Artificial intelligence-based predictive maintenance, time-sensitive networking, and big data-driven algorithmic decision-making in the economics of Industrial Internet of Things

With Industry 4.0, manufacturing has become more intelligent and data-driven, integrating heterogeneous data from numerous sources to support timely managerial decision-making. Maintenance of production machines is shifting from reactive and preventive approaches to predictive maintenance (PM), which monitors performance in the active state to identify the optimal time to intervene before breakdowns. Deploying robust PM solutions requires sophisticated edge control infrastructures and scalable applications, alongside machine learning (ML) software that adapts to anticipated changes. AI-based PM significantly affects economic efficiency and sustainability by predicting equipment failures, streamlining maintenance procedures, reducing costs, improving productivity, and extending asset lifecycles. The study applies advanced AI techniques within IIoT to proactively predict and address equipment issues before significant failures, targeting enhanced efficiency and reliability of maintenance practices. The article is structured to diagnose the research gap, introduce IIoT architecture, highlight the importance of PM, review PM evolution and AI’s role, detail AI-driven PM techniques, present results and case examples, and conclude with key findings and future directions, offering insights for researchers, practitioners, and decision-makers.

The conceptual review identifies a research gap in configuring the economics of IIoT via digital twin simulation, multi-sensory tracking, haptic and biometric sensors, geospatial big data algorithms, blockchain, image recognition, socially interconnected services, and smart contracts. It emphasizes the need for mobile location analytics, remote sensing fusion, event modeling, forecasting tools, and spatial cognition algorithms to enable big data-driven governance of cyber-physical manufacturing. Data-driven sustainable smart manufacturing leverages ML-based object recognition, deep learning-based sensing, haptic AR, interactive 3D geo-visualization, predictive modeling, and virtual simulations, supported by industrial big data and real-time sensor networks. The IIoT architecture comprises three layers: physical (sensors/actuators, RFID, accelerometers), network (protocols such as MQTT, CoAP, LoRaWAN, BLE; security; data models DSD/DI/DK/DSM), and application (analytics, ML/AI). Data extraction spans temporal, frequency, and time-frequency domains, and data are categorized into flow, configuration, and event databases. The review of maintenance approaches traces evolution from corrective and opportunistic to condition-based and predictive maintenance, incorporating resilience as a critical measure for Industry 4.0. Related work synthesizes two frameworks: (1) PHM for flexible PM, integrating CMMS, FMEA/FMECA/PFMEA, expert knowledge, time-series data via OPC UA/Modbus/MQTT/PLC protocols, threshold setting (simulation/AR/ML), and multivariate ML analyses to correlate data and predict failures; (2) AIDA, a holistic AI-driven networking and processing framework delivering real-time edge-to-cloud capabilities and time-sensitive networking (TSN), centralized network configuration (CNC) with operational and TSN subsystems, and centralized user configuration (CUC). Standards such as PTP, IEEE 802.1Q, and IEEE 802.11 enable precise synchronization, traffic classification, and QoS. Edge services include measurement/delivery, data fusion/storage (time series, Loki logs), visualization/alerting (Grafana), and provisioning/orchestration (Kubernetes/Helm), with ML methods spanning ANN, SVM, decision trees, random forests, logistic regression, XGBoost/GBM, linear/symbolic regression, reinforcement learning, and metaheuristics.

A conceptual research method was employed, focusing on literature (2020–2023) indexed in WoS and Scopus and sourced from Emerald, IEEE Xplore, ScienceDirect, and NCBI. A semantic search strategy combined keywords and phrases (e.g., IIoT applications; predictive maintenance; ML-based industrial decision-making strategies; PM challenges) using Boolean operators (AND) to retrieve narrowly defined, relevant articles. Inclusion criteria: reputable journals/conferences; studies on AI-driven PM in IIoT; real-world case studies or implementations. Exclusion criteria: lack of methodological transparency; absence of clear connection to real-world implementations. The process involved initial screening of titles/abstracts, full-text reviews, and qualitative synthesis via content and thematic analyses to extract information on AI-driven PM in IIoT. Findings were documented transparently, with synthesis relating insights to real-world implementations.

• AI-driven PM within IIoT enhances operational efficiency, reduces downtime, extends asset lifecycles, and lowers maintenance costs by proactively predicting equipment failures and enabling timely interventions. - IIoT’s three-layer architecture (physical, network, application) supports robust data collection, classification (flow/configuration/event), and analysis, enabling scalable PM solutions. - TSN provides the networking backbone for PM, ensuring on-time delivery of critical data via time scheduling, precise synchronization (PTP master/slave), flow control, QoS differentiation (IEEE 802.1Q, IEEE 802.11 TSPEC/TCLAS/TID), and scalability using standard Ethernet. - Centralized Network Configuration (CNC) and Centralized User Configuration (CUC) coordinate synchronization, resource allocation, routing, and endpoint configurations, minimizing latency and optimizing data paths. - Edge computing services (measurement/delivery with dual quality assessment; data fusion/storage including Loki logs; visualization/notifications via Grafana; provisioning/orchestration with Kubernetes/Helm) enable end-to-end observability and resilient PM pipelines. - PHM-driven PM integrates CMMS, FMEA/FMECA/PFMEA, and expert knowledge to select failure modes, define data collection (time-series via OPC UA/Modbus/MQTT/PLC), and determine thresholds (simulation/AR/ML). - ML techniques (ANN, SVM, decision trees, random forests, logistic regression, XGBoost/GBM, linear/symbolic regression, RL, metaheuristics) address classification, prediction, feature selection, RUL estimation, and maintenance policy optimization, with reinforcement learning and metaheuristics supporting action selection and parameter tuning. - Organizational requirements include systems integration across IT/maintenance/production, robust data security, and cross-functional collaboration to operationalize PM. - Business context influences PM design: equipment types and sensor modalities, asset criticality (e.g., aviation safety), and process complexity (component count, interdependencies, heterogeneous data, production modes).

The study addresses the research question of how AI-based PM, TSN, and big data analytics can optimize maintenance within IIoT by detailing architectures, data models, and algorithmic approaches that collectively enable proactive maintenance. AI models trained on historical and streaming IIoT data detect deviations from normal operations to preempt failures, while TSN guarantees timely, reliable delivery of critical maintenance data. The CNC/CUC framework ensures synchronized, low-latency routing and resource allocation, translating algorithmic decisions into network-level guarantees. PHM and AIDA frameworks provide complementary views: PHM aligns domain knowledge (CMMS, FMEA/FMECA) with data-driven thresholds and multivariate analysis; AIDA operationalizes end-to-end monitoring, storage, visualization, and orchestration at the edge-to-cloud continuum. These findings demonstrate practical pathways to improved OEE, cost savings, and resilience. Their relevance spans sectors but depends on tailoring sensors, models, and thresholds to equipment criticality and process complexity, supported by organizational integration and cybersecurity. Overall, the synergy of AI analytics and time-sensitive networking creates a robust infrastructure for predictive, scalable maintenance in Industry 4.0.

AI-based predictive maintenance within IIoT transforms maintenance from reactive/preventive paradigms to proactive, data-driven strategies. By leveraging advanced ML/AI algorithms, real-time data collection, and TSN-enabled reliable networking, organizations can predict failures, reduce downtime, optimize resource allocation, and extend asset lifecycles. The study contributes theoretically by elucidating the symbiotic relationship between AI and IIoT and practically by outlining actionable architectures (PHM, AIDA), networking configurations (CNC/CUC/TSN), and edge services (measurement, storage, visualization, orchestration). Future research should refine AI models for sector-specific generalizability, address ethical and bias concerns, strengthen data security and governance, explore hybrid approaches integrating AI with emerging technologies (e.g., digital twins, AR/VR, blockchain), and develop workforce skills for managing and interpreting complex IIoT data.

• Generalizability may vary across industrial sectors due to differences in equipment, operational conditions, and data characteristics. - Ethical considerations and potential algorithmic bias may affect decision-making in critical processes. - Data security and privacy challenges persist given real-time, large-scale data collection and analysis. - Dependence on skilled professionals for deploying, managing, and interpreting complex AI/IIoT systems. - The study is conceptual and synthesizes literature; it lacks sector-specific quantitative validation or controlled experiments.