Executive leadership exerts a critical influence on workplace safety and health (WSH) outcomes. In 2024, the Singapore government introduced the mandatory Top Executive WSH Programme (TEWP) to strengthen this capacity. However, its effectiveness remains uncertain due to the absence of validated instruments tailored to the executive tier. To address this gap, this study integrates expectancy-value theory and paradoxical safety leadership to develop and validate a seven-dimensional executive safety leadership instrument. This instrument captures two executive paradoxical safety leadership (ePSL) behavioural strategies as well as five cognitivemotivational antecedents. Using an international online panel of business owners and senior managers (N = 484), this study established initial evidence of instrument's reliability and factor structure. Then, the instrument was cross-validated using the field cohort (N = 230) and applied to evaluate the TEWP with a paired sample of 130 participants. Confirmatory factor analysis conducted on both international (CFI = 0.951; SRMR = 0.036) and field cohorts (CFI = 0.933; SRMR = 0.045) demonstrated acceptable model fit, providing initial evidence for the instrument's factorial validity across contexts. The paired sample t-test revealed statistically significant preto-post improvements across all dimensions. Explanatory regression analyses indicated that among the cognitive-motivational antecedents, only self-efficacy in adaptive decision-making consistently predicted positive shifts in both interpersonal and task-focused ePSL behavioural strategies. Practically, this study provides an instrument for TEWP evaluation; theoretically, it highlights a capability to regulate decisions under pressure as a proximal foundation for ePSL enactment, the critical need to pivot executive development toward cultivating adaptive decision-making.

Construction safety inspection remains a critical yet challenging task due to complex site environments, dynamic worker activities, and frequent visual occlusions. While recent advances in computer vision and vision-language models (VLMs) have enabled automated safety reasoning, most existing systems rely on single-agent, single-view observations, which are inherently vulnerable to occlusion, viewpoint bias, and incomplete scene understanding. There is a lack of mechanisms for consistent multi-view interpretation and reliable visual grounding. These limitations restrict the applicability in large-scale, safety-critical construction scenarios. To address these challenges, this paper proposes a multi-embodied-agent, dual-level VLM framework for construction safety inspection. The framework is structured as a four-layer architecture comprising coordinated robotic data acquisition, view-level intelligence, view-to-global object mapping, and globallevel intelligence. At the sensing level, multiple autonomous agents collaboratively collect synchronized, risk-aware, and diversity-driven multi-view visual data. At the perception level, conventional CNNbased object detection and tracking modules provide accurate localization and identity consistency, while lightweight view-level VLMs perform localized safety assessments. At the reasoning level, cross-view object observations are aligned into globally consistent identities, enabling a global-level VLM to conduct crosstime and cross-view consistency reasoning and generate reliable, explainable site-level safety decisions. To validate the feasibility of the proposed approach, experiments are conducted on a public construction safety dataset to evaluate the view-level VLM component. Quantitative and qualitative results demonstrate that the model achieves strong safety compliance reasoning and interpretable outputs even with limited training data. Overall, the proposed framework provides a scalable and robust foundation for multi-agent, VLM-enabled construction safety inspection.

Construction project risk management is a knowledge-intensive task that requires practitioners to identify risks, trace causal mechanisms, and develop mitigation strategies. Expert project managers build sophisticated mental models through years of experience, yet this valuable tacit knowledge often remains inaccessible to less experienced practitioners. While recent advances in Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) provide opportunities for intelligent decision support, current systems face two limitations: (1) knowledge bases depend on coarse-grained documents that cannot capture directional causal logic or combine diagnostic and prescriptive knowledge; (2) existing approaches are divided between purely neural systems with direction-agnostic retrieval and purely symbolic systems requiring formal query languages, lacking a unified architecture that exploits the strengths of both. This study introduces the intent-driven hybrid neuro-symbolic RAG (NSRAG) framework, built on an Expert-driven Risk Causal-Action Knowledge Graph (ERCA-KG). The ERCA-KG converts tacit expert knowledge into a cohesive causal-actionable structure that merges diagnostic causal chains with prescriptive management strategies. The NSRAG uses analytical intent classification to dynamically coordinate symbolic graph traversal operations, enabling direction-aware retrieval that matches reasoning pathways to query needs. Empirical results show that this architecture achieves 84.17% semantic knowledge graph coverage, significantly surpassing traditional vector-based and text-based baselines. Ablation studies demonstrate that intent-driven orchestration improves Normalised Discounted Cumulative Gain from 0.455 to 0.698 compared to undirected graph traversal. These findings confirm that unified causal-actionable knowledge modelling, combined with intent-driven neuro-symbolic coordination, effectively bridges the gap between implicit human expertise and explicit computational knowledge access for construction project risk management.

Construction project management programmes struggle to provide timely and personalised feedback at scale. This paper developed and evaluated an AI feedback system that combines a large language model (LLM) with retrieval-augmented generation (RAG) to deliver personalised messages. A design-based study trialled the feature in two settings, an in-person workshop and an online course, with 81 participants. Mixed methods were used through a perception questionnaire, interviews, and focus groups. Ratings were positive across constructs, with no significant differences between delivery modes. Regression analysis revealed that engagement and perceived fairness independently predicted the intention to continue using the tool. Thematic analysis identified five design considerations: clarity to reduce cognitive load, deeper diagnosis with actionable guidance, role-relevant personalisation, a motivational tone with reflective prompts, and transparency to sustain trust. This paper presents a practical LLM-RAG pipeline, provides evidence of acceptance, and offers practical guidance for practitioners on AI-generated feedback in construction management.

Current construction project risk analysis depends heavily on manual processes and subjective expertise. This approach produces risk knowledge that is inherently subjective, inconsistent, and difficult to verify or scale. From an engineering informatics perspective, this represents a critical gap in the computational formalisation of domain knowledge. Besides, existing analytical methods are inadequate as they examine discrete risks or correlations without capturing causal mechanisms. This leads to fragmented interventions that address symptoms rather than root causes. To bridge these gaps, this study develops an automated, data-driven causal analysis framework, which integrates ensemble causal discovery (ECD) with PLS-SEM validation to uncover underlying causal mechanisms directly from data. This process effectively transforms tacit risk knowledge, previously confined to expert intuition, into an explicit and verifiable computational structure. The proposed ECD method addresses the inherent instability of individual causal discovery algorithms and mitigates overfitting risks. The subsequent PLS-SEM validation provides rigorous statistical testing, ensuring the final model is both robust and interpretable. Applied to data from 229 participants in Singapore, the framework automatically established a Project Risk Causal Network (PRCN). The PRCN explained 51% of project performance variation, outperformed individual algorithms, and revealed critical risk propagation pathways. This study advances engineering informatics by: (1) automating causal knowledge extraction from project data; (2) providing robust validation for discovered causal structures; and (3) enabling targeted interventions through quantified causal mechanisms.

Due to the significant advantages of industrial robots in production, they are increasingly used in the workplace, resulting in attention being paid to the safety issues of human-robot interaction (HRI). However, existing research still has gaps in understanding the patterns of relationships between robot characteristics, robot-human errors, and the physical working environment. To address the knowledge gaps, this paper aims to identify and examine the patterns of the relationship among robot characteristics, robot-human errors, and physical working environments and investigate how the patterns evolve along the technological advances in robotics design. This paper analyses 303 HRI accident reports by applying a network analysis. Based on the cluster analysis, seven HRI incident archetypes were identified, including (1) unexpected activation, (2) faulty commands, (3) blind automation danger, (4) sensor and signal communication errors, (5) ergonomics-related injuries, (6) secondary robot intrusion, and (7) classic hazard pitfalls in robot-assisted work. Temporal analysis reveals that 'Archetype 1: unexpected activation' consistently dominated, accounting for over 60 % of accidents, and warrants the most attention in future safety management. Additionally, the increasing frequency of 'Archetype 4: sensor and signal communication errors' in later stages highlights the growing need for targeted interventions". This paper is the first to identify and categorize HRI incident archetypes systematically. It offers a useful framework for researchers and practitioners. These archetypes provide a structured tool for systematically investigating and diagnosing incidents and can also help workers and managers understand the patterns of relationships between these factors in different HRI scenarios.

Design for Safety (DfS) offers a proactive strategy to address hazards during design to reduce accidents and ill health during construction, operations, maintenance, and demolition activities. However, current DfS regulations only provide broad duties and requirements, while DfS reviews operate as an isolated process that lacks a systemic approach. Hence, a critical gap exists in adopting a systems approach to DfS, where interconnected elements work cohesively to clearly define DfS goals and iteratively plan, execute, review, and improve the system to achieve better safety outcomes. Existing workplace safety and health management systems do not adequately address DfS requirements in sufficient detail, highlighting the need for a specialized framework. This research develops a comprehensive DfS management framework (DfSMF) through mixed-methods analysis of 61 peer-reviewed articles. The study identifies 25 critical success factors organized into nine elements mapped to ISO 45001. The three most significant elements, E3 Capability Support (contributing 19.58% to the total importance of the framework), E2 DfS Planning (17.57%), and E1 Leadership, Commitment, and Engagement (16.45%), collectively constitute over half of the total weighted importance score, demonstrating that human factors and systematic planning form the foundation of effective DfS implementation. The remaining elements include E8 Performance Evaluation, E6 System Support, E7 Design Risk Management, E9 Improvement, E5 DfS Tools Support, and E4 Resources, which serve as enablers rather than primary drivers of effective DfS. This evidence-based framework offers practical guidelines for effective management and continual improvement of DfS. The DfSMF transforms the current fragmented DfS approach into a holistic management system that enhances DfS outcomes and project performance.

Although poor housekeeping leads to construction accidents, there is limited technological research on it. Existing methods for detecting poor housekeeping face many challenges, including limited explanations, lack of locating of poor housekeeping and annotated datasets. To address these challenges, this paper proposes the Housekeeping Change Detection Network (HCDN), integrating a feature fusion module and a large vision model. This paper introduces the approach to establish a change detection dataset (Housekeeping-CCD) focused on construction housekeeping, along with a housekeeping segmentation dataset. Experimental results of our Housekeeping-CCD dataset demonstrate that HCDN outperforms existing state-of-the-art (SOTA) methods, achieving average accuracy (89.32 %), mean IoU (76.97 %), and mean F-score (86.67 %). The contributions include significant performance improvements compared to existing methods, providing an effective tool for enhancing construction housekeeping and safety.

Current expert-based approaches to determining the weights of different safety management elements during contractor safety performance are time-consuming and potentially biased.Hence, this paper evaluates analyticsbased approaches, i.e., supervised learning, cluster-then-predict and two-level variable weighting K-Means (TWKM) (an extension of the traditional K-Means clustering algorithm), against the Delphi method. In collaboration with an infrastructure developer, a dataset of 461 data points and 12 features describing subcontractors' inherent risks and safety assurance performance were collected. This paper showed that supervised learning improves recall by 21 % when compared with the Delphi method. This paper also highlights that changes in input features' distributions (or covariate shifts) across construction stages and projects can reduce the recall of the supervised learning model from 93 % to 50 %. Key academic and practical contributions include the analytics-based approaches to develop weights for measuring contractors' safety performance, and strategies to manage the impact of covariate shifts on accuracy of feature weights.

There is growing interest in using safety analytics and machine learning to support the prevention of workplace incidents, especially in high-risk industries like construction and trucking. Although existing safety analytics studies have made remarkable progress, they suffer from imbalanced datasets, a common problem in safety analytics, resulting in prediction inaccuracies. This can lead to practical problems, e.g., incorrect resource allocation and improper interventions. To overcome the imbalanced data problem, we extend the theory of accident triangle to claim that the importance of data samples should be based on characteristics such as injury severity, accident frequency, and accident type. Thus, three oversampling methods are proposed based on assigning different weights to samples in the minority class. We find robust improvements among different machine/deep learning algorithms. For the lack of open-source safety datasets, we are sharing three imbalanced datasets, e.g., a 9-year nationwide construction accident record dataset, an accident and safety management dataset, and a US truck driver safety climate survey dataset, and their corresponding source codes.

Migrant workers face vulnerability and unique mental health challenges in their host countries. Housing designs may affect the residents’ mental health, yet few studies have investigated this relationship in the migrant worker dormitory context. In this study, we investigated this question by surveying the design factors of migrant workers’ dormitories and their mental health states measured by the short version of the Warwick–Edinburgh Mental Wellbeing Scale (SWEMWBS). We found that those migrant workers’ mental health is correlated with whether one has a pleasant window view, defined as green, blue (water feature) or long-distance view. The findings suggest that the elements of nature in architectural design to residents’ mental health are important in the context of migrant worker dormitories. It is vital to incorporate more elements of nature in the design and renovation of migrant worker dormitories, especially the quick-built dormitories, to improve workers’ mental health. This study is especially timely in informing regulations on the standard of dormitory design by drawing attention to the mental health considerations of such designs and highlighting the importance of elements of nature in improving migrant workers’ mental health.

The construction industry has persistently high accident rates, underscoring the urgent need for effective risk mitigation during the design phase. Design for safety (DfS) regulations are intended to eliminate or reduce hazards at the source, but their implementation often focuses on documentation rather than meaningful risk reduction, limiting their impact on safety outcomes. This study aims to classify organizational DfS compliance behaviors in Singapore's construction industry under mandatory regulation and to develop a two-axis model that explains these variations. Based on qualitative analysis of 59 semistructured interviews with industry professionals, the study categorizes compliance behaviors along two axes: the primary focus of the project team (fulfilling regulatory intent versus avoiding penalties) and the level of effort deployed to support this focus. Five types of compliance behaviors were identified: deep compliance, alternative deep compliance, enlightened surface compliance, extravagant surface compliance, and surface compliance. These findings reveal diverse organizational interpretations of DfS, from proactive risk elimination to minimal regulatory adherence. By extending the concept of surface and deep compliance from individuals to organizations, this study provides a nuanced framework for assessing and improving DfS implementation. The findings highlight the importance of aligning regulatory strategies with organizational behaviors to drive meaningful risk reduction. The proposed two-axis model categorizes DfS compliance behaviors in construction industry, identifies how situational and organizational factors shape these behaviors, and provides practical tools for regulators, trainers, consultants, and safety professionals to design targeted interventions and enhance alignment between regulatory intent and organizational safety practices.

The construction industry's high-risk environment demands effective hazard recognition strategies. Attention, a critical cognitive process, plays a crucial role in this task. Previous research focused on individual attention process, such as sustained attention, selective and divided attention. However, no research has been conducted to investigate the effects of the interplay between endogenous and exogenous factors on hazard recognition in construction settings. This paper aims to investigate the effects of the interplay between top-down (T-D) and bottom-up (B-U) attention networks on hazard recognition, using immersive virtual reality (IVR), eye tracking (ET), and electroencephalography (EEG). Two safety interventions-augmented stimuli and toolbox meetings-were tested in a dynamic IVR construction site. The results showed that both augmented stimuli and the safety toolbox meeting significantly affected B-U, T-D, and hazard recognition. This paper provided evidence that the interplay between B-U and T-D can significantly improve workers' hazard recognition performance. The results improved our understanding of the mechanisms that control selective attention and the source of guidance over attention orientation. By demonstrating that T-D and B-U processes can work together rather than in isolation, this research contributes a key theoretical insight: attentional orientation in hazardous construction environments is neither fully determined by external stimuli nor entirely controlled by internal cognitive sets. In addition, this paper highlights and calls for an integrated approach to improving worker's hazard recognition performance, by combining digital-technology-enabled stimuli with safety-goal-oriented training and managerial practices.

Introduction Decision-making failure is a critical factor influencing workers’ unsafe behavior. While previous research has predominantly concentrated on rational decision-making failures, it has largely ignored the impact of irrational decision-making failures and the mechanisms through which these failures contribute to unsafe behaviors. This oversight limits our understanding of how unsafe behavior develops. Utilizing dual-process theory and the prototype willingness model, this study distinguishes between two types of decision-making failures: unsafe behavioral intention (reflecting rational decision-making failure) and unsafe behavioral willingness (reflecting irrational decision-making failure). We examined how these failures influence workers’ unsafe behavior and explored the moderating effects of conscientiousness and consideration of future consequences on these relationships. Method Data were gathered from 446 construction workers through a questionnaire survey. Structural equation modeling and hierarchical regression analysis were used to test the hypotheses. Results The results revealed that: (1) both unsafe behavioral intention and unsafe behavioral willingness significantly and positively affected workers’ unsafe behavior; (2) unsafe behavioral intention mediated the relationship between unsafe behavioral willingness and unsafe behavior; (3) conscientiousness negatively moderated the link between unsafe behavioral intention and unsafe behavior; and (4) the moderating effect of consideration of future consequences on the relationship between unsafe behavioral willingness and unsafe behavior was not supported. Practical Applications This study provides a comprehensive perspective on the relationship between decision-making failures and unsafe behavior, elucidates the underlying mechanisms, and offers targeted strategies for managing unsafe behaviors.

Heightened global concern regarding hazard detection failure among construction workers, resulting in unsafe behaviors and potential accidents, has escalated. Nevertheless, there is still a lack of understanding regarding the causes of hazard detection failure. To bridge the gap, this study explores the mechanism of hazard detection failure from the matching perspective of hazard detection capability and hazard detection task demand. First, 35 participants were recruited to participate in an eye-tracking experiment, during which they examined 44 photos to identify potential hazards. Second, polynomial regression with response surface analysis was used to explore how different degrees of matching between hazard detection capability and task demand impact workers' hazard detection failure. Finally, the moderating effect of fatigue was examined through hierarchical multiple regression analysis. The results reveal that hazard detection failure is significantly negatively correlated with hazard detection capability and significantly positively correlated with task demand. Additionally, the primary causes of hazard detection failure are undermatching and high-level matching between hazard detection capability and task demand. Fatigue positively moderates the relationship between the degree of matching and hazard detection failure. Academically, this study significantly enhances the theoretical understanding of hazard detection failure and cognitive failure among construction workers. Practically, it provides safety managers with a structured process and actionable methods to address hazard detection failure among construction workers, including tools for evaluating workers' hazard detection capabilities and task demands.

Risk perception is crucial for minimizing injuries within the construction industry. However, existing research mostly focuses on the impact of demographic and external factors on construction workers' risk perception, with little attention to deeper cognitive factors such as cognitive capability and cognitive task demand. Moreover, even less research explores the balanced relationship between these factors, leaving the mechanism of risk perception failure unclear. To bridge this gap, this study explores the mechanism of risk perception failure from the balance perspective of risk perception capability and risk perception task demand. This study employed behavioral experiments, recruiting construction workers as participants, and required them to scan construction site photos and to self-report the probability and severity of risks. Polynomial regression with response surface analysis explored the impact of the different combinations of risk perception capability and task demand on risk perception failure. The moderating effect of risk propensity was also examined. The results find that risk perception capacity decreases risk probability and severity perception failure, while risk perception task demand increases them. Additionally, the risk probability and severity perception failure continue to increase as risk perception capacity and task demand move from "high-quality" imbalance to balance, and then to "low-quality" imbalance. Unexpectedly, a U-shaped trend is observed as both factors move from low-level to high-level balance. Risk propensity moderates the balance's effect on risk probability perception failure but not on risk severity perception failure. Academically, this study significantly advances the theoretical understanding of the mechanism of risk perception failure and even cognitive failure of construction workers. It also offers practical implications for construction safety managers to mitigate workers' risk perception failure, thereby controlling unsafe behaviors.

Construction sites are dynamic and hazardous environments where workers often struggle to maintain high levels of situation awareness (SA), essential for effective hazard recognition. While technologies exist to aid hazard perception, limited research has explored how external environmental stimuli and internal safety goals jointly influence the SA transition across perception (SA1), comprehension (SA2), and projection (SA3). This study investigates the effects of augmented stimuli and safety goals setting on SA levels, SA transition and hazard recognition. A multimodal experimental approach was employed, integrating virtual reality (VR), eye tracking, modified Situation Awareness Global Assessment Technique (SAGAT) and event-related potentials (ERPs). A novel Temporal Hybrid Situation Awareness Measurement (THSAM) method was introduced to quantify SA by linking eye-tracking data with SAGAT responses. SAGAT data showed that both augmented stimuli and safety goals improved SA across all levels. SAGAT and THSAM indicated that the combination of the two interventions led to the largest improvements across SA1, SA2, and SA3. SA transition analysis revealed that augmented stimuli effectively facilitated the shift from unawareness (SA0) to SA1. THSAM and SA transition analysis confirmed safety goals primarily enhanced SA2. ERPs analyses further indicate distinct brain activity patterns (P2 and N400) associated with each SA level. This study contributes to construction safety research by providing quantitative evidence on the cognitive and neural mechanisms underlying SA transition. It also introduces THSAM as a methodological advancement for capturing real-time SA dynamics and offers practical implications for designing integrated safety interventions that align with workers' goals and environmental demands.

Systems thinking has emerged as a necessary approach to effective management, including occupational health and safety. Considering the recent publication of the international standard ISO 45001:2018 on occupational health & safety management systems (OHSMS), this study evaluated the extent to which it incorporates systems thinking aspects derived from the literature. The qualitative assessment of the standard against eight systems thinking tenets, each of them referring to the internal and external organisational context, revealed an adequate degree of their coverage. Overall, the effective implementation of the new standard could encourage organisations to move beyond a focus on individual system components and better understand and manage their socio-technical system by contemplating the multiple interactions amongst its elements. However, our findings also indicated a tendency of ISO 45001:2018 to focus more internally, view external agents mainly as constraints and not opportunities and implicitly promote a highly systematic approach that does not recognise local adjustments to manage variability. Subject to the need for field research to investigate the above in organisations implementing the new standard, the results from this study can raise the awareness of the industry about strengths and possible gaps when implementing their OHSMS and inform international and regional bodies when introducing new standards or revising existing ones that focus heavily on socio-technical systems. Additionally, the inclusive list of systems thinking tenets developed as part of this research can function as an updated reference for studies about systems thinking in any socio-technical environment.