Predictive Maintenance Gets a Human Touch: New 5.0 Framework Links Machines, Energy and People

Predictive maintenance has long been one of the quiet triumphs of the industrial internet. Sensors listen to the vibrations of a bearing, algorithms parse the temperature signature of a motor, and maintenance crews intervene days before a catastrophic failure would otherwise halt production. Yet for all its technical sophistication, the field has remained stubbornly narrow: machines are monitored, failures are forecast, and everything else—energy, emissions, and above all the human beings who operate the equipment—has been treated as an afterthought. A new study published in Discover Sustainability argues that this narrowness is no longer tenable, and it offers both the evidence and the blueprint for something broader.

The research, led by Nada Baddou and colleagues at Sidi Mohamed Ben Abdellah University in Fez, Morocco, together with a collaborator at Moulay Ismail University in Meknes, set out to answer a deceptively simple question: how mature is the scientific literature on predictive maintenance when measured against the ambitions of Industry 5.0? This emerging paradigm, promoted by policymakers and industrial strategists alike, shifts the emphasis from pure automation and efficiency toward three pillars—sustainability, human-centricity, and resilience. Where Industry 4.0 asked how smart a factory could become, Industry 5.0 asks how responsible and robust it can be, and how well it can serve the people inside it.

To map the field, the team conducted a scoping review of the Scopus database, screening 238 candidate articles and retaining 123 that met substantive-engagement criteria. The screening was not merely a matter of counting keywords. Each retained study was assessed through a weighted scoring protocol designed to gauge how deeply it engaged with the three Industry 5.0 pillars, rather than simply mentioning them in an introduction. This distinction matters, because the authors found a pronounced skew toward shallow integration: many papers invoke sustainability or human factors as framing language, but only a minority actually operationalize all three pillars jointly within their methods, metrics, or system architectures.

The diagnosis is sobering for a field that often presents itself as the vanguard of smart manufacturing. A predictive maintenance system that reduces unplanned downtime does, incidentally, save energy and materials—fewer scrapped components, fewer emergency shipments, less wasted production. But incidental benefits are not the same as designed accountability. Without explicit pathways linking equipment degradation to quantified energy consumption and carbon dioxide emissions, a maintenance system cannot tell an operator how much greenhouse gas a deteriorating compressor is quietly adding to the plant’s footprint. Without human-in-the-loop mechanisms, the flood of algorithmic alerts becomes another source of cognitive burden rather than a tool that empowers skilled workers.

Building on these findings, the researchers propose what they call Predictive Maintenance 5.0, an implementation-oriented framework structured around three interconnected layers. The first layer addresses equipment health assessment, the familiar territory of sensor-driven degradation modeling and failure prediction. The second introduces an energy and CO2 estimation pathway that explicitly connects the physical state of machinery to sustainability indicators, translating wear and tear into projected kilowatt-hours and emissions. The third layer is perhaps the most distinctive: a structured interaction model in which a Large Language Model mediates communication between the analytical system and human operators, converting raw diagnostic outputs into explanations a maintenance technician can actually interrogate and act upon.

The choice of a Large Language Model as the interface is more than a fashionable flourish. Traditional predictive maintenance dashboards present probabilities, remaining-useful-life estimates, and anomaly scores—numbers that demand interpretation. An LLM-mediated layer can contextualize those numbers, answer follow-up questions, and adapt its explanations to the operator’s level of expertise, effectively institutionalizing the human-in-the-loop principle that Industry 5.0 champions. The framework also incorporates an adaptive learning mechanism, allowing the system to refine its models as operating conditions shift, which speaks directly to the resilience pillar: a maintenance system that cannot adapt to change is itself fragile.

To demonstrate that the architecture is more than a paper exercise, the team carried out a preliminary industrial case study using operational data from a recycled-paper mill. Paper manufacturing is an energy-intensive process with heavy rotating machinery, making it a natural proving ground for maintenance systems that must balance uptime against energy and emissions targets. The case study executed three interconnected pathways of the framework in practice: assessing equipment health from operational data, estimating the energy and CO2 impact of degradation, and running the human-in-the-loop interaction supported by LLM-generated explanations of the system’s outputs.

The human dimension of the case study was evaluated with the professional feedback of four operators who assessed the system’s outputs. The authors are careful about the scope of this evaluation—no identifiable participant information is reported, and the feedback represents an initial, formative assessment rather than a large-scale validation. Still, the exercise illustrates what human-centric predictive maintenance could look like on a factory floor: not a black box issuing decrees, but a conversational system whose reasoning about a machine’s condition can be examined, questioned, and enriched by the people who know that machine best.

Equally important for the field’s development, the researchers propose layer-specific evaluation metrics intended to support broader operationalization. This is a subtle but consequential contribution. A recurring weakness in Industry 5.0-aligned research is the absence of agreed measures: if sustainability, human-centricity, and resilience are to be engineered rather than merely proclaimed, they must be quantifiable at each layer of the system. By specifying how each layer of the framework should be assessed, the authors give other research groups and industrial adopters a structured basis for comparing approaches and for validating the framework beyond a single mill.

The broader significance of the work lies in its reframing of what predictive maintenance is for. The study’s evidence base—the 123 retained studies and their scoring—suggests that the research community has been optimizing a subsystem while the surrounding industrial paradigm has moved on. By grounding a new architecture in a systematic review and then stress-testing it against real operational data, the Moroccan team has provided both a mirror and a map: a mirror showing how far current practice falls short of the Industry 5.0 vision, and a map showing how equipment health, energy accounting, carbon estimation, and human expertise can be woven into a single operational loop. The authors describe their contribution as advancing predictive maintenance from a purely technical function toward a holistic, Industry 5.0-aligned paradigm, while offering initial empirical evidence of operational feasibility. Whether the framework spreads across energy plants, recycling facilities, and process industries will depend on the broader industrial validation the authors call for—but the direction of travel is now clearly marked, and it points toward factories where machines, sustainability metrics, and skilled workers are monitored, explained, and served together rather than in isolation.

Subject of Research: Industry 5.0-aligned predictive maintenance frameworks integrating sustainability, human-centricity and resilience

Article Title: Toward an operational predictive maintenance 5.0 framework informed by a scoping review and an industrial case study

Article References: Baddou, N., Benjelloun, M., Dadda, A., & Rzine, B. (2026). Toward an operational predictive maintenance 5.0 framework informed by a scoping review and an industrial case study. Discover Sustainability. https://doi.org/10.1007/s43621-026-04911-1

Image Credits: AI Generated

DOI: 10.1007/s43621-026-04911-1

Keywords: predictive maintenance, Industry 5.0, sustainability, human-centricity, resilience, large language models, CO2 emissions, energy efficiency, scoping review, industrial case study, recycled paper mill, machine learning