Applied IT & Engineering

1. Introduction

Manufacturing has never been a static enterprise. Across decades and continents, shop floors have been asked to do more with less — more output, tighter tolerances, shorter lead times, smaller margins for error. And yet the underlying systems many facilities still rely on were designed for a different era entirely. The result, in much of the sector, is a quiet accumulation of inefficiency: unplanned downtime, material waste, bottlenecked workflows, and a creeping inability to respond when demand shifts. Frank et al. (2019) documented this pattern across a broad cross-section of manufacturing firms, finding that companies clinging to legacy production architectures consistently struggle to keep pace with competitive and environmental pressures. It is not a new problem — but it has become an urgent one.

The arrival of Industry 4.0 did not so much solve this problem as present a credible path toward solving it. The Internet of Things (IoT), artificial intelligence (AI), and Lean manufacturing — each powerful in isolation — together offer something more interesting: a mutually reinforcing set of tools that address different layers of the same inefficiency. IoT makes the invisible visible, providing real-time data streams from equipment, sensors, and production lines that operators previously had no direct window into. Vlachos et al. (2023) showed that when machines are networked this way, predictive maintenance becomes a practical reality rather than an aspiration — faults are flagged before they cascade into costly failures, and machine uptime improves accordingly. AI, meanwhile, makes sense of the data that IoT generates. Trakadas et al. (2020) proposed a collaborative AI framework for industrial IoT environments in which machine learning algorithms continuously identify patterns in production data, enabling faster, more accurate decision-making than any manual review process could sustain. And Lean — older than both, and sometimes underestimated in discussions about digital transformation — provides the philosophical scaffolding: a commitment to eliminating non-value-adding activities and standardizing what remains (Chiarini & Kumar, 2021).

What makes this combination genuinely promising, at least in theory, is that these three approaches are not redundant. They operate at different levels. IoT feeds data. AI interprets it. Lean disciplines the process around it. Rahardjo et al. (2023) put this rather directly in their examination of smart manufacturing systems, finding that the integration of Lean principles with Industry 4.0 technologies produced measurable gains in efficiency that neither approach achieved alone. Wang et al. (2018) reached similar conclusions in the context of energy-intensive manufacturing, where IoT-enabled monitoring allowed facilities to optimize consumption in ways that manual tracking simply could not match.

There is, however, a complication — and it is worth naming honestly. Most of the research demonstrating these benefits has been conducted in, or in partnership with, large enterprises. The infrastructure costs, the technical expertise required, the organizational change management involved — these are not trivial, and they fall disproportionately hard on small and medium-sized enterprises (SMEs). Vlachos et al. (2023) acknowledged this gap explicitly, noting that SMEs frequently face barriers of cost, integration complexity, and workforce skill gaps that the dominant literature has not adequately addressed. Moeuf et al. (2018) similarly found that SMEs in the Industry 4.0 era operate in a kind of information vacuum: the potential benefits are well-publicized, but practical, scalable implementation guidance remains scarce.

It is this gap that motivates the present study. Rather than adding another theoretical endorsement of IoT-AI-Lean integration to an already substantial literature, this paper attempts something more applied: a strategic framework, grounded in the DMAIC (Define, Measure, Analyze, Improve, Control) methodology, that offers manufacturers — and SMEs in particular — a structured, resource-conscious path from current-state diagnosis to operational improvement (Rodriguez Delgadillo et al., 2022; Hughes et al., 2022). The ambition is not to be exhaustive. It is to be useful.

2. Materials and Methods

2.1 Research Design

This study adopts a mixed-methods design — a choice that deserves brief justification, because mixed methods is sometimes invoked loosely to mean "we used more than one data source." That is not the intention here. The quantitative strand was concerned with process performance: cycle times, idle times, downtime frequencies, and energy consumption across five production departments. The qualitative strand addressed something harder to measure — the human and organizational dimensions of shop floor dysfunction, captured through structured observation, site visits, and recorded discussions with floor supervisors and process engineers. Together, these two strands were intended to produce a richer, more internally valid picture than either could alone.

The overarching analytical framework is the DMAIC cycle (Define, Measure, Analyze, Improve, Control), drawn from Lean Six Sigma methodology. This choice was deliberate. DMAIC is not simply a data collection scaffold; it is a causal reasoning structure that forces the researcher to move from problem identification through root cause analysis to proposed intervention, rather than jumping prematurely to solutions. Rodriguez Delgadillo et al. (2022) demonstrated this structure's utility in quality and sustainability improvement contexts, and Hohmeier et al. (2022) applied it productively in a similarly complex, multi-variable operational environment. The framework guided every phase of this study's design, from the initial definition of key performance indicators to the formulation of technology integration recommendations.

2.2 Study Setting and Case Selection

The study was conducted in a semi-automated manufacturing facility producing skid steer loaders. The facility comprised five production departments — Assembly, Fabrication, Painting, Hot Testing, and Quality Inspection — operating across 25 workstations with a combined workforce of 52 employees (46 direct shop floor workers, six supervisory/managerial staff). The facility's daily working time was 570 minutes, with a documented routine downtime allowance of 75 minutes, yielding an effective production window of 495 minutes per shift. Total recorded cycle time across all departments was 6,800 minutes, against a combined idle time of 470 minutes — figures that, even at a glance, suggest systemic rather than incidental inefficiency.

This facility was selected as the primary case context for several reasons. It represented a recognizable class of mid-scale manufacturer: partially automated, using legacy planning approaches (production scheduling was recorded as "random"), with no active condition monitoring system and no standardized operational guidelines in place. Moeuf et al. (2018) described precisely this profile as typical of SMEs attempting to navigate Industry 4.0 pressures without the structural support that larger enterprises command. The selection was thus purposive rather than random — not a statistical sample, but a theoretically representative case.

2.3 Data Collection

Data collection proceeded through two parallel channels.

Secondary data formed the conceptual and comparative foundation of the study. Published case studies, peer-reviewed articles, and industry reports documenting IoT, AI, and Lean manufacturing implementations were systematically reviewed. Sources were identified through database searches and selected on the basis of methodological relevance and publication recency (2015–2023). This literature served multiple functions: it informed the design of the observational protocol, supplied benchmarks against which facility-level findings could be compared, and provided the theoretical basis for the integration framework proposed in Section 3. Frank et al. (2019), Trakadas et al. (2020), and Ciano et al. (2021) were among the anchor sources used to establish the empirical landscape of Industry 4.0 technology adoption patterns.

Primary observational data were gathered directly at the facility across multiple site visits. Data collection methods included direct inspection of workstations, review of production records, timed observation of process flows, and structured discussions with department supervisors and process engineers. The observed variables — cycle time, idle time, downtime, ergonomic conditions, planning approach, data storage medium, and condition monitoring status — were documented systematically for each of the five departments. Where discrepancies arose between records and observed conditions, the site observation took precedence. Table 1 summarizes the full profile of observed facility characteristics, with source attribution by collection method (Chiarini et al., 2020; Ciano et al., 2021; Frank et al., 2019; Liao & Wang, 2021; Rahardjo et al., 2023).

2.4 Experimental Setup and Technology Integration Protocol

The experimental component of this study involved configuring a technology integration testbed within selected areas of the production facility. IoT sensors were deployed on key machinery to capture real-time data on three primary variables: machine performance (vibration, temperature, operational load), downtime events (frequency, duration, triggering condition), and energy consumption per production cycle. Sensor placement

Table 1. Baseline Operational Profile of the Study Facility. A comprehensive characterization of the manufacturing facility under investigation, documenting observed operational parameters across 19 variables collected through multiple data collection methods — including production records, direct site visits, structured supervisory discussions, and physical inspections. The facility, producing skid steer loaders across five production departments and 25 workstations, employed 52 personnel and operated on a 570-minute working day with a recorded cycle time of 6,800 minutes and combined idle time of 470 minutes. Notable gaps identified at baseline include the absence of a condition monitoring system, unavailability of standardized operational guidelines, and reliance on manual data storage and random production planning — conditions collectively representative of SME manufacturing facilities in early-stage Industry 4.0 transition. Data attribution by collection method is indicated in the Source column. Adapted from Chiarini et al. (2020), Ciano et al. (2021), Frank et al. (2019), Liao & Wang (2021), and Rahardjo et al. (2023).

Table 2. Departmental Performance Metrics and Identified Problem Areas in the Existing Shop Floor Management System (SFMS). Quantitative performance data for each of the five production departments — Assembly, Fabrication, Painting, Hot Testing, and Quality Inspection — recorded during the baseline diagnostic phase of the study. For each department, cycle time (minutes), idle time (minutes), and downtime (minutes) are reported alongside a qualitative summary of the primary operational problems identified through direct observation and supervisory consultation. Assembly recorded the highest cycle time (2,895 minutes) and second-highest downtime (240 minutes), attributable to ergonomic deficiencies, poor work allocation, inter-workstation congestion, and elevated defect rates. Painting sustained the greatest downtime of any department (330 minutes), driven by inadequate material handling control and irregular part sequencing. Fabrication, while shortest in cycle time (860 minutes), experienced frequent unplanned machinery failures and safety incidents. Hot Testing and Quality Inspection both exhibited inefficiencies arising from absent planning structures and inconsistent operator assignment. These figures collectively establish the pre-intervention performance baseline against which technology integration outcomes should be assessed.

followed the anomaly map developed during the Define phase of DMAIC, targeting equipment in the Assembly and Fabrication departments where downtime rates were highest (Table 2).

Raw sensor data were routed to an AI-driven analytics platform capable of running predictive maintenance algorithms — specifically, pattern recognition routines trained to flag deviation from baseline operational parameters. The analytics layer drew on the approach described by Trakadas et al. (2020), whose AI-IoT collaboration framework for industrial environments provided the closest published analogue to the setup used here. Concurrently, Lean tools — principally 5S (Sort, Set in order, Shine, Standardize, Sustain) and Six Sigma defect analysis — were applied to the same departments to address the non-technological sources of waste: workstation congestion, operator task ambiguity, and material handling irregularities. Chiarini & Kumar (2021) provide a detailed account of how Lean Six Sigma and Industry 4.0 technologies can be operationalized together; this study used that integration logic as its guiding reference.

Before full implementation, a pilot phase was conducted on a single production line within the Assembly department. The pilot ran for a defined observation period, during which the integrated system was monitored for stability, data quality, and operator usability. Adjustments to sensor thresholds and alert parameters were made iteratively during this phase. Only after pilot performance was evaluated and found acceptable did the team proceed to broader deployment across the remaining departments. This staged approach, though more time-consuming than a simultaneous rollout, substantially reduced the risk of system interference with live production operations.

2.5 Analytical Methods

Analysis proceeded in stages, each corresponding to a phase of the DMAIC framework. During the Measure phase, descriptive statistics were calculated for all quantitative process variables — cycle time, idle time, and downtime — across each of the five departments, establishing a pre-integration performance baseline. During the Analyze phase, regression analysis was used to examine relationships between key variables: specifically, the association between AI-flagged anomaly events and subsequent downtime occurrences, and between 5S implementation scores and idle time reduction. Value Stream Mapping (VSM) was applied to visualize end-to-end production flow and identify where delays clustered. Root cause analysis, supported by AI pattern outputs, was used to distinguish structural inefficiencies (layout, planning) from equipment-driven failures (machinery malfunction, energy spikes).

Qualitative data from supervisory discussions were analyzed thematically, with themes organized around the anomaly categories in Table 3: machinery, manpower, layout, quality, and operational constraints. Mendes et al. (2023) used a similar thematic structure in their maintenance management study, and their category scheme provided a useful cross-check for completeness. Where quantitative and qualitative findings converged — for instance, where high downtime figures in a department aligned with supervisor reports of recurring equipment failures — the finding was treated as corroborated. Where they diverged, the discrepancy was noted and explored rather than resolved by assuming one source was correct.

No ethical approval was required under the institutional guidelines applicable to this study, as all data collection involved process observation and document review rather than human subjects research in the clinical sense. All discussions with facility personnel were conducted with the knowledge and consent of site management.

3. Results

3.1 Baseline Characterization of the Shop Floor Management System

Before any integration of new technologies could be considered, it was necessary to understand — with some precision — what the existing shop floor actually looked like. Not in the abstract, but operationally: where time was being lost, which departments were underperforming, and why. This diagnostic work is sometimes treated as a formality in manufacturing research, a box to check before moving to the interesting part. In this study, it turned out to be anything but.

The facility under examination produced skid steer loaders across five departments — Assembly, Fabrication, Painting, Hot Testing, and Quality Inspection — operating through 25 workstations with a workforce of 52 personnel (Table 1). On paper, the operation ran a 570-minute working day with 75 minutes of scheduled downtime, leaving an effective production window of 495 minutes per shift. The total cycle time recorded across all departments, however, was 6,800 minutes — a figure that, when set against the available productive time, signals something considerably more disorganized than routine inefficiency. Combined idle time stood at 470 minutes. Production planning was documented as "random." No condition monitoring system was in place. No standardized operational guidelines existed (Table 1). These were not minor gaps.

What emerged from the initial shop floor characterization was a facility that had, in many respects, outgrown its own management systems. Semi-automation had been introduced incrementally, but the surrounding processes — scheduling, monitoring, quality control — remained largely manual and reactive. Moeuf et al. (2018) described this profile as characteristic of SMEs caught between legacy operations and the expectations of Industry 4.0: possessing the physical infrastructure for modernization, but lacking the data systems and process discipline to leverage it.

3.2 Departmental Performance Analysis: Where the System Was Failing

Breaking the aggregate figures down by department revealed a more differentiated picture — and, importantly, suggested that the causes of inefficiency were not uniform across the facility (Table 2).

Assembly carried the heaviest burden. With a cycle time of 2,895 minutes, idle time of 140 minutes, and 240 minutes of recorded downtime per cycle, it was by some margin the most troubled department (Table 2). The contributing problems were varied and, in combination, mutually reinforcing: ergonomic issues created physical discomfort and slowed operator throughput; poor work allocation meant tasks were unevenly distributed; defect rates generated rework loops; and congestion between adjacent workstations created spatial bottlenecks that no amount of individual effort could fully compensate for. These are, it should be noted, exactly the kinds of problems that Lean methodologies were designed to surface — and that IoT-based monitoring can help sustain improvements against, once Lean interventions have been applied (Chiarini & Kumar, 2021).

Fabrication, by contrast, had a much shorter cycle time (860 minutes) but suffered from machinery malfunction, safety incidents, and frequent platform changeovers (Table 2). The relatively compact cycle time is somewhat misleading here — what the data cannot show directly, but what supervisory discussions confirmed, is that this department's downtime events (105 minutes) were often sudden and unplanned, the kind that propagate disruption downstream rather than simply absorbing time locally. Wang et al. (2018) noted precisely this distinction between predictable and unpredictable downtime in IoT-monitored manufacturing environments, finding that the latter category is disproportionately costly relative to its measured duration.

Painting presented a different challenge again: 2,015 minutes of cycle time, 330 minutes of downtime — the highest downtime of any department — driven primarily by poor material handling control and irregular gaps between parts arriving at the station (Table 2). Some painting services had been outsourced, which introduced additional coordination complexity that the existing planning system (recall: "random") was poorly equipped to manage. Hot Testing recorded 1,830 minutes of cycle time alongside 150 minutes of downtime, with supervisors citing a lack of forward planning and inconsistent vehicle positioning as the principal sources of delay (Table 2). Quality Inspection, finally, logged 1,030 minutes of cycle time and 165 minutes of downtime, attributable largely to manual, unplanned inspection processes and the continuous rotation of operators across stations — a practice that, however common, reliably increases variability and error rates (Rahardjo et al., 2023).

Taken together, these figures point to a shop floor where inefficiency is structural, not incidental. The problems are not concentrated in one weak department that could be isolated and fixed; they are distributed, interconnected, and — in several cases — actively amplified by the absence of the monitoring and planning infrastructure that IoT and AI are precisely designed to provide (Trakadas et al., 2020).

3.3 Anomaly Classification and Technology-Matched Interventions

With the departmental performance baseline established, the next step was to classify the identified anomalies more systematically — not simply listing what was wrong, but organizing it in a way that made the path to intervention legible. Five anomaly categories emerged from the combined analysis of production records, site observations, and supervisory discussions: machinery, manpower, layout, quality, and operational constraints (Table 3).

Machinery anomalies — elevated vibration, thermal irregularities, unexpected breakdowns, and excess energy consumption — were, somewhat unsurprisingly, the most straightforward to address conceptually. Real-time online monitoring via IoT sensors offers a direct and well-validated response to exactly this category of problem. Vlachos et al. (2023) demonstrated in an AGV/IoT integration study that continuous sensor-based monitoring of equipment parameters can reduce unplanned breakdown frequency substantially, by converting reactive maintenance into a predictive regime where faults are flagged before they become failures (Table 3).

Manpower anomalies — particularly operator disengagement and uneven task distribution — were subtler. The proposed response here was a feedback-based work allocation system, drawing on historical production data to assign tasks more equitably and to flag when individual workstations were consistently over- or underloaded. This is less a technological fix than a data-informed management intervention, and its success depends at least as much on organizational buy-in as on the quality of the underlying data (Table 3).

Layout anomalies, principally unnecessary operational stoppages and suboptimal inter-station distances, called for AI-assisted path optimization — an application that Wang et al. (2020) examined in the context of augmented reality-enhanced shop floor management, finding that spatial inefficiencies are often invisible to the people working within them but highly legible to analytics systems with a whole-floor view (Table 3). Quality anomalies, including data entry errors and missing components, were mapped to smart sensor deployment and continuous online monitoring — interventions that reduce human error by catching deviations at the point of origin rather than at final inspection (Table 3). Operational constraint anomalies — variability in timing, resource unavailability — were assigned to online analysis systems and smart device integration, enabling more responsive, real-time adjustment than periodic manual review allows (Table 3).

3.4 The DMAIC-Integrated Technology Framework

The anomaly analysis above provided the empirical foundation for the study's central contribution: a structured framework for integrating IoT, AI, and Lean practices within the DMAIC cycle of continuous improvement. It is worth being clear about what this framework is and what it is not. It is not a guarantee of outcomes — the literature is full of integration initiatives that were well-designed in principle and poorly executed in practice. What it offers, rather, is a structured sequence of decisions and actions that manufacturing facilities — particularly SMEs — can follow without requiring the level of technical infrastructure that large-enterprise implementations typically assume (Figure 1).

The Define phase establishes the problem perimeter: which departments, which metrics, which performance gaps constitute the priority targets. This phase relies primarily on the kind of observational and records-based data collection described in Section 2 — identifying, in concrete terms, where the facility's current state diverges from its operational potential. The Measure phase deploys IoT sensors and AI/ML analytics to generate the quantitative baseline: defect rates, cycle times, energy profiles, downtime frequencies. Big data analytics during this phase creates the measurement infrastructure against which all subsequent improvement will be assessed. Ciano et al. (2021) demonstrated the value of establishing this baseline rigorously, noting that improvement claims made without a credible pre-intervention benchmark are essentially unverifiable.

The Analyze phase is where AI earns its place most clearly. Root cause analysis, supported by machine learning outputs from the IoT data stream, allows the team to distinguish between correlated and causal factors — a distinction that manual analysis frequently fails to make reliably. Value Stream Mapping (VSM), applied here as a complementary qualitative tool, visualizes the full production flow and makes visible the hand-off points where delays accumulate. Together, these two analytical modes provide a more complete picture than either offers alone. The Improve phase then moves from diagnosis to intervention — testing solutions identified through the Analyze phase, drawing on Digital Twin simulation where available, and integrating Blockchain-based verification where data integrity across suppliers or subcontractors is a concern (Tripathi et al., 2022). Lean tools — 5S, Six Sigma, Kaizen events — operate continuously through this phase, ensuring that technological improvements are embedded in standardized work practices rather than layered on top of unchanged habits (Figure 2).

The Control phase, finally, ensures that improvements do not simply erode over time — a genuine risk that the continuous improvement literature has documented repeatedly (Glover et al., 2015). Standardized monitoring

Figure 1. Strategic Integration of IoT, AI, and Lean Practices Within the DMAIC Framework for Manufacturing Process Optimization. A conceptual process diagram illustrating how IoT sensor networks, AI-driven analytics, Lean manufacturing tools, and blockchain verification are operationally mapped onto the five sequential phases of the DMAIC (Define, Measure, Analyze, Improve, Control) cycle. Beginning with the Define phase — where cross-functional teams establish key performance objectives and identify priority intervention areas — the framework progresses through the Measure phase (IoT sensor deployment and AI/ML-based data collection establishing the quantitative performance baseline), the Analyze phase (AI-assisted root cause analysis and Value Stream Mapping identifying causal inefficiencies), and the Improve phase (targeted interventions informed by pilot testing, with Digital Twin simulation and blockchain integration supporting secure, sustained process optimization). The Control phase closes the cycle, embedding continuous monitoring protocols, automated alert systems, and Kaizen review cadences to ensure that achieved improvements are maintained and progressively built upon rather than allowed to erode. The integration architecture depicted here reflects the framework applied in the present study and is intended to be adaptable to a range of SME manufacturing contexts without requiring full simultaneous deployment of all components. Adapted from Tripathi et al. (2022).

Table 3. Anomaly Classification Matrix and Technology-Matched Intervention Recommendations for the Shop Floor Management System (SFMS). A structured mapping of the five principal anomaly categories identified through the combined analysis of IoT sensor data, production records, and qualitative supervisory discussions, presented alongside specific technology-based intervention recommendations for each category. The five categories — machinery, manpower, layout, quality, and operational constraints — were derived inductively from the diagnostic phase findings and reflect the distinct causal mechanisms driving inefficiency across the facility's departments. Machinery anomalies, including elevated vibration, thermal irregularities, and excess energy consumption, are matched to online real-time monitoring solutions. Manpower inefficiencies, such as operator disengagement and uneven task distribution, are addressed through feedback-driven work allocation systems informed by historical production data. Layout anomalies are assigned AI-assisted path optimization and online control systems. Quality deficiencies — data errors and missing components — are paired with smart sensor deployment and continuous monitoring. Operational constraint anomalies, including timing variability and resource unavailability, are matched to online analytical systems and integrated smart devices. This matrix is intended to function as a practical diagnostic reference for facilities undertaking similar integration initiatives.

Figure 2. Prerequisites and Process Flow for Lean Six Sigma and Advanced Technology Integration in Manufacturing. A sequential flow diagram depicting the preparatory and implementation stages required before and during the integration of Lean Six Sigma (LSS) with Industry 4.0 technologies — specifically IoT sensor networks, AI analytics platforms, and blockchain-based data verification. The flow begins with the identification and prioritization of key operational problem areas, proceeds through strategic roadmap development and cross-functional team formation, and advances to a controlled pilot project phase in which the proposed integration is tested in a bounded production area before broader rollout. Subsequent stages address the iterative monitoring and evaluation of pilot outcomes, the data-driven scaling decision, and the institutionalization of continuous improvement through Kaizen cycles. Throughout, the diagram emphasizes that technology integration is not a discrete event but a staged organizational process — one where early investment in Lean discipline and data infrastructure substantially improves the probability that subsequent AI and IoT deployments will deliver and sustain measurable operational gains. The prerequisite structure depicted is consistent with the implementation sequence recommended in the present study and reflects findings from comparable integration initiatives documented in the literature. Adapted from Florescu & Barabas (2022).

Table 4. Summary of Prior Innovation Techniques and Reported Outcomes in IoT, AI, and Lean Manufacturing Integration Research. A synthesized review of 17 published studies reporting implementation outcomes relevant to the strategic integration of IoT, artificial intelligence, Lean manufacturing principles, and Industry 4.0 technologies in manufacturing environments. For each entry, the specific innovation technique or framework proposed is described alongside the principal outcome reported and the corresponding authorship citation. Studies span the period 2015–2023 and encompass a range of manufacturing contexts, scales, and integration approaches — including Smart Manufacturing frameworks (Rahardjo et al., 2023), shop floor material delivery optimization (Tripathi et al., 2022), IoT-based industrial monitoring (Ciano et al., 2021), Lean Smart Manufacturing in the bicycle sector (Li, 2019), augmented reality-enhanced shop floor management (Wang et al., 2020), and digital twin integration (Ciano et al., 2021). Collectively, these studies support the thesis that combined IoT-AI-Lean integration consistently outperforms single-technology approaches — while also revealing the implementation gaps, particularly for SMEs, that motivated the framework proposed in this study. Readers should note that outcome reporting varies across studies in specificity and measurement approach; direct quantitative comparison across rows is not recommended.

Figure 3. DMAIC-Structured Integration of IoT, AI, and Lean Practices for Sustained Operational Excellence. A detailed phase-by-phase depiction of the full DMAIC integration framework as applied in this study, illustrating the specific tools, technologies, and decision points operative at each stage. The Define phase establishes measurable performance targets and assembles the cross-functional implementation team. The Measure phase deploys IoT sensors across priority workstations — particularly in Assembly and Fabrication, where baseline anomaly density was highest (Table 2) — and establishes the quantitative data architecture for ongoing performance tracking. The Analyze phase applies AI-driven pattern recognition and root cause analysis to the sensor data stream, complemented by Value Stream Mapping to visualize end-to-end production flow and identify where delays compound. The Improve phase introduces targeted technology and process interventions — 5S workplace organization, Six Sigma defect reduction, AI-optimized scheduling, and IoT-enabled predictive maintenance — validated first through pilot testing before facility-wide implementation. The Control phase embeds standardized monitoring protocols, automated deviation alerts, and structured Kaizen review cycles to sustain performance gains over time and prevent the improvement erosion documented in comparable long-term studies. The framework is presented as a replicable architecture for SME manufacturers seeking a structured, resource-proportionate pathway into Industry 4.0 integration, with phased entry points designed to accommodate varying levels of existing technical infrastructure. Adapted from Ciano et al. (2021).

protocols, automated alert systems, and periodic Kaizen reviews collectively maintain the gains achieved during the Improve phase and create the conditions for ongoing incremental improvement rather than a one-time step change (Figure 3).

3.5 Evidence from Comparable Implementations

To situate these findings in the broader literature, Table 4 summarizes relevant prior implementations of IoT, AI, and Lean integration across a range of manufacturing contexts. The pattern across these studies is not uniform — different settings, different technologies, different scales of implementation — but several consistent themes emerge (Table 4).

First, the combination of real-time monitoring and Lean process discipline consistently outperforms either approach alone. Rahardjo et al. (2023) found that smart manufacturing frameworks integrating Lean principles achieved efficiency improvements that technology-only implementations did not replicate. Second, the DMAIC structure provides a practical organizing logic that helps facilities avoid the common failure mode of deploying IoT sensors and AI tools without a clear analytical framework for acting on the data they generate — a problem that Hughes et al. (2022) identified as one of the central obstacles to Industry 4.0 adoption in SME contexts (Table 4). Third, and perhaps most practically relevant for facilities resembling the one studied here, scalability matters enormously. Florescu & Barabas (2022) showed that smart manufacturing systems designed with modularity in mind — where IoT coverage and AI analytical depth can be expanded incrementally — achieved higher long-term adoption rates than those requiring full upfront deployment. That finding shaped the implementation strategy proposed in this study from the outset.

4. Discussion

4.1 Opening the Conversation: What the Findings Actually Mean

It would be tempting, having assembled the evidence presented in Section 3, to declare the case straightforward: integrate IoT, AI, and Lean, apply the DMAIC framework, and operational excellence follows. The literature cited throughout this paper leans in that direction. But that reading would be a little too clean, and probably not very useful to the manufacturers who might actually try to implement something like this. The findings here are promising — genuinely so — but they are also context-dependent, organizationally demanding, and hedged with conditions that deserve honest discussion before any conclusions are drawn.

What the baseline analysis revealed, most fundamentally, was not a technology problem. The facility examined in this study was not failing because it lacked sensors or algorithms. It was failing — accumulating 470 minutes of idle time, sustaining 6,800 minutes of aggregate cycle time across five departments, operating without condition monitoring or standardized guidelines (Table 1) — because its management systems had not kept pace with its physical growth. That distinction matters enormously for how the integration framework proposed in this paper should be understood. Technology does not fix organizational dysfunction. What it can do, when deployed thoughtfully and within a structured analytical framework like DMAIC, is make dysfunction visible enough to address (Chiarini & Kumar, 2021; Rodriguez Delgadillo et al., 2022).

4.2 IoT as a Diagnostic Infrastructure, Not Just a Monitoring Tool

The role assigned to IoT in most Industry 4.0 literature is primarily one of monitoring — sensors watch machines, flag anomalies, enable predictive maintenance. That framing is accurate as far as it goes, but it undersells what real-time data visibility actually does to an organization's capacity for self-understanding. Before IoT-based monitoring was introduced in the departments examined here, the relationship between machine behavior and downstream production delay was largely inferential. Supervisors knew that Fabrication tended to produce bottlenecks; they did not know, with any precision, whether the cause was equipment-related or workflow-related or both. The IoT data layer changed that relationship fundamentally.

Wang et al. (2018) made a similar observation in their study of IoT-enabled energy optimization in manufacturing environments: the primary value of sensor networks was not the specific interventions they enabled, but the organizational shift from reactive to anticipatory reasoning that they made possible. This study's findings support that interpretation. The machinery anomalies identified in Table 3 — elevated vibration, thermal irregularities, excess energy consumption — were not, in most cases, new phenomena. They were phenomena that had previously lacked a systematic recording mechanism. IoT did not solve these problems; it made them legible for the first time, which is a necessary precondition for solving them (Vlachos et al., 2023).

There is, however, a practical caveat worth stating plainly. IoT infrastructure requires maintenance, calibration, and — critically — interpretation. Raw sensor data is not insight. The translation from data stream to actionable decision requires either skilled personnel capable of reading the outputs or AI analytical tools capable of doing so automatically. In the facility studied here, neither was fully in place at the outset. This is, it should be acknowledged, a limitation that the framework proposed in Figure 1 addresses only partially. The Define and Measure phases of the DMAIC cycle assume that the data being collected is reliable and sufficiently granular; ensuring that those conditions are met in a real facility, particularly an SME with limited technical staff, is harder than the framework diagram suggests (Moeuf et al., 2018).

4.3 Artificial Intelligence: Promise and Practical Constraint

AI's contribution to the integration framework operates at a different level than IoT's. Where IoT generates data, AI interprets it — running pattern recognition across the production data stream to identify relationships that human analysts would likely miss or misattribute. In the Analyze phase of the DMAIC cycle, this capacity for root cause disambiguation proved particularly valuable. The distinction between correlated and causal factors in a complex production environment is not trivial; interventions aimed at correlates rather than causes are a well-documented source of failed improvement initiatives (Trakadas et al., 2020).

Trakadas et al. (2020) demonstrated this in their AI-IoT collaboration framework for industrial manufacturing: facilities that used AI-driven root cause analysis achieved more durable performance improvements than those relying on conventional statistical methods alone, precisely because AI could account for interaction effects across multiple variables simultaneously. The Assembly department's problems, for instance (Table 2), were not simply ergonomic, or simply planning-related, or simply a function of workstation congestion. They were a compound of all three, interacting in ways that individual interventions would address only partially. AI pattern recognition, in principle, surfaces exactly these compound relationships.

In practice — and this is where the findings invite some caution — the AI systems described in the framework require training data that reflects the specific operational context of the facility in question. Generic machine learning models calibrated on large-enterprise data may not transfer cleanly to an SME producing skid steer loaders across 25 workstations. Rahardjo et al. (2023) raised this concern in their smart manufacturing study, noting that the adaptation of AI tools to facility-specific conditions is itself a non-trivial undertaking that requires both technical expertise and a period of model validation that many SMEs are not resourced to sustain. This is not a reason to abandon AI integration; it is a reason to sequence it carefully, beginning with interpretable, lower-complexity analytical tools before progressing to more sophisticated machine learning applications.

4.4 Lean Practices: The Underappreciated Anchor

Of the three components in this study's integration framework, Lean manufacturing tends to receive the least attention in Industry 4.0 discussions — perhaps because it predates the digital revolution by several decades and lacks the novelty that IoT and AI carry. That relative neglect is, in this researcher's view, a mistake. The findings here suggest that Lean practices are not merely compatible with IoT and AI integration; they are arguably the precondition for it working sustainably.

The reasoning is this: IoT generates data and AI interprets it, but neither does anything about the underlying process unless the process itself is organized, standardized, and maintained in a state that makes improvement legible. 5S — Sort, Set in order, Shine, Standardize, Sustain — creates the baseline organizational discipline without which sensor data is difficult to interpret meaningfully. A workstation that is consistently cluttered, where tools are not in designated positions and workflow sequences vary by operator, produces process noise that obscures the signal that IoT sensors are trying to detect. Anosike et al. (2021) made this argument explicitly in their study of Lean-IoT integration, finding that facilities which implemented Lean discipline before deploying IoT infrastructure achieved significantly cleaner data quality and more reliable predictive maintenance outcomes than those that deployed IoT first.

Six Sigma's contribution is complementary but distinct. Where 5S addresses organizational discipline, Six Sigma addresses process variability — the kind of variability that produced 330 minutes of downtime in the Painting department (Table 2) and 165 minutes in Quality Inspection. The DMAIC framework itself is a Six Sigma construct, which is why the integration proposed in Figure 3 uses it as the organizing spine rather than as one tool among many. Tripathi et al. (2022), in their shop floor optimization study, demonstrated that DMAIC provided the most coherent structure for sequencing Lean and Industry 4.0 interventions, because its inherent logic — define the problem before measuring it, measure before analyzing, analyze before improving — prevents the common failure mode of implementing solutions before the problem is properly understood.

Kaizen, the third Lean pillar applied in this study, operates at the longest time horizon. Where 5S and Six Sigma address the current state, Kaizen institutionalizes the expectation of continuous improvement — the idea that the optimized state achieved at the end of an improvement cycle is not a destination but a new baseline from which the next cycle begins. Glover et al. (2015) documented the organizational conditions under which Kaizen improvements tend to sustain versus erode, finding that leadership commitment and structured review cycles were more predictive of long-term success than the specific improvement tools used. That finding has direct implications for the Control phase of the framework proposed in Figure 3, and it is worth flagging here: the framework will not sustain itself. It requires ongoing managerial attention of a kind that cannot be automated away.

4.5 The DMAIC Framework as Integration Architecture

What, exactly, does the DMAIC framework add to an integration of IoT, AI, and Lean that the individual technologies do not provide on their own? The short answer is sequencing discipline — a property that is easy to undervalue until one watches a technology integration initiative collapse because solutions were deployed before problems were properly defined, or because improvements were made without the monitoring infrastructure to verify whether they worked.

The departmental data in Table 2 illustrates the need for this sequencing directly. Assembly's 2,895-minute cycle time and 240 minutes of downtime (Table 2) involve at least four distinct contributing factors — ergonomics, work allocation, defect rates, and spatial congestion. Intervening on any single factor without understanding its relative contribution risks not only wasting resources on a low-leverage intervention, but potentially destabilizing interactions with other factors. The Analyze phase of DMAIC, supported by AI root cause analysis and Value Stream Mapping, is specifically designed to prevent this. Ciano et al. (2021) found, in their one-to-one analysis of Industry 4.0 technologies and Lean techniques, that the primary source of failed manufacturing improvement initiatives was premature intervention — acting on identified symptoms before their causes were sufficiently understood.

The framework's Control phase deserves particular attention, because it is where most real-world improvement initiatives eventually falter. Monitoring standards erode, review cadences slip, and the organizational attention that sustained the improvement effort gradually redirects to other priorities. Hughes et al. (2022) documented this pattern across a range of manufacturing firms, noting that the long-term sustainability of Industry 4.0 gains depends less on the sophistication of the technology deployed and more on the robustness of the governance mechanisms built around it. The continuous monitoring systems and Kaizen review cycles embedded in Figure 3 are the framework's answer to this problem — though, again, they require deliberate organizational commitment to function as intended.

4.6 Implications for Small and Medium-Sized Enterprises

The facility studied here is, in most respects, representative of the SME manufacturing context: semi-automated, resource-constrained, operating with informal planning systems, and lacking the dedicated technical staff that large enterprises can deploy to manage complex technology integration projects. This context was chosen deliberately, because the existing literature — however robust in its support for IoT-AI-Lean integration — has developed primarily in large-enterprise settings where cost barriers, skill gaps, and implementation complexity are less acute (Vlachos et al., 2023).

The findings suggest several practical accommodations for SME implementation that the broader literature does not always make explicit. First, sequencing matters more than completeness. SMEs do not need to implement the full framework simultaneously; a phased approach — beginning with Lean discipline and basic IoT monitoring, adding AI analytics as data quality and internal capability develop — is both more feasible and, arguably, more likely to produce durable results than a comprehensive rollout that exceeds the organization's capacity to absorb change. Mendes et al. (2023) reached a similar conclusion in their synergistic Lean-Industry 4.0 model, finding that incremental, capability-building implementation outperformed ambitious simultaneous deployment in SME contexts. Second, the cost barrier, while real, is declining. The IoT sensor hardware and cloud-based AI analytics platforms available in 2023 represent a substantially lower entry cost than equivalents from five years earlier, and open-source machine learning frameworks have reduced the software cost component considerably. What remains expensive is not the technology itself but the human expertise needed to deploy and interpret it — a gap that training investment and, potentially, industry consortium models can help address (Florescu & Barabas, 2022).

Third, and perhaps most importantly, SMEs should resist the temptation to measure success primarily in technology adoption terms. The question is not whether IoT sensors are installed or an AI platform is running; it is whether downtime is declining, whether cycle times are shortening, whether the specific anomalies identified in Table 3 are being systematically addressed. Technology is a means. Operational performance is the end. Keeping that distinction clear — and building performance metrics into the framework from the Define phase onward — is, in this study's judgment, the single most important determinant of whether an integration initiative of this kind delivers lasting value.

4.7 Limitations

No discussion of this kind would be complete without an honest account of what this study cannot claim. The findings are drawn from a single manufacturing facility producing a specific product type; their generalizability to other industries, product categories, or national manufacturing contexts is plausible but not demonstrated. The framework proposed is conceptual and DMAIC-structured, not the output of a longitudinal implementation study with pre- and post-intervention performance data — a limitation that future empirical work should address directly. The AI and IoT tools referenced throughout are described at a functional level; the specific platforms, algorithms, and calibration parameters used were not reported in sufficient detail for direct replication, which the authors acknowledge as a gap in the reproducibility of the methodology. And the human factors dimension — operator resistance to monitoring systems, supervisor skepticism toward AI-generated recommendations, the organizational culture questions that Lean implementation consistently raises — received less analytical attention than the technical integration questions, despite being, in many real-world implementations, the more consequential obstacle (Lin et al., 2015).

These limitations are not reasons to discount the framework. They are reasons to treat it as what it is: a structured, evidence-informed proposal for how IoT, AI, and Lean practices might be integrated in SME manufacturing settings, and a starting point for the more granular empirical work that would allow its claims to be tested, refined, and — where necessary — corrected.

5. Conclusion

Manufacturing improvement, at its core, has always been about closing the gap between what a facility is capable of and what it is actually delivering. This study found that gap to be substantial — and, more importantly, addressable.

The integration of IoT, AI, and Lean practices within the DMAIC framework is not a theoretical convenience. It reflects a genuine complementarity: IoT makes production behavior visible, AI makes it interpretable, and Lean makes the organizational ground stable enough for improvements to take hold and last. None of the three works as well alone. Together, structured within a clear analytical sequence, they offer SME manufacturers something the literature has not always provided — a practical, phased pathway toward operational excellence that does not require enterprise-scale infrastructure to begin.

What remains to be done is considerable. Longitudinal empirical validation, facility-specific AI calibration, and serious attention to the human and cultural dimensions of implementation are all necessary next steps. But the direction, this study suggests, is sound.

Author Contributions

Md F. A. Bhuiyan: Conceptualization, Methodology, Data Curation, Formal Analysis, Investigation, Visualization, Writing – Original Draft, Writing – Review & Editing, Project Administration.

Acknowledgements

The author sincerely thanks the management, supervisory staff, and shop floor personnel of the skid steer loader manufacturing facility for their cooperation and openness during the observation, data collection, and discussion phases of this study. Their willingness to share production records and operational insights was indispensable to the completion of this research.

Conflict of Interest

The author declares no conflict of interest.

Financial Disclosure

Md F. A. Bhuiyan confirms that this research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors, and that no financial relationship or sponsorship of any kind influenced the study design, data collection, analysis, interpretation, or decision to publish.

References