2.1 Evolution and limitations of existing safety models

The complexity of modern aviation maintenance environments necessitates a shift from linear accident models toward frameworks capable of addressing emergent behaviours and performance variability in socio-technical systems. Two of the most prominent models are STAMP (Systems-Theoretic Accident Model and Processes) and FRAM (functional resonance analysis method). However, SMS frameworks often fail to fully capture dynamic human factors such as psychological stress, environmental variability and team dynamics. At the same time, these safety models increasingly require integration with dynamic hazard detection mechanisms, moving beyond retrospective analysis towards real-time system adaptability [Reference Kıvanç, Tuzkaya and Vayvay62, Reference Xiong, Wang, Wong and Hou114].

2.1.1 STAMP: a systems-theoretic approach to safety control

STAMP, developed by [Reference Leveson66], represents a significant departure from traditional failure-event models, in the sense that it conceptualises safety as an emergent property of a system’s control architecture, rather than as the outcome of discrete technical failures (Fig. 1). In STAMP, accidents occur due to component malfunctions and inadequate enforcement of safety constraints across a hierarchical structure of interacting components, technical, organisational and human. STAMP is structured around the idea that each layer of a socio-technical system, ranging from frontline operators to management and regulators, exerts control through constraints, feedback loops and decision-making processes. Two critical sub-methodologies have emerged within STAMP:

• • CAST (Causal Analysis based on STAMP) that focuses on identifying how systemic control failures propagate through an organisation, often uncovering mismatches between planned safety structures and actual practices.

• • STPA (System-Theoretic Process Analysis) is designed to identify potential hazard scenarios proactively and inform design interventions before an accident occurs [Reference Allison, Revell, Sears and Stanton3].

Figure 1. The system safety engineering process components based on STAMP. Adapted from Ref. (Reference Leveson66).

Despite its conceptual strength, STAMP faces practical limitations in contexts that demand rapid or dynamic responses, such as aircraft maintenance. While STPA offers predictive insight into potential hazards, it requires extensive qualitative data and time-intensive modelling. It also lacks built-in mechanisms for quantifying uncertainty, particularly the kind that emerges from fluctuating psychological states, resource constraints or inter-team coordination issues under time pressure. These shortcomings hinder its scalability and real-time applicability in fast-paced operational environments such as line or base maintenance, where uncertainty must be assessed and acted upon rapidly.

2.1.2 FRAM: a model of functional variability and emergent risk

The FRAM, proposed by [Reference Hollnagel41], follows a different, yet complementary approach by shifting the focus from component failure to everyday performance variability (Fig. 2).

Figure 2. A display of FRAM. A hexagon representing a function, with the six aspects of input (I), output (O), preconditions (P), resources (R), control (C), and time (T). In public domain [Reference Hollnagel41].

Nomenclature:

• • I: that which the function processes or transforms or that which starts the function

• • O: the result of the function, can be an entity or a state change

• • R: resource(s) required for the processing performed by the function

• • T: time required for the processing performed by the operational unit

• • C: constraints and controls on the operational unit (procedures, methods, etc.)

• • P: conditions that must be satisfied before a function can be carried out

It posits that both safe and unsafe system outcomes emerge from interactions between routine functions and that risk arises when these functional outputs resonate or interact in unpredictable ways. FRAM models work as nodes with six aspects (input, output, preconditions, resources, control, and time), emphasising that variability in one function can propagate through the system, potentially amplifying other minor deviations into major disruptions. This makes it particularly well-suited for analysing socio-technical systems like aircraft maintenance, where adaptive behaviour is often required to handle changing operational demands, environmental conditions or incomplete information. The strengths of FRAM lie in its ability to:

• • Map complex interactions between human and technical functions.

• • Capture non-linear emergent phenomena in operations.

• • Recognise that variability is a normal, not an exceptional, condition in human performance.

However, as Tian et al. [Reference Tian, Caponecchia and Kourousis104] highlight in their systematic review of 108 FRAM applications, including 26 in aviation, FRAM is primarily qualitative and lacks precise mechanisms for time-sensitive quantification or risk calculation. Its application requires intensive manual modelling and contextual knowledge, which impedes its utility in real-time decision-support systems or predictive safety monitoring. Furthermore, it struggles to incorporate dynamic human-centric uncertainty, such as fluctuating psychological load, task ambiguity, or shifting team structures – conditions common in aviation maintenance operations.

3.1 Literature review methodology

A systematic literature review was undertaken, following a structured seven-stage protocol adapted from Booth et al. [Reference Booth, Sutton and Papaioannou13], Grant and Booth [Reference Grant and Booth36] and Okoli [Reference Okoli79, Reference Okoli80], as shown in Fig. 3.

Figure 3. The procedural steps of the literature review.

Search strategies combined Boolean logic across four academic databases (ProQuest, ScienceDirect, SAGE and EBSCOhost), applying human error, uncertainty, safety management, maintenance, STAMP, FRAM and SMS keywords. Articles were selected according to language (English or French), publication quality (peer reviewed) and domain relevance (safety critical systems), focusing on sources from 2006 to 2025 for applied studies, with broader inclusion criteria for theoretical frameworks. From an initial pool of 2244 sources, 73 articles were ultimately included in the final synthesis. Figure 4 presents the typological breakdown of the included sources, indicating that the majority were peer-reviewed journal articles (68.5%), followed by scholarly monographs (20.5%), book chapters (6.8%) and technical or institutional reports (4.1%). Thematic clustering revealed that while 64% of sources addressed human error or human factors, fewer than 7% attempted any quantitative operationalisation of uncertainty, and none provided validated tools for field-level application in aircraft maintenance. These findings align with the observations made by Reason [Reference Reason90] and Liu [Reference Liu68], who highlighted the limitations of retrospective error models and the absence of predictive frameworks.

Figure 4. Percentage distribution by publication type.

As illustrated in Fig. 5, the thematic distribution of the selected studies reveals a prevailing research concentration on general human factors topics, particularly those situated within safety-critical industries and defence-related safety systems. In contrast, only a marginal proportion of the literature directly estimates or quantifies uncertainty. These findings also underscore the limitations of dominant human error taxonomies – slips, lapses and violations, as outlined by Reason [Reference Reason90] and Wickens et al. [Reference Wickens, Hollands, Banbury and Parasuraman110]. These classifications describe error as a retrospective artifact rather than a dynamic construct, influenced by evolving environmental, organisational and psychological factors. As a result, they fail to capture the cyclical, situated drivers of uncertainty that characterise real-world aviation maintenance environments.

Figure 5. The sources arranged by themes.

3.2 Research gap

Despite significant advancements in aviation safety research, the current body of literature remains constrained in its ability to fully capture human behaviour’s complexity, fluidity and unpredictability in aircraft maintenance. A systematic review of 73 key studies (Table 1 and Table B1 in Appendix B) revealed several persistent gaps across theoretical, methodological and practical dimensions, particularly in relation to human-centric uncertainty and its influence on safety outcomes.

1. 1. Widely adopted safety models such as STAMP and FRAM emphasise systemic structures, control constraints and functional variability. However, they fail to represent the probabilistic and emergent nature of human behaviour, especially under conditions of uncertainty. These models lack mechanisms for modelling real-time variability in psychological, environmental and team-based factors that directly influence operational safety.

2. 2. Existing approaches to human factors, including HRA and human factor analysis and classification system (HFACS), rely predominantly on qualitative or taxonomic methods, which, although insightful, do not lend themselves to proactive, quantitative risk prediction. This absence of quantitative modelling limits our ability to anticipate error emergence and implement timely interventions.

3. 3. Most safety models in aviation maintenance remain static and retrospective in nature. They are not designed to evolve dynamically with frontline conditions or integrate live operational data. In high-risk environments like aircraft maintenance, where situational variables shift rapidly, this rigidity diminishes the practical utility of such models in real-time decision-making.

4. 4. Prevailing positivist and deterministic paradigms underpinning most safety research, assume error to be an observable, linear output of failure chains. This view neglects the subtleties of human adaptation, context-driven decisions, and the subjective experience of uncertainty. Consequently, the field lacks operator-centric research that explores how frontline personnel perceive, navigate, and mitigate uncertainty before it manifests as error.

The shift from traditional compliance-driven SMS frameworks toward continuous, predictive safety systems has become increasingly vital in managing growing operational complexity and uncovering latent risks [Reference Kıvanç, Tuzkaya and Vayvay62]. Effectively addressing both foreseeable ‘grey rhino’ threats and rare, high-impact ‘black swan’ events demand a more sophisticated, real-time approach to uncertainty modelling and hazard detection [Reference Xiong, Wang, Wong and Hou114]. This evolving landscape highlights an urgent research imperative: developing a novel safety framework that quantitatively models human uncertainty as a dynamic and context-aware variable. Such a framework should be rooted in real-world operational insights, remain adaptable to shifting field conditions and enhance existing safety paradigms such as SMS, STAMP and FRAM.

Table 1. Gap domains in uncertainty modeling

Table 2. The interviews outline

3.3 Conceptual response: the UQAM framework

In response to the abovementioned gaps, this study introduces the (UQAM framework, a novel, hybrid model designed to mathematically represent and operationally manage human-centric uncertainty within safety-critical maintenance environments. At the core of UQAM is the IUE, a computational mechanism that synthesises field-level assessments across eight key dimensions of uncertainty:

• • technical knowledge

• • procedural familiarity

• • psychological state

• • environmental conditions

• • team dynamics

• • resource availability

• • task complexity

• • time pressure

These dimensions emerged through a constructivist analysis of interviews with 49 licensed aircraft maintenance professionals representing civil and military sectors. Unlike traditional models, UQAM does not treat uncertainty as an abstract background condition but as a measurable real-time input into safety decision making. The IUE quantifies these variables into a scalar uncertainty score that supervisors can visualise, interpret and act upon during task planning and execution. This transforms uncertainty from a latent risk into a manageable parameter, enabling predictive interventions, targeted oversight and continuous feedback within safety management systems. The UQAM framework is both theoretically grounded and practically validated. It integrates the interpretivist understanding of human experience with the precision of mathematical modelling, thereby bridging the methodological division between soft human factors inquiry and hard systems engineering. Furthermore, its compatibility with existing safety models, such as SMS, STAMP, and FRAM, offers a pathway for real-time enhancement of these frameworks, transforming them from retrospective taxonomies into adaptive, anticipatory systems. In essence, UQAM proposes a paradigm shift: from error-centric to uncertainty-aware safety governance. It equips maintenance organisations with a practical tool for quantifying human variability, enhances situational awareness at the supervisory level and contributes to the long-term evolution of predictive safety science in aviation. In the next section, the paper presents the methodological design, empirical findings, and validation outcomes that support the implementation of this new framework and argues for its integration into future-oriented aviation safety governance.

3.4 Research philosophy and methodological approach

Anchored in the interpretivist paradigm, this research posits that uncertainty in aircraft maintenance stems not solely from technical anomalies but also from contextual, experiential and inherently subjective dimensions of frontline practice. The methodology embraces the complexity of socio-technical systems and recognises safety as an emergent property shaped by dynamic operational contexts [Reference Creswell20]. Consistent with the perspective of [Reference Lincoln, Lynham and Guba67], it emphasises reflexivity, context-awareness, and the co-construction of meaning. The adopted mixed-method approach (Fig. 6) unfolded across three interconnected phases, balancing theoretical depth with empirical validation. This design ensures the resulting framework is both firmly rooted in real-world practice and adaptable for broader system-level applications.

1. 1. Qualitative exploration: In-depth, semi-structured interviews and field note observations were conducted to capture frontline perspectives on uncertainty, error genesis and safety systems.

2. 2. Model development: Thematic patterns were translated into a conceptual framework, leading to the design of the UQAM process and the derivation of the IUE.

3. 3. Computational formalisation: A structured mathematical formulation operationalised uncertainty as a scalar quantity, grounded in real-world inputs.

Figure 6. The mixed method research approach.

3.5 Data collection and participants

4.1 Thematic analysis and qualitative findings

Qualitative data were analysed using Braun and Clarke’s [Reference Braun and Clarke15] six-phase thematic analysis methodology, with a reflexive approach guided by Kiger and Varpio [Reference Kiger and Varpio59]. This combination provided a flexible yet rigorous structure to inductively generate themes while ensuring that the analytic process remained grounded in participants’ lived experiences. As presented in Fig. 7 the analytical workflow was structured around the following iterative steps:

1. 1. Familiarisation with the data: Audio recordings were transcribed verbatim and repeatedly reviewed alongside field notes to ensure a comprehensive understanding of contextual nuances, body language and environmental cues.

2. 2. Initial coding: Open inductive coding was performed manually and iteratively in NVivo, generating more than 150 unique codes across the dataset. Codes reflected descriptive and interpretive insights, often capturing relational dynamics and psychological states.

3. 3. Theme development and refinement: The codes were reviewed, clustered and abstracted into higher-order thematic categories. Themes were refined through recursive comparison and aligned with theoretical constructs and emergent operational realities.

4. 4. Conceptual mapping and synthesis: Thematic patterns were synthesised into a conceptual map of uncertainty-inducing conditions in aircraft maintenance. This map informed the structure of the UQAM framework and directly influenced the development of the IUE.

5. 5. Participant validation: A reflexive member-checking process was conducted with a sub-sample of 10 participants, who reviewed thematic summaries and confirmed their accuracy, clarity, and resonance with real-world practice. Their feedback led to the refinement of terminology and the clarification of two overlapping dimensions.

Through this robust analytical process, eight recurrent and interdependent factors emerged as primary dimensions of operational uncertainty. These factors were cited across civil and military contexts and consistently reflected in tasks ranging from routine line maintenance to complex base-level inspections. The final thematic structure is presented in Table 3 and serves as the conceptual foundation for the UQAM framework. These factors were not isolated or static; they operated as dynamic, interacting variables that fluctuated based on task type, personnel configuration and real-time operational pressures. This thematic foundation enabled the translation of rich qualitative insights into the formalised mathematical structure of the IUE, ensuring that the final model was theoretically sound, empirically grounded and operationally relevant.

Table 3. Primary dimensions of operational uncertainty in aircraft maintenance

Figure 7. Thematic map of uncertainty in maintenance tasks.

Figure 8. The UQAM process.

4.2 Design of the UQAM process

The UQAM process offers a structured, dynamic and predictive framework for quantifying operational uncertainty in aviation maintenance tasks. Grounded in real-world practice and designed for adaptability, UQAM integrates human-centric factors with mathematical modelling to support proactive safety management. As seen in Fig. 8, the process begins with targeted data collection on eight key uncertainty factors (e.g., technical knowledge, team dynamics, time pressure), which are then mapped into a dynamic framework capable of incorporating evolving variables, such as new technologies or environmental changes. Each factor is assigned a dynamic coefficient, calibrated using empirical data, expert judgement and contextual relevance. These weighted factors are synthesised using the IUE to compute a total scalar uncertainty score. This score is compared against a predefined ToU to determine whether corrective actions are required. Heatmaps visualise the distribution of factor-level uncertainty, enhancing interpretability and team-based decision-making. The process supports real-time feedback loops, updating uncertainty levels dynamically as tasks evolve. The visualisation of uncertainty contributions parallels the need to dynamically map factor propagation across maintenance environments [Reference Bohrey and Chatpalliwar11]. In its final phase, UQAM could incorporate AI-driven analysis to refine coefficient weights and thresholds over time, enabling continuous learning and alignment with shifting operational risk profiles. UQAM transforms uncertainty from a latent background risk into a measurable, actionable and adaptive safety metric through its modular design, real-time responsiveness and integration with predictive analytics.

4.3 The Integrated Uncertainty Equation (IUE)

At the core of UQAM lies the IUE, a computational model that quantifies uncertainty using task-level data. The equation mathematically represents uncertainty as a function of the interaction between factor scores and contextual coefficients.

4.3.2 Equation derivation

The IUE aims to quantify the total uncertainty $U$ associated with a specific aircraft maintenance task. This is achieved by aggregating the influence of eight uncertainty-inducing factors, each weighted by a context-sensitive coefficient. The objectives of the IUE are to (1) capture the individual contribution of each factor, (2) reflect the relative contextual importance of that contribution (via a weight), (3) preserve the independence and non-linearity of contributing dimensions and (4) aggregate all components into a scalar, interpretable measure of total task uncertainty.

Let:

• • $i \in \left\{ {1,2, \ldots ,n} \right\}$ denote the index for each uncertainty factor. In this model, $n = 8$ .

• • ${F_i} \in \left[ {1,5} \right]$ : the score for the $i$ -th factor, based on observation or expert assessment.

• • ${C_i} \in \left[ {0.5,2.0} \right]$ : the coefficient (weight) reflecting the contextual importance of factor $i$ .

• • $U$ : the task’s total scalar uncertainty score computed.

To ensure non-negativity and to amplify the influence of higher scores (i.e., high-risk factors), each factor value is squared:

$$F_i^2$$

This approach aligns with formulations found in kinetic energy, statistical variance and machine learning cost functions. Each squared factor score is then multiplied by its respective contextual coefficient:

$$F_i^2 \cdot {C_i}$$

This forms a weighted squared value that reflects both the severity and situational relevance of each uncertainty factor. To compute the combined uncertainty, all weighted squared terms are summed:

$$\mathop \sum \limits_{i = 1}^n F_i^2 \cdot {C_i}$$

To produce a normalised, interpretable scalar score, the square root is applied:

$$U = \sqrt {\mathop \sum \limits_{i = 1}^n \left( {F_i^2 \cdot {C_i}} \right)} $$

Where:

• • $U$ is the total uncertainty associated with the maintenance task.

• • ${F_i}$ is the uncertainty score for factor $i$ , where 1 represents very low uncertainty and 5 represents very high uncertainty.

• • ${C_i}$ is the real-time coefficient for factor $i$ , ranging from 0.5 (minimal contextual impact) to 2.0 (high contextual impact).

• • $n = 8$ , corresponding to the eight core uncertainty dimensions listed in Table 3.

Range of Output ( $U$ ):

• • Minimum Value: $U = \sqrt {\sum F_i^2 \cdot {C_i}} = \sqrt 4 = 2$ , represents uniformly low uncertainty.

• • Maximum Value: $U = \sqrt {\sum F_i^2 \cdot {C_i}} = \sqrt {400} = 20$ , represents uniformly high uncertainty.

4.4 Thresholds of Uncertainty (ToU)

To interpret the output of the IUE in operational terms, ToU were developed collaboratively with safety managers and engineering supervisors. These thresholds translate scalar uncertainty scores into actionable decision categories and are tailored to specific task types, as presented in Table 4. During the 12-month field validation, the ToU values were not arbitrarily set but were empirically calibrated through an iterative, data-driven process. Supervisors assigned factor scores and contextual coefficients for each maintenance task evaluated, which were then input into the IUE to produce U-values. These values were tracked alongside task performance indicators, such as deviations, delays or first-time execution issues, to assess whether elevated uncertainty levels align with observable operational risks. When U-scores consistently preceded errors or procedural breakdowns, threshold levels were adjusted downward to improve sensitivity. Conversely, if high scores were repeatedly recorded without corresponding risk manifestations, thresholds were raised to prevent over-alerting. This refinement process ensured that ToU boundaries accurately reflected both real-world variability and organisational risk tolerance.

Table 4. Risk categories and recommended actions based on uncertainty score

The final thresholds guide supervisory decision-making through a three-tiered risk model:

• • Low Risk ( $U \le 7.0$ ): Task proceeds as planned. No additional oversight is required.

• • Moderate Risk ( $7.1 \lt U \le 14.0$ ): Task should proceed with enhanced monitoring or resource adjustments, such as experienced personnel assignment or procedural reinforcement.

• • High Risk ( $U \gt 14.0$ ): Task execution should be deferred, reassigned, or escalated to a higher authority due to elevated uncertainty and potential risk exposure.

These thresholds are intentionally dynamic and responsive. They are reviewed regularly based on ongoing operational data, evolving regulatory guidance and frontline feedback. This adaptability ensures that uncertainty assessment remains aligned with aircraft maintenance environments’ complex, fluid realities.

5.1 Validation design and procedure

A supervisor-led, checklist-based scoring mechanism was implemented with safety and compliance officers. For each task, supervisors assessed eight predefined uncertainty-inducing factors $\left( {{F_1}} \right)$ to $\left( {{F_8}} \right)$ , assigning a score between 1 and 5, and a dynamic coefficient between 0.5 and 2.0. These values were then input into the IUE:

$$U = \sqrt {\mathop \sum \limits_{i = 1}^8 \left( {F_i^2 \cdot {C_i}} \right)}$$

where:

• • ${F_i}$ represents the task-specific uncertainty score for factor $i$

• • ${C_i}$ denotes the real-time contextual weight for factor $i$

Risk categorisation followed pre-established ToU, developed and refined with engineering supervisors. Supervisors were instructed not to intervene based on U-scores during the trial to preserve objectivity. Instead, scores and observations were collected for post-hoc analysis, threshold calibration and model tuning.

5.2 Evaluated maintenance tasks

Four recurring maintenance tasks were selected to test the model’s responsiveness across varied operational contexts. These tasks offered a spectrum of procedural regularity, technical depth and environmental variability ideal for evaluating the IUE’s performance resolution and flexibility. To ensure comprehensive model validation, the tasks were deliberately chosen to reflect varying levels of technical complexity, predictability and human-centric uncertainty.

5.3 Task-specific results

5.4 Heatmap analysis of uncertainty factors

Figure 9. Factor contribution heatmap: engine oil changes.

In addition to scalar uncertainty scores, the UQAM framework provides visual diagnostics through heatmaps that represent the relative contribution of each factor $({F_1}$ – ${F_8})$ , to the total uncertainty score (U). These visualisations offer a granular view of the dominant sources of task-related uncertainty, enabling supervisors and safety managers to pinpoint areas for targeted intervention, such as focused training, procedural reinforcement or environmental adjustments. Heatmaps were generated based on supervisor-assigned scores during live assessments for each of the four tasks evaluated. As illustrated in Figs 9–12, the heatmaps clearly differentiate the uncertainty profiles across tasks:

• • Engine oil change: Heatmaps showed occasional spikes in time pressure $({F_8}$ ) and psychological state $({F_3}$ ), particularly during turnaround scenarios, despite the task’s overall stability. These factors accounted for over 60% of total uncertainty in outlier cases.

• • Lubrication: Factor contributions remained uniformly low across all observations. The heatmap validated the task’s minimal uncertainty profile, with no individual factor exerting a dominant influence, confirming task predictability and model reliability in low-variance contexts.

• • Freewheel inspection: High contributions were observed from technical knowledge $({F_1}$ ) and task complexity $({F_7}$ ), especially when inspections involved first-time execution or unfamiliar system configurations. These factors accounted for over 70% of the uncertainty, reinforcing the need for skill-based mitigation strategies in complex diagnostic tasks.

• • Driveshaft inspection and lubrication: Environmental conditions $({F_4}$ ) and team dynamics $({F_5}$ ) emerged as dominant sources of uncertainty. Variability in workspace ergonomics and multi-person coordination increased the operational risk profile. The heatmap revealed distributed, multi-factor uncertainty, reflecting the complexity and interdependence inherent in this task.

Figure 10. Factor contribution heatmap: lubrication.

Figure 11. Factor contribution heatmap: freewheel inspection.

These heatmap outputs not only confirmed the validity of the eight-factor model but also demonstrated its diagnostic value in visualising uncertainty propagation. By making the source and weight of uncertainty visible at the task level, the UQAM heatmaps support real-time risk communication, enhance supervisory awareness, and provide a feedback mechanism for continuous safety improvement.

5.5 Model sensitivity

The field validation demonstrated that the IUE is highly responsive to the variability and complexity inherent in real-world maintenance operations. Across the four evaluated tasks, the IUE consistently reflected task-specific uncertainty dynamics, distinguishing between routine, moderate and elevated risk conditions without producing false alerts in stable contexts (Table 5). One of the model’s key strengths is its sensitivity to anomaly conditions. During the trial, the IUE successfully flagged elevated uncertainty in cases involving procedural deviation, task unfamiliarity, misdiagnosis and constrained work environments, each of which correlated with known operational disruptions. These outcomes confirm the IUE’s utility as a measurement tool and an early warning indicator for supervisory decision-making. The model’s scalability and adaptability were confirmed through its successful application across tasks with widely differing profiles. By dynamically incorporating contextual coefficients $\left( {{C_i}} \right)$ and enabling modular adjustments, the IUE allowed for fine-tuned application without recalibrating the mathematical structure. This modularity supports deployment across diverse maintenance environments, from line to base operations. The model also proved to be practically usable.

Table 5. Uncertainty observations, thresholds, and key contributors per task

Supervisors in the trial reported that the scoring process was intuitive and easily integrated into pre-task briefings or planning workflows. The heatmap visualisations were particularly valued for their ability to clearly communicate uncertainty sources, aiding task-level decisions and long-term safety planning.

Furthermore, the scalar output produced by the IUE was shown to integrate seamlessly with existing digital safety infrastructures, including SMS dashboards. Its real-valued output is also well-suited for future incorporation into AI tools and machine learning systems, opening the door to data-driven safety forecasting and autonomous decision support. The validation confirmed that the UQAM framework represents a substantive advancement in predictive safety modelling. Its ability to quantify human-centric uncertainty in real time and to do so in a flexible, scalable and interpretable way positions it as a valuable tool not only for post-hoc analysis but for shaping frontline interventions before error materialises. However, further testing in civilian Part-145 environments, integration with AI risk agents, and cross-sector validation are recommended to enhance generalisability and system-wide applicability. The model lays a foundational pathway for incorporating uncertainty as a live operational variable in future-ready safety architectures such as SMS, STAMP and FRAM.

5.5.1 Sensitivity analysis using the driveshaft inspection and lubrication task

To evaluate the responsiveness of the IUE to fluctuations in individual uncertainty factors, a sensitivity analysis was conducted using the driveshaft inspection and lubrication task. This task, identified as the most operationally complex during field trials, provides a representative test case due to its multi-factor uncertainty profile and elevated baseline risk score.

Using a one-at-a-time (OAT) approach, each input score ( ${F_i}$ ) was independently varied from its baseline to extreme low (1) and high (5) values, while holding contextual coefficients ( ${C_i}$ ) and other factors constant. The resulting changes in the total uncertainty score ( $U$ ) were used to calculate the impact range, reflecting each factor’s marginal influence.

As shown in Table 6, the most influential variables were Environmental Conditions ( ${F_4}$ ) and Team Dynamics ( ${F_5}$ ), which produced the highest impact ranges (up to $ \pm 40{\rm{\% }}$ ). Factors like Technical Knowledge ( ${F_1}$ ) and Task Complexity ( ${F_7}$ ) also showed considerable leverage, supporting the field-observed findings. Less sensitive variables included Resource Availability ( ${F_6}$ ) and Time Pressure ( ${F_8}$ ), though still relevant in high-stress contexts.

Table 6. Sensitivity of IUE Score to input factors (baseline $U = 9.27$ )

6.1 Theoretical contribution

UQAM addresses longstanding theoretical limitations in dominant safety models – SMS, STAMP and FRAM – by introducing a formal mechanism to measure and model human-centric uncertainty in real time. While these models acknowledge complexity and emergent risk, they largely lack the capacity to assign operational value to uncertainty before errors occur. The IUE bridges this gap by transforming subjective, qualitative risk factors into measurable, scalar outputs that are actionable during planning and execution. This aligns with contemporary calls for nonlinear, probabilistic and adaptive models in safety-critical systems. Moreover, by grounding the framework in field-based qualitative data and operational context, the study offers a theoretical model that respects human factors’ interpretive dimensions while enhancing their integration into systems engineering and predictive analytics.

6.2 Methodological innovation

Methodologically, UQAM represents a hybrid model that blends constructivist inquiry with computational modelling. Using thematic analysis to derive eight operational uncertainty factors ensured that the model reflects lived frontline experience, not just abstract constructs. The IUE’s mathematical formulation, validated through a year-long field study, demonstrates how mixed-methods research can move from conceptual exploration to formalised, scalable tools. This hybrid approach also resolves a key limitation in the human factors field: the lack of quantitative methods for operationalising uncertainty in day-to-day maintenance activities. UQAM thus sets a precedent for developing rigorous, field-responsive models in other high-risk domains.

6.3 Practical applications and system integration

In practical terms, UQAM delivers a decision-support tool that is both intuitive and operationally impactful. The factor-based scoring system and associated heatmap visualisations provide clear, real-time insights into the sources and severity of task-related uncertainty. This supports informed supervisory decisions such as reallocating resources, adjusting schedules or delaying high-risk tasks. The framework is also designed for seamless integration into existing safety systems. As proposed in Figs 13–15, by embedding scalar uncertainty scores into SMS workflows and enhancing STAMP and FRAM with real-time inputs, UQAM strengthens feedback loops, enhances hazard visibility and supports dynamic intervention, transforming these frameworks from retrospective audits into proactive safety engines. Furthermore, the IUE’s real-valued outputs are AI-compatible, enabling future applications, such as intelligent task planning, automated threshold tuning, and predictive risk detection through machine learning pipelines. By embedding AI-driven knowledge bases and predictive diagnostic tools into maintenance ecosystems, promises to transform aviation safety from reactive to proactive risk governance [Reference Kabashkin and Perekrestov56].

Figure 12. Factor contribution heatmap: driveshaft inspection.

Figure 13. UQAM potential integration into STAMP.

Figure 14. UQAM potential integration into FRAM.

Figure 15. Reshaping the SMS with UQAM.

While the UQAM framework demonstrated strong usability and interpretability in field trials, certain limitations merit attention. The scoring of uncertainty factors, although guided by structured checklists, remains partially subjective and may vary depending on the assessor’s experience, fatigue or situational awareness. Future iterations may benefit from incorporating automated or sensor-based inputs to complement human judgement. Additionally, the introduction of new scoring and visualisation tools may encounter cultural or operational resistance from supervisors accustomed to established routines. Early engagement, training and integration into existing workflows are essential to ensure acceptance and sustained use.

6.4 Regulatory and industry relevance

Given the industry’s increasing push toward predictive, performance-based safety management, UQAM aligns with evolving expectations under EASA Regulation 1321/2014, EMAR-145, and ICAO Annex 19. Its emphasis on real-time assessment, frontline usability and feedback-driven calibration directly supports emerging adaptive and evidence-based safety oversight requirements. UQAM provides a pathway for transitioning from compliance-driven audits to risk-intelligent operations – a shift vital to sustaining safety in increasingly complex maintenance ecosystems.