	Unit	Primary Evidence Source	Role
Crusher loading time	Time required to load the crusher per pallet in repeated time studies	min/pallet	Manual time study; implementation validation	H1
Setup/changeover time	Time required to complete validated wire/net setup–change tasks	min/change	Manual time study; SMED records	H1
Pallet transport time	Internal transfer time between workstations for pallet movement	min/operation	Spaghetti/VSM observation; time study	H1
Productivity	Average daily production output under representative operation	kg/day	Production logs; plant KPI records	H2
Monthly production cost	Total plant production cost per month	EUR/month	Plant cost records; consolidated KPI tables	H2
Breakdown frequency	Unplanned failure events causing production disturbance within one month	events/month	Maintenance and failure records	H3
Sustained unplanned downtime	Cumulative monthly duration of unplanned stoppages after stabilization	h/month	Maintenance records; annual consolidated KPI table	H3
Scrap rate	Defective output relative to total production	%	Quality and production reports	H3
Annual energy consumption	Annual plant energy use across the production system	MWh/year	Smart metering/energy monitoring records	H4
Safety incidents	Recorded work-related incidents reported by the plant	events/month	Safety log/incident records	H4

defines the KPI set used to evaluate the intervention and links each metric to its evidence source and hypothesis role.

Data were collected at daily, monthly and annual aggregation levels using repeated time studies, production records, maintenance logs, MES-supported operational data, and smart monitoring of plant performance and energy use. The resulting cyber layer provided ongoing visibility into process behavior and enabled feedback-informed intervention decisions. The CPS’s role was therefore expressed through improved visibility, faster response, and sustained KPI improvement rather than through an independent cyber performance metric.

The evidence base combined repeated operation-level time studies, daily production logs, monthly maintenance, and cost records, quality records, annual energy records, and consolidated plant KPI tables. Operation-level bottleneck indicators were retained when supported by repeated time studies and implementation validation. Plant-level indicators were retained when consistent with consolidated production, maintenance, quality, cost, and energy records. No statistical outlier filtering such as a 3σ or IQR rule was applied to undisclosed raw series; instead, inconsistent implementation-stage values were excluded and the final conservative KPI set was used as the source of truth.

Figure 4 provides a detailed CPS/IT-OT data-flow architecture for the implemented human-supervised manufacturing framework. It separates physical/OT signals, edge and plant-interface functions, cyber-layer aggregation and visualization, supervisory rules, and human-supervised feedback so that the physical-to-cyber and cyber-to-physical information paths are explicit.

Physical/OT data from crusher loading, setup/changeover, pallet conveyance, machine operation, and quality/safety logs flow upward via common industrial communication protocol classes (OPC UA, Modbus TCP and MQTT) through an acquisition and IT/OT integration layer (PLC/SCADA/MES signal capture, data validation, and aggregation buffer) to the cyber/IT data and analytics layer. There, KPIs (OEE, cost, and energy) are computed using a time-series historian and MES database, with analytics services and an API layer. A dashboard and rules layer provides visualization through the plant-standard dashboard interface, a threshold engine (OEE, downtime, or quality), and an alert queue with an audit log.

Human-supervised feedback—alarms, work orders, scheduling priorities, and PDCA/CAPA actions—flows back to operators, supervisors, and maintenance staff under human authority. A threshold update loop supports continuous improvement. The architecture expresses a plant-level diagnose–intervene–monitor–feedback logic and explicitly operates as a supervisory CPS; no autonomous high-speed closed-loop control is implemented, and native OT safety interfaces remain unchanged.

Industrial communication between physical/OT equipment and the cyber layer relied on common industrial communication protocol classes, including OPC UA, Modbus TCP, and MQTT. Time-series process data were stored in a dedicated process historian, while aggregated KPI data and MES records were maintained in a relational database. KPI dashboards were implemented using a plant-standard business intelligence interface, providing shift/daily visual indicators without exposing proprietary vendor identities or software versions.

Table 5. CPS-oriented architecture and IT/OT technical specification classes.

Layer	Components	Data/Interface Type	Sampling/Aggregation Class	Latency/Refresh Interpretation	Main Function
Physical/OT layer	Crusher loading, setup/changeover, pallet transport, machine operation, maintenance activities	PLC/SCADA-compatible equipment signals, operator entries, production/maintenance/quality records	Event, shift, daily and monthly records depending on KPI; no millisecond control-loop claim	Plant-record/dashboard refresh cycle; operational near-real-time means minutes-to-shift visibility, not sub-second control	Executes production and material flow
Cyber layer	MES-supported collection, KPI dashboard, energy monitoring, maintenance logs	MES/work-order records, energy meter records, quality and downtime logs	Daily/monthly KPI aggregation; local event flags for deviations	Dashboard visibility used for supervisory response and PDCA review	Aggregates and visualizes KPI state; computes OEE and derived indicators
Feedback layer	Deviation detection, alarms, CAPA/PDCA, maintenance planning	Threshold comparison and human-confirmed alerts	At each monitoring window W and when critical deviations are reported	Response time recorded operationally; not validated as a cyber metric	Turns deviations into corrective actions
Human layer	Operators, supervisors, maintenance, quality, energy/CAPA owners	Operator validation and authority levels	Continuous during operation; formal review by shift/day/month as relevant	Human authority remains final except existing safety procedures	Interprets, authorizes and closes corrective actions

expands the earlier conceptual CPS table by adding technical specification classes, data sources, sampling/aggregation level, latency interpretation, and human authority boundaries.

Vendor-specific manufacturer names, sensor models, server identifiers, software versions, and proprietary plant-interface details are anonymized under the industrial confidentiality boundary. For reproducibility, the manuscript reports the functional IT/OT architecture, data-source categories, aggregation levels, and decision logic. The system should not be interpreted as a sub-second autonomous control system; rather, near-real-time is used in the operational supervisory context of minutes-to-shift visibility for deviations and KPI updates. No version-dependent software routines were used in the reported calculations.

To clarify how the cyber layer operated in the implemented human-supervised CPS environment, the KPI monitoring and feedback routine is shown as a graphical algorithmic flowchart in Figure 5. The figure illustrates the data pathway from physical/OT signals to cyber-layer KPI computation and from KPI deviations back to supervisory action. It is intentionally presented as a human-supervised decision-support procedure rather than as a fully autonomous closed-loop controller, because the implemented system generated dashboard updates, alerts, work-order/CAPA actions, and PDCA feedback, while final operational authority remained with operators, supervisors, and maintenance personnel. The figure provides the algorithmic representation of the KPI monitoring and supervisory feedback logic and clarifies the pathway from KPI deviation to human-supervised corrective action.

The flowchart describes the complete monitoring cycle: data acquisition from the physical/OT layer, validation, and KPI computation in the cyber layer, threshold-based deviation detection with safety-critical escalation, human-supervised assignment and execution of corrective actions, persistence checking, and PDCA/CAPA escalation. The loop repeats after the defined monitoring interval and preserves human authority for all operational decisions.

Table 6. Research questions, hypotheses and validation logic.

Item	Statement	Main Validation Metrics
RQ1	How, and to what extent, can a human-supervised CPS combining bottleneck analysis, lean methods, selective automation, preventive maintenance and KPI monitoring improve insulation-material production?	Integrated before–after KPI package; OEE consistency check; descriptive interpretation
H1	The CPS-oriented framework reduces non-value-added time in critical operations.	Crusher loading time; setup/changeover time; pallet transport time; cycle time
H2	The CPS-oriented framework improves production output and cost efficiency.	Productivity; monthly cost; annual output; unit cost
H3	The CPS-oriented framework improves operational stability and quality.	Breakdown frequency; sustained downtime; scrap rate; availability/OEE
H4	The CPS-oriented framework improves sustainability-related performance.	Annual energy consumption; energy per unit; safety incidents

links the research question and hypotheses to the validation strategy used in Section 3.

To support the formal evaluation of these hypotheses, the OEE-based model and the algorithmic feedback logic introduced earlier in Section 2 use a set of mathematical symbols and variables. For clarity, all symbols are listed in

Table 7. Nomenclature and mathematical symbols used in the model and in the graphical algorithmic flowchart.

Symbol	Meaning	Unit/Type
A	Availability component of OEE	dimensionless/%
C	Annual production cost	EUR/year
c	Specific production cost	EUR/kg
E	Annual energy consumption	kWh/year or MWh/year
e	Specific energy consumption	kWh/kg
$K_{pre}$	Baseline KPI vector before implementation	data vector
$K_t$	KPI vector at time t	data vector
$L_t$	Alarm/deviation level at time t	categorical
N	Number of consecutive windows for persistence checking	count
OEE	Overall Equipment Effectiveness	dimensionless/%
P	Performance component of OEE	dimensionless/%
Q	Daily good output	kg/day
$Q_a$	Annual output	kg/year
$Q_{cap}$	Theoretical daily line capacity	kg/day
$R_t$	PDCA/CAPA record at time t	record
$S_t$	Physical/OT data streams and plant records at time t	data vector
W	KPI aggregation window	shift, day, month
Y	Quality yield component of OEE	dimensionless/%
$a_t$	Corrective action or work order at time t	action record
Δt	Monitoring interval	time
$\Theta$	Plant-specific KPI threshold vector	data vector

.

To formalize the production logic and provide an internal consistency check between plant KPIs, the study introduced an OEE-based surrogate model following established OEE formulations in the manufacturing literature [42,43]. The model was not used to replace the plant KPI system or to claim high-frequency autonomous control; it was used to verify whether reported productivity changes were numerically coherent with measured OEE values.

$$\mathrm{Q}={\mathrm{Q}}_{\mathrm{c}}\mathrm{a}\mathrm{p}\times \mathrm{O}\mathrm{E}\mathrm{E}={\mathrm{Q}}_{\mathrm{c}}\mathrm{a}\mathrm{p}\times \mathrm{A}\times \mathrm{P}\times \mathrm{Y}$$

(1)

$$OEE=A\times P\times Y$$

(2)

$$c=C/Q_a$$

(3)

$$e=E/Q_a$$

(4)

$$Q_{c}ap=Q/OEE$$

(5)

$$S_x=\left[\right(f(x+\Delta x)-f\left(x\right))/f(x\left)\right]/[\Delta x/x]$$

(6)

In Equations (1) and (2), A represents availability, P is performance and Y is quality yield. Availability was anchored in validated downtime/availability records, while quality yield was anchored in the reported scrap rate. The performance factor was treated as the implied residual OEE component required to reconcile realized output with theoretical line capacity under stabilized operating conditions. Output-normalized cost and energy indicators were derived from annual plant records by normalizing annual cost and annual energy use by annual output.

Table 8. Data aggregation, disclosure limitations and statistical interpretation.

Indicator Group	Original Data Basis	Aggregation Used	Outlier/Data-Quality Treatment	CV/Raw-Data Status	Interpretation
Time study indicators	Repeated operational time studies and implementation records	Validated before–after values for critical operations	Obvious recording errors checked by operator/supervisor verification	Raw replicate-level SD/CV not available for disclosure	Descriptive before–after comparison
Production and cost indicators	Monthly reporting windows before and after intervention (n = 12 + 12) and annual consolidated records	Monthly means and annual consolidated totals	Cross-check against consolidated KPI records; no formal 3σ/IQR exclusion reported	Full raw monthly dataset commercially sensitive; CV not computed	Descriptive and practically meaningful difference only
Maintenance and quality indicators	Monthly counts/logs and annual consolidation	Events/month, h/month and scrap %	Reconciled with maintenance/quality logs	SD/CV not disclosed	Descriptive difference supported by triangulation
Energy and safety indicators	Energy records and safety incident logs	Annual energy and monthly safety incident rates	Consistency checks against energy and safety records	Raw time-series unavailable for publication	Descriptive sustainability and safety outcomes
Inferential statistics	Not applicable to the available manuscript package	No paired t-test, ANOVA, Wilcoxon, confidence interval, Durbin–Watson or decomposition reported	Avoid pseudo-replication from aggregated single-case data	Not computed	Results intentionally labeled as descriptive and practically meaningful

clarifies aggregation levels and statistical interpretation, avoiding any implication that descriptive industrial data were analyzed as a fully controlled replicated experiment.

Because the submitted manuscript package contains consolidated plant-level KPI values rather than the full raw monthly dataset, the results are interpreted as descriptive and practically meaningful before–after differences. Classical inferential tests, confidence intervals, coefficient-of-variation values and time-series diagnostics were therefore not reported, because the raw monthly observations required for such analyses were not available for disclosure. The study reports aggregated values and triangulates time studies, MES records, maintenance logs and energy-monitoring outputs; no paired t-tests, repeated-measures ANOVA, Wilcoxon signed-rank tests, Durbin–Watson tests, time-series decomposition or CV calculations were performed on unpublished raw monthly observations.

Additional evidence-source mapping, CAPEX assumptions and calculation notes are provided in Appendix A.

3. Results

The results retain only annual consolidated KPI tables and the most internally consistent operation-level measurements as the source of truth. More optimistic target-state or workshop-stage values reported during implementation were not used as headline results. This conservative approach is important in a CPS-oriented manuscript because the credibility of the cyber–physical claim depends on alignment between digital performance visibility, recorded KPI values and validated plant-level outcomes.

The first result block addresses H1. Repeated time studies and post-implementation analysis showed substantial reductions in all three major bottleneck operations. Crusher loading time decreased from 30 min/pallet to 10 min/pallet (−66%), validated setup/changeover time decreased from 30 min to 15 min (−50%) and pallet transport time decreased from 10 min to 4 min (−60%). Cycle time decreased from 45 min/unit to 30 min/unit. These results indicate that the intervention shortened the most time-intensive non-value-adding activities identified in the baseline stage. Figure 6 visualizes the validated local time reductions that support H1.

The second result block addresses H2. Daily productivity increased from 7864 kg/day to 9000 kg/day (+14.5%). Annual production increased from 2,872,360 kg to 3,285,000 kg, corresponding to an additional 412,640 kg/year. Monthly production costs decreased from EUR 200,000 to EUR 180,000 (−10%), and annual production costs decreased from EUR 2,400,000 to EUR 2,160,000, representing annual savings of EUR 240,000. These outcomes show that the intervention improved both throughput and cost efficiency. Figure 7 summarizes the main validated plant-level changes used as headline results.

The third result block addresses H3. Breakdown frequency decreased from 5 to 3 events/month (−40%), equivalent to a decrease from 60 to 36 breakdowns/year. Sustained unplanned downtime decreased from 10 h/month to 5 h/month after PDCA stabilization. Scrap decreased from 5% to 3% (−40%). The combination of lower breakdowns, lower downtime, higher availability, and lower scrap indicates improved structural stability rather than a simple increase in production speed.

The fourth result block addresses H4. Annual energy consumption decreased from 500 MWh/year to 450 MWh/year (−10%), and energy consumption per production unit decreased from 30 kWh/unit to 25 kWh/unit. Reported safety incidents decreased from 3/month to 0/month during the 12-month post-implementation observation period, consistent with reduced manual handling and automation of loading and transport tasks. Figure 8 visualizes the sustainability- and safety-related outcomes that support H4.

Table 9. Consolidated validated pre- and post-implementation results used for the main claims of the manuscript.

KPI	Pre-Implementation	Post-Implementation	Change
Crusher loading time	30 min/pallet	10 min/pallet	−66%
Validated setup/changeover time	30 min	15 min	−50%
Pallet transport time	10 min	4 min	−60%
Productivity	7864 kg/day	9000 kg/day	+14.5%
Annual production	2,872,360 kg	3,285,000 kg	+412,640 kg
Monthly production costs	EUR 200,000	EUR 180,000	−10%
Annual production costs	EUR 2,400,000	EUR 2,160,000	−10%
Cost per production unit	EUR 50/unit	EUR 45/unit	−10%
Breakdown frequency	5/month	3/month	−40%
Annual breakdowns	60/year	36/year	−40%
Sustained unplanned downtime	10 h/month	5 h/month	−50%
Scrap rate	5%	3%	−40%
Overall Equipment Effectiveness	75%	85%	+10 p.p.
Machine availability	85%	95%	+10 p.p.
Productivity per hour	160 units/h	220 units/h	+60 units/h
Cycle time	45 min/unit	30 min/unit	−15 min/unit
Order-processing speed	5 days	3 days	−2 days
Energy consumption per unit	30 kWh/unit	25 kWh/unit	−5 kWh/unit
Annual energy consumption	500 MWh/year	450 MWh/year	−10%
Estimated environmental impact	1000 t CO2/year	900 t CO2/year	−10%
Safety incidents	3/month	0/month	−100%

consolidates the before–after KPI results used as the main evidence base of the manuscript.

Using Equation (5), the back-calculated theoretical capacity was 10,485 kg/day in the pre-intervention state and 10,588 kg/day in the post-intervention state. The relative discrepancy between the two estimates was only 0.98%, indicating good internal consistency between observed productivity values and independently reported OEE values. The implied performance factor remained approximately stable at 92–93%, suggesting that the main improvement mechanism was the combined gain in availability and quality yield rather than an overstatement of nominal running speed.

Table 10. OEE decomposition, internal consistency check, derived efficiency indicators, and OEE sensitivity scenarios.

Indicator/Scenario	Pre	Post/Predicted	Interpretation
Availability A (%)	85.0	95.0	Anchored in validated availability records
Quality yield Y (%)	95.0	97.0	Implied from validated scrap rates
Implied performance factor P (%)	92.9	92.3	Residual OEE component under plant-level aggregation
Back-calculated theoretical capacity (kg/day)	10,485	10,588	0.98% discrepancy across states
Specific production cost (EUR/kg)	0.836	0.658	−21.3% vs. baseline
Specific energy consumption (kWh/kg)	0.174	0.137	−21.3% vs. baseline
Predicted output at OEE = 0.80 (kg/day)	—	8470	Scenario below validated post-state
Predicted output at OEE = 0.90 (kg/day)	—	9529	+5.9% vs. OEE = 0.85 state
Predicted output at OEE = 0.92 (kg/day)	—	9741	+8.2% vs. OEE = 0.85 state

summarizes the OEE decomposition, model-based consistency check, and post-state sensitivity scenarios.

4. Discussion

The results support all four hypothesis groups. Non-value-added time was reduced in the most critical operations, throughput increased, production costs declined, breakdown frequency and scrap fell, and energy and safety indicators improved in parallel. This pattern suggests that the intervention did not merely shift losses from one part of the system to another; instead, the production system became faster, more stable, less wasteful and safer.

A key explanatory point is that the intervention was built around measured bottlenecks rather than technology-first implementation. Crusher loading, setup/changeover, pallet transport, breakdowns, and scrap were identified as dominant sources of loss before the capital and organizational measures were selected. The CPS layer did not replace this logic; it strengthened it by improving performance visibility, deviation response, and PDCA-based decision quality.

Compared with SMED-centered studies, the present case reports a setup improvement but extends the evidence base to loading, logistics, breakdowns, scrap, energy, and safety. Compared with energy-monitoring and digital-twin studies, it is less technologically ambitious but stronger as a conservative before–after industrial KPI validation. Compared with insulation circularity studies, its contribution lies at the factory-operation level rather than at the product-system or recycling-pathway levels.

The CPS’s claim is deliberately limited. The implemented architecture should be interpreted as a human-supervised CPS, not as a fully autonomous digital twin, device twin or closed-loop controller in the strict control-engineering sense. The cyber layer primarily provided data acquisition, aggregation, visualization, deviation classification and feedback support, while final intervention authority remained with operators, supervisors and maintenance personnel. This position aligns with CPS frameworks that include digital, physical and human components [1,2,3,4,5], and it clearly distinguishes the present case from more digitally ambitious architectures that implement autonomous scheduling, high-fidelity model synchronization or device-twin control [30,32].

The intervention also has managerial implications. First, the most effective improvements targeted measured bottlenecks rather than fashionable technologies. Second, a mixed package combining lean methods, selective automation, maintenance and KPI monitoring appears more robust than isolated tool deployment. Third, improvement projects should use KPI architectures that include throughput, cost, reliability, quality, energy and safety rather than only output and cost [6,7,8,9,25,26,33,44].

Table 11. Economic sensitivity to hidden implementation costs.

Scenario	Adjusted Investment	Assumption	Interpretation
Base equipment package	EUR 73,000	EUR 240,000/year annual cost reduction	Indicative payback ≈ 3.7 months; reported project amortization ≈ 9 months due to phased implementation
+10% hidden implementation cost	EUR 80,300	Software, dashboard tuning, training and early maintenance allowance	Still below first-year annual savings
+20% hidden implementation cost	EUR 87,600	More conservative allowance for software/training/maintenance	Still below first-year annual savings
+30% hidden implementation cost	EUR 94,900	High allowance for undisclosed lifecycle costs	Still positive under the observed annual savings, but full lifecycle appraisal remains future work

extends the economic discussion by adding a conservative sensitivity check for hidden implementation costs such as software, dashboard refinement, training and maintenance.

For transfer to comparable process industries, the case suggests the following sequence: quantify physical bottlenecks using time studies, VSM and maintenance records; define a conservative KPI architecture before intervention; prioritize actions that jointly reduce waiting, manual handling and unplanned stoppages; connect the affected operations to a cyber layer for KPI visibility; validate the intervention with a consolidated pre/post evidence set; and use model-based consistency checks to test whether throughput, OEE, cost and energy figures remain coherent. The same logic could support maintenance and inspection-related applications, for example, by linking defect-classification methods for fire-door inspection or semi-supervised fault detection for air-handling units to future CPS monitoring and CAPA workflows [40,41].

Several limitations should be acknowledged. First, the study is based on a single industrial case, which limits statistical generalizability and supports analytical transferability rather than broad statistical inference. Second, the current manuscript package contains consolidated plant-level observations rather than the full raw monthly dataset; therefore, the before–after differences are interpreted primarily as descriptive and practically meaningful, not as statistically tested effects. Third, the cyber layer was evaluated primarily through its contribution to plant-level performance improvement rather than dedicated cyber–physical metrics such as latency, event-detection accuracy, model fidelity or automated response quality. Fourth, a full discounted lifecycle appraisal, including software, training, maintenance and cybersecurity costs, was outside the scope of the present study. Fifth, AI-assisted predictive maintenance and defect-classification modules were discussed as future extensions but were not implemented.

5. Conclusions

This paper presented a data-driven industrial case study of insulation-material production optimization within a human-supervised cyber–physical manufacturing system. The intervention combined bottleneck analysis, lean methods, selective automation, preventive maintenance, and KPI monitoring to address losses related to crusher loading, setup/changeovers, pallet transport, breakdowns, scrap, and energy use. Across the validated KPI set, the plant improved simultaneously in operational, economic, sustainability, and worker-safety dimensions.

The most important validated outcomes were the crusher loading reduction from 30 to 10 min/pallet, validated setup/changeover reduction from 30 to 15 min, pallet transport reduction from 10 to 4 min, productivity increase from 7864 to 9000 kg/day (+14.5%), monthly production cost reduction from EUR 200,000 to EUR 180,000 (−10%), breakdown-frequency reduction from 5 to 3 events/month (−40%), scrap reduction from 5% to 3% (−40%), annual energy reduction from 500 to 450 MWh (−10%) and a reduction in reported safety incidents decreasing to zero during the 12-month post-implementation observation period.

The main contribution lies in demonstrating that insulation-material production can be optimized when lean interventions, selective automation and KPI-based monitoring are integrated through a feedback-supported, human-supervised CPS architecture. The study does not present a CPS as an abstract digital label, a fully autonomous closed-loop controller or a full digital twin; it presents a practically implemented manufacturing environment in which cyber and physical components jointly supported measurable improvement while final corrective action authority remained with humans.

In addition to the intervention framework itself, the paper contributes a compact OEE-based model that cross-checks throughput consistency and reveals stronger unit-level gains in cost and energy efficiency than are visible from the absolute before–after totals alone.

Future work should extend the framework through longer observation windows, multi-site replication, fuller inferential statistics based on shareable monthly data, explicit lifecycle cost appraisal, dedicated cyber–physical performance metrics, AI-assisted predictive maintenance modules and more rigorous carbon-accounting methodology.