Sensing, Analytics, and Trust: An Integrated AI-IoT-Blockchain Framework for Cleaner Production

Abstract

The integration of artificial intelligence (AI), the Internet of Things (IoT), and blockchain may provide a viable approach to tackle persistent operational and informational challenges in cleaner production. This conceptual review synthesizes existing literature and presents an integrated AI-IoT-blockchain framework mapped across the four sequential stages of cleaner production: source reduction, process control, end-of-pipe treatment and recycling, and full-chain traceability. The literature indicates that IoT enables real-time sensing, AI drives predictive and prescriptive analytics, and blockchain ensures tamper-proof record-keeping and stakeholder trust. Together, these technologies may help address long-standing barriers including fragmented data, delayed responses, and a lack of verifiability. Despite challenges such as high costs, technical fragmentation, and organizational resistance, several emerging strategies have been proposed in the literature to address these challenges. These include modular deployment, federated learning, permissioned blockchains, and regulatory sandboxes. The framework’s underlying architecture appears transferable across sectors, subject to industry-specific adaptation, supporting sustainable manufacturing, the circular economy, and low-carbon development.

1. Introduction

Cleaner Production, according to the United Nations Environment Programme (UNEP) [1], is defined as the continuous application of an integrated preventive environmental strategy to processes, products, and services. Its goal is to increase overall efficiency and reduce risks to humans and the environment [2,3]. This definition reflects a full-process management philosophy that fundamentally distinguishes cleaner production from conventional end-of-pipe approaches [4]. Its core objectives of reducing costs and improving efficiency are achieved primarily through preventive actions rather than reactive remediation. In practice, an integrated management system that embeds cleaner production practices can promote resource efficiency, pollution prevention, life-cycle transparency, and regulatory compliance. To structure the analysis in this review, we organize this full-process management philosophy into four sequential stages: source reduction, process control, end-of-pipe treatment and recycling, and full-chain traceability. These stages are proposed as an organizing framework for this review and are not intended to replace the established UNEP taxonomy. Collectively, they cover the entire production chain, from what enters the system and how it is processed, to what remains unavoidable and how all activities are documented and verified. The specific tasks of these four stages are illustrated in Figure 1.

With the growing demand for sustainable development, the concept of cleaner production has gained widespread recognition. However, the implementation of traditional cleaner production approaches has been reported to encounter multiple barriers [5]. According to Vieira and Amaral (2016) [6], these barriers can be classified into internal and external categories. Internal barriers include organizational culture, poor communication, economic constraints, and low education attainment. External barriers include insufficient public awareness, inadequate economic incentives, and limitations of existing methodologies. While some of these barriers are systemic, others are technical in nature. Although these macro-level barriers are critical, they largely lie beyond the scope of technological intervention. This review does not attempt to address all of them. Instead, we focus on a subset of operational and information-related problems that emerge within the production process itself. As discussed in the literature, these problems are potentially addressable through digital technologies. Specifically, these problems manifest across four cleaner production stages. Source reduction remains heavily experience-driven, lacking real-time, quantitative tools to evaluate alternatives, which leads to suboptimal substitution decisions and operational resistance. Process control relies on manual sampling and offline laboratory analysis, resulting in delayed interventions and economic inefficiencies. End-of-pipe treatment and recycling operations struggle with high contamination rates and low sorting accuracy. They are deprived of real-time feedback from upstream stages, which ultimately undermines closed-loop resource recovery. Full-chain traceability and trusted management suffer from fragmented and tamper-prone record-keeping systems. These systems create data silos, erode stakeholder trust, and complicate environmental certification. Taken together, these challenges result in fragmented information, delayed responses, untrustworthy data, and ultimately a failure to achieve the full potential of cleaner production.

Recent advances in digital technologies [7], specifically the Internet of Things (IoT), Artificial Intelligence (AI), and blockchain, offer promising solutions to these long-standing challenges in cleaner production. IoT enables real-time, continuous data collection through interconnected sensors, replacing manual, intermittent, and labor-intensive monitoring. This directly addresses the problem of delayed intervention and provides reliable, high-frequency data on material flows and operational conditions [8]. AI brings data-driven decision-making, predictive analytics, and automation capabilities. It tackles the limitations of experience-based judgment, offline analysis, and slow response times, enabling proactive rather than reactive management [9]. Blockchain provides decentralized and tamper-proof record-keeping. This directly addresses the issues of fragmented data silos, lack of stakeholder trust, and fraud risks in environmental certification [10]. Importantly, these technologies do not merely improve individual stages in isolation. Rather, they have the potential to create the foundation for an integrated, transparent, and intelligent system in which cleaner production principles can be reliably executed from source to end-of-pipe. In the existing literature, most current studies concentrate on single-technology applications, such as IoT-based monitoring systems, AI-driven optimization models, or blockchain-enabled traceability platforms. A smaller subset of research has explored dual-technology combinations, including AI-enabled IoT for smart process control and IoT-blockchain integration for secure data logging [11]. More recently, several studies have begun to examine the integration of all three technologies, namely AI, IoT, and blockchain, in specific contexts such as environmental management [12], food traceability [13], and aquaculture [14]. However, a systematic, stage-by-stage synthesis of how these three digital enablers can work together across the full spectrum of cleaner production, from source reduction to end-of-pipe treatment and recycling and across full-chain traceability, remains limited. This review aims to organize existing evidence into a unified three-layer framework and map it onto the four cleaner production stages.

Given this rationale, this review contributes to the literature by synthesizing existing applications of AI, IoT, and blockchain across the four cleaner production stages, and by organizing them under a unified conceptual framework. Specifically, this review: (1) maps existing applications of AI, IoT, and blockchain across the four cleaner production stages, covering single-, dual-, and triple-technology use cases; (2) introduces a three-layer (sensing–analytics–trust) architecture as an organizing logic for understanding how these technologies complement each other; and (3) outlines potential cross-sectoral application pathways, using a comparative table to illustrate how the framework may be adapted to industries such as energy, agriculture, and manufacturing.

To anchor the analysis, a structured literature search was conducted in Web of Science, Scopus, and IEEE Xplore for peer-reviewed articles published between 2016 and 2026, using combinations of keywords related to cleaner production, AI, IoT, and blockchain. Additional relevant studies were identified through backward snowballing from key reviews. Studies were considered for inclusion if they addressed cleaner production applications of AI, IoT, or blockchain in production, supply-chain, or environmental management contexts. The evidence base of the studies presented in the tables was then classified according to the taxonomy defined alongside the tables.

2. Core Technologies and Their Stage-Specific Applications in Cleaner Production

2.1. Source Reduction

Source reduction is the first stage of cleaner production. Its core objective is to minimize or substitute toxic, polluting, and high-environmental-impact raw materials at the source. This must be accomplished before these materials enter the production system. In doing so, pollution and waste are eliminated at their origin. Unlike end-of-pipe treatment, source reduction emphasizes the principle that “prevention is better than remediation”. If problems are eliminated at the source, the subsequent stages face substantially less pressure.

However, in traditional practices, source reduction faces deep operational and information-related challenges. Traditionally, raw material selection and optimization have relied heavily on operator experience and heuristic decision-making, for example, manual visual inspection, rule-based acceptance criteria [15], or empirical knowledge to adjust formulations [16]. This experience-dependent approach suffers from inherent limitations: subjectivity and inconsistency; limited resolution for subtle quality variations; poor scalability and transferability across operators or facilities; and an inability to balance cost, performance, and environmental impact simultaneously, often resorting to simplified rules of thumb.

Simply adding more sensors or storing more data cannot overcome these limitations. Instead, they call for a fundamentally different approach, one capable of learning from historical patterns, processing high-dimensional information, and consistently optimizing across multiple objectives. This is precisely the challenge for which artificial intelligence is uniquely suited [17]. In the context of source reduction, AI has been shown to address these limitations through three complementary mechanisms [18]. These mechanisms follow the natural sequence of raw material handling: quality inspection, quality grading, and material formulation design.

In terms of quality inspection, Tao et al. (2026) developed an improved YOLO11-based object detection model [19], named METW-YOLO, to detect visible mold on oil-bearing raw materials such as peanuts, corn, and soybeans. Under simulated conveying conditions, the model achieved 95.4% precision and 95.8% recall [19]. By screening out deteriorated batches before they enter processing, AI prevents downstream quality losses and resource waste at the intake stage.

Once a batch passes the initial inspection, the process extends to quality grading. Nuzzi et al. (2026) applied laser speckle pattern imaging combined with machine learning to classify raw cow milk samples, achieving 95% classification accuracy [20]. Similarly, Liu et al. (2016) developed a Weibull distribution-based statistical modeling approach with a semi-supervised classifier for rice quality grading [21]. This approach achieved 98% accuracy across 6250 samples from five rice varieties. At a deeper level, quality grading extends to detecting abnormal or hazardous conditions that human inspection cannot identify. Aqeel et al. (2024) developed a hyperspectral imaging system with convolutional neural networks to detect chemical adulteration in milk with formalin, boric acid, glucose, and salicylic acid [22]. The system achieved 97% classification accuracy across 49 classes of adulteration levels. This capability prevents harmful substances from entering the food chain and directs raw materials to their appropriate end-uses.

With inspection and grading providing reliable data on raw material quality, AI then enables formulation design. Unlike inspection and grading, which focus on incoming raw materials, formulation design proactively optimizes ingredient combinations. This optimization targets quality, nutrition, and sustainability. Yakoubi (2024) integrated molecular docking with machine learning to optimize protein-ligand interactions in plant-based cheese [23], achieving 71.59% prediction accuracy via SVM. The resulting formulation contained 85% higher protein content than existing commercial products [23]. This represents an exploratory formulation optimization result rather than a demonstrated source-reduction success.

Beyond the food industry, these three AI-driven mechanisms have been widely applied in other sectors. In the materials industry, Ferro et al. (2026) demonstrated the formulation design mechanism by developing a deep neural network and genetic algorithm framework [24]. This framework was used to optimize ductile cast iron composition, reducing critical raw material dependency while maintaining mechanical properties. Their model achieved an R² > 0.98 for strength prediction and reduced the Alloy Critical Index by over 30% across three case studies [24]. In the construction industry, Shi et al. (2025) demonstrated the quality inspection and grading mechanisms by developing an AI-powered visual grading system for reclaimed lumber [25]. This system achieved 99–100% precision in detecting surface defects and reduced dimensional measurement errors from approximately 7% to below 2.5% through image calibration.

Notably, these AI applications have improved source reduction decisions. However, the trust issue between suppliers and processors remains unresolved. On this basis, blockchain has been suggested as a potential means to fill this gap by providing immutable records of raw material provenance and quality certifications.

2.2. Process Control

Process control is the second stage of cleaner production. Its core objective is to optimize production operations through real-time monitoring, parameter adjustment, and predictive decision-making, thereby reducing waste and emissions during manufacturing. Unlike reactive troubleshooting, process control emphasizes proactive intervention. When deviations are detected and corrected in real time, quality loss and resource waste can be minimized before they occur.

However, in traditional practices, process control faces persistent challenges. Traditionally, it has relied heavily on fixed operational set-points, periodic manual sampling, and delayed feedback control [26]. Operators adjust parameters based on predefined schedules or experience, rather than on real-time data-driven decision-making. This approach suffers from several inherent limitations. For example, it is labor-intensive and inefficient, and cannot respond rapidly to fluctuations in raw material quality or processing conditions; it fails to detect subtle anomalies before they escalate into defects or equipment failures; and it often resorts to conservative parameter settings such as excessive chemical dosing or energy input to ensure product quality within specification, leading to resource waste and unnecessary emissions.

To address these limitations, recent studies have turned to AI-driven approaches. AI has demonstrated significant value by learning patterns from historical data, enabling predictive modeling that outperforms traditional methods. However, AI alone relies on offline data and cannot perceive real-time changes in production conditions, causing its predictions to drift from actual conditions. This limitation is overcome by integrating AI with the IoT [27]. IoT sensors continuously gather real-time operational data, such as temperature, pressure, pH, and flow rate, and feed them to AI models. AI then processes these live data streams to forecast trends, detect anomalies, and recommend optimal adjustments. This AI + IoT integrated framework [28,29], where IoT provides real-time sensing and AI provides cognitive analytics, has been applied across multiple sectors, forming the foundation for intelligent process control.

In the food industry, AI alone has demonstrated significant potential for process optimization [30,31]. Horiuchi et al. (2004) applied an artificial neural network (ANN) to predict acidification time in cheese fermentation [32]. This model reduced the mean absolute error to approximately 7 min (3% of total process time) with a correlation coefficient of 0.92, significantly outperforming conventional methods that rely on manual expertise and traditional kinetic modeling [32]. This purely AI-driven approach, however, relies on offline data and cannot adapt to real-time process variations. Building on this, the integration of IoT sensors enables real-time monitoring and dynamic control. An AI-driven real-time monitoring and predictive control system was developed by Ray et al. (2026), using IoT sensors to measure electrical conductivity, pH, temperature, and total dissolved solids [33]. Among the six models evaluated (FFNN, GRU, GPR, LSTM, SVR, and Transformer), the FFNN model achieved the highest predictive accuracy, with an R² of 0.999 on validation data and 0.990 on held-out test data, enabling operators to predict yogurt fermentation outcomes and decide the optimal termination point in real time [33]. It should be noted, however, that such high R² values are uncommon in biological systems and may indicate some degree of overfitting, as the original authors themselves acknowledged that deeper architectures tended to overfit under data-limited conditions. Extending this further, computer vision, IoT sensors, and machine learning can be integrated with closed-loop PID control to automatically regulate temperature and stirring speed in traditional fermented dairy production, achieving an R² > 0.95 with a fully integrated hybrid feature set [34].

This technological integration and system transformation are not confined to the food industry. Rather, the combination of AI and IoT has been widely deployed in diverse sectors such as manufacturing [35], agriculture [36], and renewable energy. This deployment has given rise to closed-loop control systems that continuously sense, analyze, and adjust process parameters in real time [37,38].

In manufacturing, real-time defect detection and predictive maintenance replace traditional periodic inspections and reactive repairs. In lithium-ion battery manufacturing, IoT sensors continuously collect production data. Meanwhile, AI-driven systems also play a key role. They enable real-time defect detection through computer vision, predict equipment failures through machine learning algorithms, and simulate production processes through digital twins. In a widely cited industrial case study from CATL’s Liyang facility, this integration collectively reduces defect rates from parts per million to parts per billion, while also lowering energy consumption and increasing output [39]. However, these figures are derived from a single case study and may not be representative of typical industry performance. For industrial machinery, vibration, temperature, and rotational speed sensors feed data into traditional machine learning models. An IoT-enabled predictive maintenance system based on this architecture achieves a 50% reduction in unplanned downtime in automotive manufacturing and an 11.25% increase in overall production uptime [40].

A similar logic applies in agriculture [41], where environmental monitoring and predictive forecasting optimize resource allocation. A distributed IoT sensor network monitors soil moisture, nutrient levels, microclimate, and crop health in real time. The data streams are processed by an ensemble learning-based predictive model for crop yield forecasting, accounting for spatial and temporal variability to generate high-resolution predictions. These predictions enable optimized irrigation, fertilization, early disease detection, and informed market decisions [42].

In renewable energy, the same sensing–analysis–control loop supports real-time energy optimization that balances supply and demand. In solar energy, an IoT-enabled storage system collects real-time data on voltage, current, temperature, and irradiance, along with weather parameters such as humidity, wind speed, and cloudiness. AI models process these data for predictive energy modeling, fault detection, and load adjustment. By combining IoT-based real-time monitoring with AI-driven predictive analytics and adaptive control, this dual-technology approach achieves 15% to 20% efficiency gains compared to traditional static solar energy systems [43].

However, the process data generated and analyzed by IoT and AI are typically stored in centralized systems that lack tamper-proof assurance. Integrating blockchain would complement this dual-technology setup by creating an immutable audit trail of parameter adjustments and deviation logs, supporting verifiable compliance and root-cause analysis.

2.3. End-of-Pipe Treatment and Recycling

End-of-pipe treatment refers to the processing of pollutants and wastes after they have been generated and before they are discharged into the environment [44]. In the organizational framework of this review, it is included as the final production-chain stage, acknowledging that in practice not all waste can be eliminated at the source. It serves as the final line of defense after source reduction and process control. Unlike source reduction, which prevents waste at the source, or process control, which minimizes waste during production, end-of-pipe treatment addresses the unavoidable residues that remain [45]. End-of-pipe treatment encompasses three major categories based on the environmental medium: wastewater treatment, waste gas treatment, and solid waste treatment and disposal. Its core objectives are threefold: (1) reducing the toxicity and volume of pollutants before discharge; (2) recovering valuable resources from waste streams; and (3) closing material loops to achieve circular utilization.

This emphasis on closing material loops naturally leads to a circular approach, which manifests in two complementary directions. On the one hand, wastes generated during production, such as water and sludge from wastewater treatment, can be recycled back into the front-end of the same process [46]. On the other hand, one industry’s waste can become another industry’s resource [47,48]. For instance, Bishayee et al. (2023) demonstrated that crushed waste teapot fragments can be effectively used as an adsorbent for the end-of-pipe treatment of secondary-treated coking wastewater, removing fluoride, cyanide, phenol, ammonia nitrogen, and nitrate from industrial effluent [49]. When combined with circular economy principles, end-of-pipe treatment thus transforms waste from an endpoint problem into a resource opportunity. This can occur either within the same production loop or across different industrial sectors.

In traditional practice, conventional end-of-pipe treatment and resource recovery systems faced multiple bottlenecks, including outdated monitoring methods with delayed responses and an inability to achieve real-time optimization; a lack of synergy among water, air, and solid treatment media; the failure to close-loop recycle secondary pollutants; and low recovery efficiency and high costs for valuable components. The integration of AI and IoT addresses these limitations by enabling real-time monitoring, predictive control, and automated sorting [50,51], essentially applying the same sensing–analysis–control loop to waste streams and treatment operations.

The traditional challenge is particularly acute for high-strength industrial waste streams. In the food industry, cheese whey, a by-product of cheese manufacturing, exemplifies this problem [52]: over 190 million tonnes are produced annually (equivalent to approximately 9 tonnes of whey per tonne of cheese produced [53], based on FAOSTAT global cheese production data [54]), yet only 50% is processed into environmentally safe products [55]. When discharged untreated, whey causes severe aquatic pollution due to its extremely high organic loads (COD in the range of 0.8–102.0 g/L and BOD in the range of 0.6–60.0 g/L) [56]. The decomposition of these organic pollutants in receiving water bodies depletes dissolved oxygen, triggers algal blooms, causes fish mortality, and facilitates the propagation of pathogenic microorganisms. Traditional biological treatments, such as anaerobic digestion, convert whey into biogas and biochar but suffer from pH instability, nutrient imbalances, and a lack of real-time process control, leading to suboptimal performance and high operational costs. Yu and Li (2025) addressed these limitations by integrating AI into industrial wastewater treatment [57], employing a convolutional neural network to monitor pollutant levels and reinforcement learning to optimize aeration, chemical dosing, and sludge retention time. This AI-driven approach reduced energy consumption by 7.02%, greenhouse gas emissions by 18%, and total nitrogen concentrations by 15%, while also lowering operating costs by 10% and extending equipment lifespan by 20% [57].

Looking beyond AI alone, Yu and Li (2025) further noted that the integration of multiple smart technologies [57], including IoT, AI, and blockchain, represents a critical direction for enhancing transparency, traceability, and automation in wastewater management, ultimately supporting long-term sustainability goals. This multi-technology integration has since been extended to other areas of end-of-pipe treatment and circular utilization.

At the AI-IoT level in the solid waste treatment domain, Arun (2025) proposed a waste material recovery framework integrating IoT sensors with deep learning for automated waste classification and sorting [58], achieving 98.23% recovery efficiency, 97.22% classification accuracy, and a 30% reduction in operational costs, based on a dataset of 15,000 images across 30 waste categories, substantially outperforming traditional methods. As these figures were obtained under controlled experimental conditions, their direct transferability to heterogeneous real-world waste streams warrants further investigation, particularly given the considerable variability in composition and contamination levels. Building on this, Aremu et al. (2026) added blockchain to create a fully integrated AI-IoT-blockchain framework for coal gangue recycling [59]. In their implementation, IoT sensors collect real-time material data; an AI model (ResNet-50 CNN) achieves 95% ± 0.5% classification accuracy; and a blockchain ledger ensures immutable traceability of transactions and carbon emissions. This integration demonstrates how the three technologies can work together in end-of-pipe treatment, where blockchain ensures that recycling transactions, carbon credits, and compliance records remain fully auditable.

2.4. Full-Chain Traceability

Full-chain traceability refers to the ability to track a product and its components across the entire supply chain. This covers the full journey from raw material sourcing through production, processing, transportation, and storage to final consumption, while also encompassing the flow of materials, information, and finances from origin to delivery [60,61]. In the cleaner production framework, traceability serves as the information backbone that connects source reduction, process control, and end-of-pipe treatment into a coherent, accountable system. Its core objectives are threefold: (1) enhancing accountability and root-cause analysis; (2) supporting regulatory and sustainability compliance; (3) building stakeholder and consumer trust through end-to-end transparency. Liu and Nie (2025) further noted that in distributed manufacturing scenarios [62], full-chain traceability plays a critical supporting role in optimizing production cycles, reducing energy consumption, and achieving circular economy goals.

Traditional traceability systems face multiple barriers. They rely on paper-based records, fragmented databases, and manual data entry, leading to incomplete, inconsistent, and easily manipulable data [14,63]. Different stakeholders operate in information silos with limited interoperability [60]. When a quality or safety incident occurs, tracing the non-conformance or defect source can take days or weeks. During this time, problematic products or materials continue to flow through the supply chain, exacerbating economic losses and potential liability risks. Furthermore, the “black-box” nature of AI models can create distrust among supply chain partners. As Duong pointed out, many product attributes such as organic production are credence qualities that cannot be fully verified even after consumption or use, and traditional traceability methods struggle to resolve such information asymmetry [64]. These limitations create an urgent need for blockchain technology.

Blockchain technology is a decentralized, distributed ledger that records transactions in cryptographically linked blocks [60]. It offers four core characteristics. First, immutability ensures that once data is recorded, it cannot be altered, guaranteeing tamper-proof records [14,62]. Second, transparency allows authorized participants to access a single shared version of the truth [14]. Third, decentralization eliminates the need for a central authority, as the peer-to-peer network maintains consensus [62]. Fourth, smart contracts are self-executing codes that automate verification and payment processes when predefined conditions are met [65]. These characteristics make blockchain particularly well-suited for supply chains where trust and verifiability are paramount, nowhere more so than in the food industry [66].

Saurabh and Dey (2021) developed a modular blockchain-enabled information system for the grape wine supply chain [65], integrating IoT sensors, RFID tags, and smart contracts to achieve disintermediation, traceability, and trust. Their conjoint analysis identified traceability as the second most important factor influencing blockchain adoption, with a relative importance exceeding that of price and trust. Duong’s empirical study further found that blockchain-enabled traceability significantly enhanced consumer trust in organic food retailers and producers, thereby increasing purchase intentions [64].

Seafood products face unique traceability challenges due to complex, fragmented supply chains involving multiple intermediaries, frequent custody transfers, and high risks of fraud and spoilage [67]. IoT sensors can monitor temperature, humidity, and handling conditions along the cold chain, while blockchain records each custody transfer from catch to distribution, creating an immutable audit trail. Pilot implementations in Southeast Asia have demonstrated that this integration reduces fraudulent activities by 30%. While encouraging, this finding is derived from a limited number of pilot sites, and the magnitude of fraud reduction is likely to be context-dependent, varying with supply chain structure, regulatory frameworks, and stakeholder engagement levels. Furthermore, blockchain-enabled shipping containers have substantially reduced spoilage of temperature-sensitive seafood, directly addressing one of the industry’s most costly problems [14].

Beyond food safety, blockchain technology has been widely adopted across other industries to ensure supply chain transparency and traceability. In agriculture, blockchain can record organic certification, fair trade compliance, and pesticide application data, enhancing consumer trust in product authenticity [68]. In construction, blockchain enables traceability of building materials and sustainability certification of critical raw materials [69,70]. In manufacturing [71], blockchain supports circular economy tracking and traceability of recycled content [62]. In transportation and logistics [72], the integration of blockchain with IoT and AI enables real-time and tamper-proof monitoring of shipment location, temperature, and humidity, significantly improving supply chain transparency and efficiency [73]. In healthcare, blockchain has been applied to track medicine origins, ensure authenticity, and reduce counterfeit drugs. By providing a decentralized and immutable dataset, blockchain strengthens transparency and trust in pharmaceutical supply chains. Industry analyses indicate that enhanced traceability can mitigate the economic burden associated with counterfeit drug circulation [74,75].

Across these diverse sectors, a common value proposition of blockchain can be identified. It creates a shared, immutable, and auditable record of product provenance that builds trust among stakeholders and enables accountable and transparent supply chains.

2.5. An Integrated AI-IoT-Blockchain Framework Based on Three Layers

The preceding sections have examined how AI, IoT, and blockchain, either individually or in pairwise combinations, support cleaner production across four stages: source reduction, process control, end-of-pipe treatment and recycling, and full-chain traceability. Among these, AI drives raw material optimization and predictive control; IoT enables real-time sensing of processes and waste streams; and blockchain secures tamper-proof traceability. However, across these stages, the three technologies remain largely fragmented rather than being integrated into a unified operational logic. This fragmentation limits the synergies that could arise from their coordinated deployment and prevents the formation of a seamless digital thread from source to end-of-pipe.

On this basis, we synthesize the preceding discussion into an integrated AI-IoT-blockchain framework structured around three foundational layers: sensing, analytics, and trust [76]. Rather than merely combining three technologies, this framework assigns each to a distinct but complementary role. IoT serves as the sensing layer, capturing real-time, high-resolution data for acquisition. AI serves as the analytics layer, transforming sensed data into predictions, optimizations, and decisions. Blockchain serves as the trust layer, providing immutable, auditable, and transparent record-keeping. This layered conceptualization is not arbitrary; it is supported by a growing body of literature that has independently examined the synergistic integration of these three technologies across various domains.

Table 1. Representative applications of the AI-IoT-blockchain integrated framework across different domains.

Domain	Sensing Layer	Analytics Layer	Trust Layer	Key Outcome	Evidence Base
Solid waste [59]	IoT sensors for real-time material monitoring	ResNet-50 CNN for automated sorting	Blockchain ledger for transactions and carbon records	Reduced emissions, energy use, and water contamination	Simulation
Aquaculture [14]	IoT-enabled cold-chain monitoring	AI-driven fraud detection	Blockchain-based custody transfer records	Reduced fraud and spoilage	Multi-site deployment/Pilot/Lab validation
Agriculture [78]	IoT sensors, UAVs, edge devices for real-time farm data collection	AI (ML/DL) for predictive analytics, disease detection, resource optimization	Blockchain for data security, traceability, and smart contracts	Enhanced productivity, resource efficiency, and supply chain transparency	Multi-site deployment/Pilot/Simulation
Energy [77]	IoT sensors for feedstock traceability and process monitoring	AI (Random Forest) for preconsensus anomaly detection	Hybrid PoS-PBFT blockchain for compliance MRV	Low-cost digital MRV for SMEs under RED II and CORSIA	Multi-site deployment/Pilot/Simulation
Food [79]	IoT sensors, smart packaging for real-time microbial monitoring	AI (ML/DL) for pathogen detection and risk prediction	Blockchain for immutable safety data records	Enhanced traceability, proactive risk management from reactive to predictive	Lab validation

summarizes how studies in aquaculture [14], solid waste recycling [59], energy [77], agriculture [78], and food [79], have converged toward a similar architectural logic. In each case, IoT provides the sensing function, AI provides the analytical capability, and blockchain provides the trust mechanism. These cross-domain examples illustrate the generalizability of this layered architecture and provide empirical grounding for our proposed framework.

Evidence base categories are defined as follows. Multi-site deployment applies when the study discusses two or more independent physical sites with their actual deployment and application cases. Single case study applies when the study focuses on one specific enterprise or organization with its actual deployment and application case, while other cases are mentioned only as background or for comparison. Pilot applies when the study discusses a pilot project, covering its design, scale, or preliminary results. Simulation applies when the study uses simulation or computational modeling as its core evidence for discussion or analysis. Lab validation applies when the study uses laboratory-based method validation or performance evaluation results as its core evidence for discussion.

A study is classified according to the type or types of evidence it discusses, regardless of whether the evidence originates from the authors’ own work or from cited sources. If a study presents multiple types of evidence, all applicable labels are listed jointly. The same classification criteria are applied in the subsequent tables.

While Table 1 has illustrated the mapping from individual technologies to functional layers, Figure 2 further shows how the three layers collectively span the entire production chain from source reduction through process control and end-of-pipe treatment to full-chain traceability. Importantly, the three layers do not correspond to different stages; rather, they operate simultaneously across all four stages, creating a complete, end-to-end digital thread that connects the entire cleaner production system into a unified, intelligent, and accountable system. This three-layer architecture, which draws on the broader Industry 4.0 literature on layered digital architectures [76], is proposed here as the authors’ own conceptual synthesis of the reviewed applications.

The integrated framework follows a closed-loop logic. IoT sensors continuously collect real-time data on raw material quality, production parameters, waste stream status, and supply chain movements. These data streams are fed into AI models that predict trends, detect anomalies, optimize parameters, and support decision-making [80]. Simultaneously, all critical data and transactions are recorded on an immutable blockchain ledger, enabling verifiable traceability and automating business execution through smart contracts. This closed-loop system bridges data acquisition, intelligent analysis, and trusted record-keeping, with feedback from AI and blockchain continuously informing subsequent sensing and action.

Two representative cases, coal gangue recycling and aquaculture supply chains, illustrate the practical viability of this framework in different industrial contexts. In coal gangue recycling, IoT sensors for real-time material monitoring, an AI-driven ResNet-50 CNN for automated sorting, and a blockchain ledger for transaction and carbon accounting have been integrated. This integrated approach has achieved measurable outcomes, including a 30% reduction in CO₂ emissions, 34% energy savings, a payback period of 3.6 years for a 2000 t/day facility, and an ROI of 22% [59]. Similarly, in aquaculture supply chains, Chandran et al. reported that IoT-enabled cold-chain monitoring, AI-driven fraud detection [14], and blockchain-based custody transfer records collectively reduced fraudulent activities, substantially minimized spoilage of temperature-sensitive seafood, and achieved 25% cost savings in feed operations.

However, the evidence base varies considerably across sectors. For the waste management and aquaculture sectors, the framework has demonstrated implementability, with integrated deployments achieving quantifiable cost savings and operational improvements [14,59]. For the agriculture, food, and energy sectors, full integration across all four cleaner production stages remains at an early validation stage, despite conceptual development supported by proof-of-concept pilots, simulation-based analyses, and systematic literature reviews [77,78,79]. For the chemicals, textiles, and electronics sectors, the framework remains at an exploratory stage, requiring further empirical research to establish the technical and economic case for implementation. The transferability of the framework to these sectors represents a promising direction for future research, contingent upon context-specific validation.

2.6. Cross-Sector Transferability of the Framework

To examine the current state of integrated AI-IoT-blockchain applications in the literature and to clarify the positioning of this study relative to existing work,

Table 2. Position of this study against prior integrative studies on AI, IoT, blockchain integration Source: drafted by the authors.

Reference	Scope	Stages Covered	Sectors	Evidence Base
[12]	Integrated AI-IoT-blockchain framework for precision environmental management	Process control	Agriculture, smart cities	Multi-site deployment
[13]	Integrated AI-IoT-blockchain framework for food authenticity and traceability	Full-chain traceability	Food	Lab validation
[14]	Integrated AI-IoT-blockchain framework for aquaculture	Process control, Full-chain traceability	Aquaculture	Multi-site deployment/Pilot/Lab validation
[59]	Integrated AI-IoT-blockchain framework for coal gangue recycling	End-of-pipe treatment and recycling	Solid waste management	Simulation
[77]	Integrated AI-IoT-blockchain framework for low-cost digital MRV in bioethanol supply chains	Full-chain traceability	Energy	Multi-site deployment/Pilot/Simulation
[79]	Integrated AI-IoT-blockchain framework for food microbial analysis and safety	Process control, Full-chain traceability	Food	Lab validation
This study	Integrated AI-IoT-blockchain framework for cleaner production; proposes a three-layer architecture mapped onto the full cleaner production chain	Source reduction, Process control, End-of-pipe treatment and recycling, Full-chain traceability	Cross-sector transferability (energy, agriculture, manufacturing, etc.)	Multi-site deployment/Pilot/Simulation/Lab validation

compares representative triple-technology integration studies from recent years across four dimensions: scope, cleaner production stages covered, sectors involved, and evidence base. As shown in Table 2, most existing studies focus on one or two cleaner production stages and are concentrated in specific sectors such as aquaculture and food [13,14,79], while coverage of all four cleaner production stages remains relatively limited. Building on this understanding, this study attempts to map triple-technology integration onto the entire cleaner production chain, proposing a “sensing–analytics–trust” three-layer architecture. Furthermore, it explores the transferability of this framework to industries such as energy, agriculture, and manufacturing, where characteristics similar to those of the food sector are observed.

To explore the potential transferability of the conceptual framework to other sectors, we take the food industry as a useful entry point. The food sector exhibits several characteristics that are shared by many other industries, including raw material quality directly determining final product quality, the need for real-time monitoring of critical process parameters such as temperature, pH, and fermentation status; the generation of high-polluting by-products requiring circular utilization, and long supply chains involving multiple stakeholders. These shared characteristics make the food industry a suitable starting point for examining how the same conceptual logic might be adapted to other industrial contexts.

Building on this,

Table 3. Application Pathways of the Framework to Three Transferable Industries. Source: drafted by the authors.

Target Industries	Common Characteristics	Technology Applications	Expected Benefits	Evidence Base
Energy Production	Real-time operational data; supply-demand balance; multi-party transactions	IoT: Irradiance/wind sensors; grid monitors; smart meters AI: PV/wind forecasting (LSTM); load forecasting; fault diagnosis; storage dispatch Blockchain: Green certificates/carbon credits on-chain; provenance tracking; P2P trading with smart contracts	Wind/solar curtailment reduced; improved forecast accuracy [82]	Multi-site deployment/Pilot
Agriculture	Environmental data (soil/weather/pests); precision inputs; long traceability chains	IoT: Soil sensors (moisture/EC/pH/NPK); weather stations; drone/satellite imagery AI: Yield prediction; pest/disease detection (CNN); fertilization/pesticide optimization; irrigation optimization (RL) Blockchain: Seed-to-harvest traceability; certification on-chain; input records; smart contracts	Water and agrochemical consumption reduced; yield prediction accuracy improved; supply chain traceability time shortened [83]	Multi-site deployment/Pilot
Manufacturing	Equipment data (vibration/temperature/current); process optimization; multi-tier supply chains	IoT: Vibration/temperature/current sensors; AOI vision; PLC/SCADA AI: Predictive maintenance; quality prediction (SPC); scheduling optimization; yield root-cause analysis Blockchain: Raw material traceability; quality records on-chain per process step; maintenance logs; Digital Product Passport	Equipment reliability and operational efficiency improved [39]. Unplanned downtime reduced through predictive maintenance [40]; faster supply chain traceability	Multi-site deployment/Single case study/Simulation

maps the application pathways of the framework to three additional industries, namely energy, agriculture, and manufacturing, which share the above characteristics with the food sector [81]. For each industry, the table outlines the common characteristics, corresponding technology applications, and expected benefits, illustrating how the framework may be adaptable across diverse industrial contexts.

In summary, although energy, agriculture, and manufacturing differ from the food industry in terms of specific processes and product forms, they share the same core characteristics: dependence on raw material quality, real-time process monitoring, optimization of resource inputs, and long-chain multi-stakeholder coordination. Therefore, the three-layer “sensing–analytics–trust” architecture appears adaptable across sectors, with industry-specific adaptation expected in sensor deployment, AI model design, and blockchain record structures. The three-layer sensing-analytics-trust architecture draws on the established AI-IoT-blockchain stack in the Industry 4.0 literature [76]; the distinct contribution of this review lies in mapping this layered logic onto the four cleaner production stages and outlining cross-sector application pathways.

3. Scaling Challenges and Strategic Solutions

Despite the demonstrated potential of the AI-IoT-blockchain framework across the four cleaner production stages, its widespread adoption is hindered by a complex set of barriers that are intrinsically linked to the three functional layers—Sensing, Analytics, and Trust. Importantly, these challenges are not static; ongoing technological advances are progressively mitigating them, creating a self-reinforcing cycle of improvement.

3.1. Sensing Layer (IoT)

At the Sensing layer (IoT), data quality and cost remain the primary challenges. Industrial environments generate noisy, incomplete, or inconsistent data due to sensor drift and environmental interference, undermining the reliability of downstream analytics [14]. This directly affects process control through suboptimal parameter adjustments and end-of-pipe treatment through inefficient chemical dosing. Meanwhile, high deployment costs limit real-time sensing adoption across all stages, particularly for small and medium-sized enterprises (SMEs) [14].

Recent technological breakthroughs are alleviating these constraints. On the cost side, emerging photonic chip-based sensing technologies suggest potential cost reductions in sensor hardware. For example, Perceptra, an MIT spin-off, has developed a photonic chip-based Raman sensor that, according to a company announcement, achieves significant reductions in size and cost compared to conventional lab-grade systems, while maintaining comparable sensitivity [84]. This makes source reduction quality inspection economically viable for more facilities and expands sensor coverage in full-chain traceability. On the data quality side, machine learning-based field calibration techniques have been shown to substantially improve the accuracy of low-cost particulate matter sensors [85]. Similar approaches have been demonstrated in industrial contexts where sensor drift compensation is applied to FET-based ion sensors and pH monitoring in continuous water quality assessment. Koziel et al. (2025) [86], for example, reported coefficients of determination reaching up to 0.89 under real-world urban monitoring conditions, illustrating the potential of ML-based calibration for broader sensing applications. Simultaneously, edge AI enables on-device data cleaning and feature extraction [87,88,89], ensuring that only high-quality data streams reach AI models—directly improving process control reliability and end-of-pipe monitoring accuracy.

3.2. Analytics Layer (AI)

The Analytics layer’s barriers center on model opacity and data silos. The “black-box” nature of advanced AI models erodes operator trust when models recommend material substitutions or process adjustments without clear rationale [57]. This is particularly critical in cleaner production contexts, where human operators need to validate AI-driven decisions against their domain knowledge to ensure safety and compliance. A separate but equally important barrier is that training robust models requires large datasets that remain fragmented across stakeholders. A further concern is the escalating energy demand of AI itself: training large-scale deep learning models and operating data centers consume substantial electricity, and this environmental footprint can partially offset the sustainability gains that AI enables in cleaner production applications [82]. However, empirical evidence indicates that the operational energy cost of AI can be outweighed by the efficiency gains it enables. In a cement plant case study, Jena et al. (2026) [90] reported that AI-driven optimizations reduced specific energy consumption by 10.16% and carbon footprint by 10.11%, with a payback period of 2.3 years. These figures demonstrate measurable process-side environmental and economic benefits in an energy-intensive manufacturing context. However, whether these process savings outweigh the AI system’s own training and inference energy footprint remains an open question, as the latter was not quantified in the study.

Explainable AI (XAI) techniques such as SHAP and LIME address the opacity challenge by providing both global feature importance rankings and local explanations for individual predictions [91]. XAI has been shown to enhance user confidence and facilitate troubleshooting in IoT-enabled industrial settings [91], directly benefiting source reduction, process control, and end-of-pipe treatment through improved transparency and regulatory acceptance. Federated learning (FL) tackles data silos by enabling decentralized training across multiple sources without raw data exchange [92,93]. Data heterogeneity across participating nodes, including inconsistent labeling, missing values, and distributional skewness, remains a challenge that can degrade model performance if not properly addressed [88]. Despite this, FL allows suppliers to aggregate quality insights without exposing proprietary formulations, enables collaborative process optimization across production lines, and supports full-chain traceability through distributed model improvement while preserving data sovereignty.

3.3. Trust Layer (Blockchain)

The Trust layer’s challenges center on two interconnected tensions. First, the transparency-privacy trade-off means that public blockchains, by design, expose transaction data to all participants, which can reveal proprietary production parameters, supplier identities, and cost structures—information that manufacturers are typically unwilling to share [9,64,94]. Second, the high energy consumption of certain blockchain architectures, particularly Proof-of-Work (PoW) consensus [95], conflicts with cleaner production goals by introducing a significant environmental overhead that may offset the sustainability gains achieved elsewhere.

Privacy concerns are being addressed through permissioned blockchains with role-based access control [96] and zero-knowledge proofs [97]. These enable selective data disclosure, allowing supply chain partners to verify provenance and compliance without exposing sensitive formulations or cost structures. Hyperledger Fabric-based consortium blockchains have been successfully deployed in building photovoltaic life-cycle management [98,99], demonstrating how permissioned architectures can support secure multi-stakeholder data exchange while preserving commercial confidentiality [100]. This capability directly enhances full-chain traceability and builds trust among stakeholders who would otherwise resist participation. Energy consumption, meanwhile, is being mitigated through alternative consensus mechanisms. Hybrid PoS-PBFT consensus has demonstrated substantial energy savings compared to Proof-of-Work in SME-scale applications, ensuring that the framework’s sustainability gains are not offset by infrastructure overhead [77]. More broadly, the choice of blockchain architecture matters: permissioned and PoS-based systems consume substantially less energy per transaction than PoW networks [95], making them better suited for industrial cleaner production contexts where environmental performance is a primary concern.

3.4. Cross-Layer Challenges and Dynamic Feedback

Beyond layer-specific barriers, cross-layer frictions further undermine the framework’s performance. A data interoperability gap arises from incompatible data formats and communication protocols among IoT devices, AI platforms, and blockchain systems from different vendors, leading to high integration costs and fragmented data flows [101,102,103]. Compounding this technical fragmentation is an organizational capability gap: the interdisciplinary talent capable of spanning sensing, analytics, and trust domains remains scarce, and such skills rarely reside within a single team or organization, slowing adoption particularly among SMEs [14]. Regulatory uncertainty further complicates the picture, as underdeveloped legal frameworks for data ownership, cross-border data flow, and smart contract enforceability create significant hurdles for cross-jurisdictional deployments [69,104].

To resolve technical incompatibilities, industry consortia have developed unified frameworks such as Hyperledger [98,99], which provide common interfaces for multi-stakeholder systems, enabling IoT devices, AI platforms, and blockchain systems from different vendors to interoperate more seamlessly. To tackle organizational and social barriers, cross-disciplinary training programs and university-industry partnerships are building the needed workforce [105]. Regulatory sandboxes provide a controlled environment where businesses can test integrated applications without immediate risk of penalties [106,107], allowing regulators to observe and develop appropriate rules while enabling innovation.

Economic viability is a critical enabler for the adoption of the integrated framework, yet detailed cost data remain scarce in the reviewed literature. Most studies report whole-framework economics rather than layer-disaggregated capex or opex figures. Across the available case studies and simulations, the economic performance of integrated AI-IoT-blockchain deployments varies considerably with sector, scale, and technology readiness. SME-scale simulation-based modeling in the bioethanol supply chain reported a 20% average ROI, with 49% of scenarios reaching payback within five years [77]. A case study from a manufacturing green supply chain, integrating AI, IoT, and collaborative robotics with blockchain-enabled procurement, achieved a 2.5-year payback and annual savings of €200,000, alongside a 20% reduction in energy consumption and a 15% decrease in CO₂ emissions [108]. These figures are consistent with the magnitude of economic benefits reported in the coal gangue recycling case [59], where a 3.6-year payback and 22% ROI were achieved. While these figures are derived from a mix of actual deployment data and simulation-based estimates, and should therefore be interpreted as indicative rather than definitive, they suggest that the framework can be economically viable under favorable conditions, with payback periods broadly ranging from 2.5 to 5 years depending on the sector, scale, and specific technologies deployed. The absence of systematic, layer-specific cost reporting in the literature limits the generalizability of these estimates, highlighting a clear direction for future empirical research.

Beyond economic viability, the deployability of the integrated framework also hinges on alignment with existing technical standards and cybersecurity frameworks for operational technology. On the interoperability side, OPC Unified Architecture [109] has been established as a core standard for digital asset representation and communication in Industry 4.0 [110], providing a semantic information model for heterogeneous manufacturing environments. For supply chain traceability, private blockchain-based food safety distribution systems have demonstrated successful integration with GS1 Electronic Product Code Information Services standards [111], enabling tamper-proof tracking of environmental data and location information across the logistics chain [112]. From a management perspective, international standards such as ISO 50001 [113] and ISO 14001 [114] provide foundational frameworks for organizational action and digital support in enhancing energy efficiency and environmental performance [115]. On the cybersecurity front, connecting production equipment to networked AI and blockchain systems creates exposure to operational technology (OT)-specific threats. The IEC 62443 [116] family of standards has been proposed as a key framework for implementing cyber defense in industrial environments, providing systematic risk assessment and security measures to protect industrial automation and control systems [117]. A relevant policy development is the EU Ecodesign for Sustainable Products Regulation [118], which introduces Digital Product Passports for a wide range of products. A distributed ledger and zero-knowledge proof-enabled framework has been proposed to implement privacy-preserving Digital Product Passports in compliance with the ESPR, anchoring event records and enabling stakeholders to verify compliance predicates without revealing commercially sensitive operational data [119]. These standards and policy frameworks are not intended as implementation requirements for this conceptual review, but they establish the boundary conditions that future deployments would need to address.

Beyond these static measures, the dynamic feedback among the three layers also deserves attention. AI predictions can adjust IoT sampling rates, increasing frequency when deviations are detected and reducing it during normal operation to conserve energy [120]. Meanwhile, blockchain-verified data can be fed back into AI retraining as high-quality ground truth. However, this closed-loop logic introduces temporal dynamics that must be carefully managed. Data freshness [121], defined as the elapsed time from sensor data collection to its commitment on the blockchain, directly affects process control responsiveness. Kim et al. (2024) demonstrated that blockchain processing latency can significantly degrade data freshness [121] and proposed a reinforcement learning-based sensing decision algorithm to enable intelligent sampling adjustments based on channel conditions, blockchain latency, and energy status. Similarly, Lee et al. (2021) identified blockchain consensus latency [122], particularly the endorsement, ordering, and validation phases in permissioned networks, as a critical factor affecting the timeliness of data available for AI inference. Managing these temporal dynamics is essential to ensuring that the closed-loop logic translates into tangible, cleaner production outcomes. Timely process adjustments reduce resource waste and emissions, while verifiable data records support compliance and stakeholder trust across the entire production chain. The foregoing discussion is consolidated in

Table 4. Mapping of barriers and mitigation strategies to specific layers and cleaner production stages.

Specific Challenge	Affected Layer(s)	Affected Cleaner Production Stage(s)	Mitigation Strategy	Key References
Data quality issues (noise, drift, incompleteness)	Sensing (IoT)	Process control, End-of-pipe	Edge AI data cleaning; ML-based field calibration	[14,86,87,88,89]
High deployment and maintenance costs	Sensing (IoT)	All stages	Photonic chip-based sensors; AI-driven cost reductions	[14,84]
Model opacity (black-box) eroding operator trust	Analytics (AI)	Source reduction, Process control, End-of-pipe	Explainable AI (SHAP, LIME)	[57,91]
Fragmented training data across stakeholders	Analytics (AI)	Source reduction, Process control, Traceability	Federated learning	[92,93]
Escalating energy demand of AI infrastructure	Analytics (AI)	Process control	Energy-efficient AI architectures	[82,90]
Transparency-privacy trade-off	Trust (Blockchain)	Traceability	Permissioned blockchains; zero-knowledge proofs	[9,64,94,96,97]
High PoW energy consumption	Trust (Blockchain)	Traceability	PoS/PBFT hybrid consensus; permissioned architectures	[77,95]
Data interoperability gaps across layers	All layers	All stages	Unified frameworks (e.g., Hyperledger); industry standards	[98,99,101,102,103]
Scarce interdisciplinary talent	All layers	All stages	Cross-disciplinary training; university-industry partnerships	[14,105]
Undeveloped regulatory frameworks	All layers	Traceability	Regulatory sandboxes; multi-stakeholder governance	[69,104,106,107]

, which maps each barrier and its corresponding mitigation strategy to the specific layer and cleaner production stage it affects, providing a structured reference for researchers and practitioners.

In summary, while the AI-IoT-blockchain framework faces multiple challenges in scaling cleaner production applications, these barriers appear addressable through coordinated technical, economic, and policy measures. With continued research, policy support, and cross-industry collaboration, the framework offers a promising direction for advancing cleaner production across diverse sectors.

4. Conclusions and Future Perspectives

In this review, we trace the evolution of AI, IoT, and blockchain in cleaner production from single-technology applications to fully integrated frameworks. The reviewed evidence suggests that the three technologies play complementary roles across the four cleaner production stages, and that integrated deployments have reported benefits including enhanced traceability, real-time process optimization, and verifiable record-keeping, which single-technology or pairwise applications do not typically address. The framework creates a complete digital thread in which IoT provides real-time sensing, AI delivers cognitive analytics, and blockchain ensures trusted, auditable record-keeping. While case studies such as coal gangue recycling and aquaculture traceability illustrate technical feasibility and economic benefits in their respective contexts, widespread adoption faces several obstacles. These include high capital and energy costs; technical barriers such as interoperability issues and a lack of unified standards; and organizational challenges, including scarce interdisciplinary talent, AI’s black-box nature, and underdeveloped regulatory frameworks for data ownership and smart contracts. To overcome these barriers, strategic solutions are available, such as modular deployment, cloud-based AI and blockchain services, federated learning, permissioned blockchains with role-based access, and regulatory sandboxes.

Nevertheless, several aspects of this review also point to clear opportunities for refinement. While we have conceptually argued for the framework’s cross-sector applicability, this remains a logical inference rather than empirical validation. The framework has not yet been tested in domains such as chemicals, textiles, or electronics. Each of these industries presents unique technical and regulatory challenges that may require tailored adaptations. In addition, our analysis relies on case studies drawn from publicly available literature, which introduces the possibility of publication bias, potentially skewing the perceived effectiveness of the integrated approach.

These opportunities for refinement directly inform our proposed future research agenda. Future research should prioritize three directions: cross-sector empirical validation in industries such as chemicals, textiles, and electronics; life-cycle economic and environmental assessment to quantify cost–benefit and carbon reduction trade-offs; and governance and legal design addressing data sovereignty, smart contract enforceability, and cross-border supply chain rules. Lighter-weight technologies, including edge AI and IOTA-based blockchains, also merit investigation for SME adoption. The key strength of the framework lies in its architectural transferability: the sensing-analytics-trust logic may be applicable wherever production processes generate data, require decisions, and demand verification, though its practical utility in specific sectors remains to be demonstrated through empirical validation. With continued research, policy support, and cross-industry collaboration, this integrated framework holds strong promise for advancing sustainable manufacturing, the circular economy, and low-carbon development.