Engineering

Abstract: Ensuring safety of lithium-ion batteries (LIBs) requires accounting for complex degradation mechanisms and uncertain failure conditions that can lead to thermal runaway. This study developed an integrated risk assessment methodology combining Fault Tree Analysis (FTA) with fuzzy logic to identify the main failure pathways and to estimate the likelihood of thermal runaway initiation conditions for a generic Li-ion battery system. The fault tree incorporated abuse conditions (mechanical, electrical, thermal) and cooling system faults, including accelerating mechanisms such as ageing and other stress conditions, which are often neglected in other studies. Expert judgments were translated into fuzzy sets, aggregated using the Similarity Aggregation Method, and converted into probabilistic values for top-event evaluation. The importance analysis identified BTMS failure, cycling ageing, cooling-system fouling, and high ambient temperature as the most influential modeled basic events. Comparison with previously published studies confirmed consistency in the magnitude and relevance of key failure events. The methodology provided a robust and general framework for LIB safety assessment under uncertainty data conditions, independently on specific cell chemistry or application and the framework is directly transferable to battery packs and battery management systems of electric vehicles, ships, and aircraft.

Abstract: Proton exchange membrane (PEM) water electrolysis is a key technology for converting renewable electricity into green hydrogen, and the availability of electrolyser stacks directly determines the hydrogen output and economics of such plants. This paper extends a previously reported ten-state semi-Markov reliability model of a PEM electrolyser into a twenty-state framework that better reflects field behavior. The extension adds a derated, partial-output operating state for each of four degradation mechanisms (membrane, electrode, chemical and thermal), a spares-waiting state that separates logistics delay from active repair, a scheduled preventive-maintenance branch, and a rare catastrophic-shock branch, all treated as competing risks at the operative state. Closed-form steady-state expressions are derived for availability, mean time to system failure (MTSF) and maintenance-crew busy periods. A delta-method procedure converts sensitivity coefficients into a confidence interval on availability, and a profit function balances hydrogen revenue against inspection, derating, logistics and repair costs. Using plant rates from a pilot-scale electrolyser and literature-informed values for the new parameters, the model gives a steady-state availability of 0.8804 (95% CI: 0.778–0.983), an MTSF of about 5270 h and an illustrative profit rate of $383.28/h. The results show that models omitting preventive maintenance and shock-type failures overstate electrolyser availability, with direct consequences for hydrogen supply planning and plant economics.

Abstract: Safety risks of large-scale water conservancy, transportation, and energy infrastructure emerge from coupled geological conditions, extreme weather events, and construction disturbances throughout construction and operation phases. These risks are characterized by concealment, wide-area distribution, and dynamic evolution. Limited by sensor deployment locations and personnel accessibility, conventional ground sensors and manual inspections struggle to detect regional geological hazards outside project boundaries. Space–air–ground remote sensing observations, including Synthetic Aperture Radar (SAR) satellites, optical satellites, and Unmanned Aerial Vehicle (UAV) imagery, support wide-area non-contact monitoring and capture spatial information such as slope deformation, land cover anomalies, and landslide precursors. However, single-source remote sensing is vulnerable to cloud and fog contamination and temporal decorrelation. It also cannot reflect internal structural conditions, which demands feature-level fusion with ground-based monitoring data. This study takes multi-source remote sensing as the core modality and integrates ground-based time-series sensors, on-site audio-visual recordings, and engineering archive texts. A space–air–ground collaborative multimodal safety risk situational awareness framework is established for major infrastructure projects. Following the pipeline of feature extraction, cross-modal spatiotemporal alignment, and adaptive attention fusion, the framework enables risk perception, deduction and assessment, and forms a closed-loop prevention chain of “cognition–deduction–decision–optimization”. Simulations and field measurements show that the proposed model achieves an overall risk perception rate of 96%, early warning accuracy above 94%, and a 63% reduction in response latency compared with conventional schemes. The remote sensing branch achieves a recall rate of 92.7% for geological hazard identification. Ablation experiments demonstrate that removing remote sensing inputs reduces the overall perception rate to 90.2%, accompanied by a notable increase in undetected geological hazards. This work delivers a technical solution for intelligent risk prevention and control across the full life cycle of major projects. It advances the shift of engineering safety management from passive response to proactive early warning, and provides theoretical and practical support for remote sensing-enabled infrastructure safety monitoring.

Abstract: Current attempts to instill internal alignment mechanisms into modern AI systems remain brittle when applied to autonomous, tool using agents facing out of distribution adversarial attacks, multi-step evasion strategies, and context smuggling. To date, none of the attempts at instilling internal alignment safety protocols have computationally modeled the approach developed by evolution to guide humans away from rogue behaviors and towards socially aligned ones. In human neurobiology, behavioral safety is enforced through hierarchical, intrinsic executive inhibition mediated by the prefrontal cortex (PFC). When PFC function is impaired by disease, trauma, or developmental conditions, disinhibition, poor judgment, and rule-violating behaviors emerge. Drawing on this biomimetic principle, we propose and prototype a structural containment framework-the Artificial Prefrontal Cortex (aPFC)-that provides an external layer of executive control over LLM based autonomous agents. Using Gemma 4-e4b and a Python based implementation, this pa- per presents a prototypical executive control architecture that remained robust under diverse adversarial conditions, including authority framing jailbreaks, multi node collusion, stealth payload mutation, context smug- gling, buffer flushing, and long horizon evasion. The architecture intro- duces a decoupled Executive Control module composed of four functional analogues inspired by human PFC organization: (1) a deterministic reflex layer providing zero latency constraint enforcement for known vulnera- bilities; (2) a coordination and perseveration detector that terminates unauthorized multi agent swarming and repeated evasion loops; (3) a se- mantic working memory evaluator that uses an LLM to assess the intent and consequences of proposed actions; and (4) an event driven temporal tracker that consolidates multi step behaviors to mitigate buffer flushing and “temporal amnesia” attacks. Empirical validation in an air-gapped simulation environment demon- strated that the system autonomously intercepted authority framing jail- breaks, multi node collusion, stealth payload mutations, and long horizon evasion tactics. These findings suggest that securing autonomous AI sys- tems requires moving beyond current internal model alignment and toward structural executive containment, providing a scalable and model agnostic blueprint for next generation AI safety.

Abstract: Carbon capture, utilization, and storage (CCUS) supply chains depend on injection nodes whose loss of capacity can propagate upstream as curtailment and network disruption. This design science study develops a digital twin ready predictive risk artifact that converts subsurface measurements into governed states rather than treating model probability as automatic authority. Eighteen slip events from two rough and one smooth fault were represented by seven mechanics aligned variables. Source domain leave one event out probabilities were mapped to Watch (60 s), Mitigate (30 s), and Hold (10 s) states using source calibrated negative exposure quantiles and persistence rules. A cross fitted Mahalanobis support gate identified states outside source experience. Rough to smooth Watch and Mitigate policies intercepted all six events with median warnings of 16.5 and 3.5 s and no pre horizon intervention burden. Smooth to rough policies also intercepted all twelve events but incurred 22.9 and 42.2 s/event of burden. At the 99th percentile support cutoff, support eligible coverage within 30 s was 91.7% versus 36.4% in reverse, and within 10 s it was 75.0% versus 11.7%. These specimen limited results support an Industry 5.0 architecture that separates predictive risk, domain validity, human authority, and physics based fallback.

Abstract: The ubiquitous integration of lithium-ion batteries (LIBs) into modern society, powering applications from portable electronics to grid-scale energy storage and electric mobility, has been paralleled by a growing concern over their safety. The high energy density that makes LIBs so advantageous also renders them susceptible to catastrophic failure modes, most notably fire and explosion. This review provides a comprehensive and multidisciplinary analysis of the phenomenon of battery fires. We begin by delineating the fundamental root causes of thermal runaway, the critical event precipitating combustion, including mechanical abuse, electrical abuse, thermal abuse, and manufacturing defects. The review then delves into the intricate mechanisms of thermal runaway, tracing the sequence of exothermic reactions from Solid Electrolyte Interphase (SEI) decomposition to the violent reactions of cathode materials and electrolyte. A detailed examination of the complex combustion chemistry of LIBs is presented, highlighting the roles of gaseous, liquid, and solid components as fuels and the generation of toxic and flammable effluents. The propagation of failure within a battery module or pack is discussed, underscoring the challenges in containing single-cell incidents. We systematically catalog and critically evaluate the current state-of-the-art mitigation strategies, spanning intrinsic safety approaches (novel electrolytes, solid-state batteries, advanced electrode materials) and extrinsic safety systems (battery management systems, thermal management systems, and fire suppression technologies). Finally, the review identifies persistent knowledge gaps and outlines future research directions, emphasizing the need for multi-scale modeling, advanced diagnostics, and the development of inherently safe next-generation battery chemistries. This work aims to serve as a foundational reference for researchers and engineers across academia and industry, fostering the development of safer energy storage systems.

Abstract: The present work provides a detailed experimental and numerical study on suppression of a heptane medium-scale pool fire by applying water mist. The customarily described well defined structure of the flame does not exist any longer by application of water mist. It is observed that the size and shape of the flame vary randomly and continuously due to sudden and short mist vaporization periods. An injection of the finely divided water mist (50μm) into a buoyancy-induced fire has a combustion-supporting effect of a heptane pool fire. In no mist case, contribution of the radiation heat flux to the pyrolysis rate of liquid fuel seems predominant. However, with addition of water mist, an increasing trend in burning rate is related mainly to the significance of convective heat flux due to moving of the flame region towards the liquid surface. This behavior suggests that extinguishment is achieved by total clearance of the liquid surface rather than from reduction in flame-to-liquid surface heat feedback. The numerical work provides the detailed temperature and species fields in addition to radiation/convection heat fluxes over the liquid surface during the addition of a water mist by using various drag force models.

Abstract: Noise barrier tunnels are increasingly used to mitigate transportation noise in densely populated areas; however, their enclosed or semi-enclosed acoustic environments can create significant challenges for emergency evacuation broadcasting during fire incidents. In particular, highly reflective tunnel surfaces and complex sound propagation characteristics may degrade speech intelligibility and reduce the effectiveness of emergency information delivery. This study investigated the acoustic performance of directional loudspeakers for emergency broadcasting systems in noise barrier tunnels through measurements conducted in both a semi-anechoic chamber and an actual tunnel environment. Two directional loudspeakers with diameters of 5 and 15 inches were evaluated under different installation heights, source–receiver distances, and directivity angles. Sound pressure level (SPL) and Speech Transmission Index (STI) were used as the primary performance indicators. The results obtained in the semi-anechoic chamber showed that the 15-inch loudspeaker generally achieved the target SPL of 80 dB(A) or higher under most measurement conditions, whereas the 5-inch loudspeaker exhibited greater variations depending on distance and directivity angle. In the actual noise barrier tunnel, however, speech intelligibility was substantially reduced because of excessive reverberation. Reverberation times of 12.4 s at 125 Hz and 9.8 s at 250 Hz were measured, and the maximum STI value remained below the target performance criterion of 0.60. The results indicate that achieving sufficient SPL alone is insufficient to ensure effective emergency communication in highly reverberant noise barrier tunnels. Instead, acoustic design strategies should prioritize reverberation control through sound-absorbing treatments and appropriate loudspeaker placement. These findings provide fundamental data for the acoustic design and development of emergency evacuation broadcasting systems capable of improving speech intelligibility and evacuation safety in noise barrier tunnels.

Abstract: Building fires and industrial thermal disasters rank among the most devastating hazards confronting modern societies, causing thousands of fatalities, billions of dollars in economic losses, and lasting environmental damage each year. The increasing complexity of built environments — including high-rise structures, underground facilities, and chemical processing plants — demands a commensurate advancement in emergency rescue technologies. This review provides a comprehensive examination of state-of-the-art technologies employed in the detection, suppression, and rescue operations associated with building fires and industrial thermal disasters. Key domains covered include: early fire detection systems integrating the Internet of Things (IoT), artificial intelligence (AI), and machine learning; advanced fire suppression technologies such as water mist, gaseous agents, and high-expansion foam; autonomous and semi-autonomous rescue robotics; unmanned aerial vehicles (UAVs) equipped with thermal imaging; Building Information Modeling (BIM) and digital twins for emergency planning and evacuation simulation; personal protective equipment (PPE) advancements including biometric wearables; and communication and situational-awareness systems. Special attention is given to industrial thermal hazards including petrochemical fires, hazardous material incidents, BLEVE events, and dust explosions. The paper concludes with a critical synthesis of current research gaps and recommendations for future work, emphasizing integrated, data-driven, and human-centered approaches to emergency rescue.

Abstract: Background: Technical operations at oil and gas project sites — drilling and well intervention, construction and commissioning, turnaround maintenance, hot work, confined-space entry, hydrotesting, and heavy lifting — concentrate high-energy hazards, transient multi-employer workforces, and abnormal plant states within compressed schedules. Safety intervention here is pursued through three largely separate traditions: barrier management, behaviour-based observation, and management-system certification. Objective: To synthesise the evidence and standards base for each, evaluate what is known about intervention effectiveness in technical operations specifically, and propose an integrated framework assigning each intervention class to the failure mode it can plausibly control. Methods: Structured review following PRISMA-ScR, searching Scopus, Web of Science, ScienceDirect, and the ASCE and OnePetro libraries for records published 1990–2026, with a purposive corpus of public inquiry reports and normative standards (IOGP, API, ISO, IEC, NORSOK, OSHA, HSE). Forty-nine sources were retained (Figure 1). Results: Most intervention evaluation evidence addresses personal safety (recordable injury rates) while the catastrophic risk is process safety — an asymmetry unresolved since Piper Alpha. Contractor fragmentation, not individual error, is the dominant proximate condition in serious project-site events, yet the contracting interface has no controlled evaluation in the retained corpus. Behaviour-based safety has reasonable evidence for repetitive, observable, low-energy tasks and weak theoretical warrant for the judgement-dense work of turnarounds and well operations. Permit-towork and energy-isolation systems are the most consistently implicated failed barriers, and their failures are failures of administration under schedule pressure rather than of design. The highest-reliability interventions available displace administrative reliance with physical constraint — double-block-and-bleed with monitored bleed, spool removal, individual employee locks, verified inerting, SILrated instrumented protection, redundant tested well barrier envelopes, hydrorather than pneumatic testing, and design-stage elimination of entry and hot work — and all are decided in engineering design and procurement rather than at the site. Three groups of technical operations show traceable intervention success, and each pairs an engineered barrier with a competence requirement and an independently testable verification criterion. Conclusions: Intervention effort is misallocated relative to the risk profile: it over-invests in observing individual behaviour and under-invests in barrier verification, interface governance, and competence assurance across the contracting chain. We propose a four-layer framework — barrier integrity, competence, disclosure, governance — with a matched leading-indicator set, five separately manipulable contractor-interface components, and a steppedwedge design by which they could be tested across multi-site portfolios.

Abstract: Kill-web absorption resilience depends on whether directed mission paths remain available after disruption. This study addresses how to allocate a limited fortification budget to mission-function nodes so that an adaptive interdictor must incur the greatest possible cost to disconnect all valid mission paths. We formulate Mission-Path Cut Fortification ( MPCF), which maximizes the post-fortification minimum node-cut cost. Node-capacity splitting converts the interdiction response for a fixed fortification set into an equivalent maximum-flow minimum-cut problem. A compact mixed-integer linear program provides the primary exact solution, and an independent cut-generation formulation cross-checks the optimum in cut space. Across 135 graph-budget pairs from 45 prespecified synthetic networks, the mean optimality gaps of MPCF-Greedy and PathFrequency are 2.76% and 9.50%, respectively. Across 810 combinations under six role-composition profiles, the two exact formulations obtain identical objective values; MPCF-Greedy has a mean gap of 1.98%, whereas the best non-MPCF method under each profile has a mean gap of 8.11%–10.12%. Exact optimization also agrees with exhaustive enumeration on 72 small fortification instances and 72 capacity-interdiction instances, and cut-space verification agrees on 30 public-network instances. These results show that mission-path fortification benefits from joint optimization of the protected node set and the adaptive minimum-cut response.

Abstract: Although conventional hammering inspection remains a standard practice across various industries, its effectiveness is often limited by a heavy reliance on the inspector’s subjective skill, leading to qualitative and inconsistent results. To address these limitations, this study introduces a digital hammering inspection system integrated with Acoustic Emission (AE) sensors, designed for enhanced quality control and long-term aging management. The system has been extensively implemented in the diagnostic evaluation of social infrastructure, encompassing both metallic and concrete structures. By leveraging the principles of natural vibration, the system provides an objective, quantitative, and recordable screening process that significantly reduces human error and ensures high-speed data acquisition. Furthermore, its lightweight design and portability facilitate seamless field application, enabling comprehensive, on-site diagnostics that are both reliable and scalable for diverse engineering environments.

Abstract: Discrepancies between announced regional seismic intensity and shaking experienced indoors may affect household mitigation and risk communication. We report on a three-year field experiment (2021–2024) in the Greater Tokyo Metropolitan Area combining citizen-hosted low-cost sensors with subjective reports. Up to 92 volunteers participated; sensor hosts kept plug-in micro-electromechanical systems (MEMS) seismometers at home, and participants completed questionnaires after 96 target earthquakes. The system yielded 5,724 responses to 8,054 delivery attempts (response rates of 80.6% in 2022 and 67.6% in 2023), with 96% returned within 72 hours. Findings were robust to unrounded measures. First, municipal Japan Meteorological Agency (JMA) seismic intensity < perceived seismic intensity < indoor seismic intensity; indoor seismic intensity exceeded municipal JMA seismic intensity by roughly one-third of a class (median +0.34 against unrounded instrumental values), with a 1.4-class variation across structures. Second, local intensity was the primary predictor of feeling an event (odds ratio 2.57 per class), whereas the floor-height association was weaker. Third, repeated feedback did not produce detectable calibration; over-reporting biases (+0.4 to +0.7 classes) persisted over up to 96 events. These findings support indoor seismic intensity as a complement to regional metrics for indoor seismic-hazard evaluation, last-mile early warning, and risk communication.

Abstract: Predictive maintenance has emerged as a key strategic approach across numerous industrial sectors, particularly in the current context marked by the systematic integration of artificial intel-ligence (AI) technologies into asset management processes. Within the railway domain, maintenance plays a critical role in ensuring both operational reliability and safety, with maintenance-related activities accounting for up to 40% of the overall budget allocated across the V-cycle of rolling stock development. Given the essential contribution of rail transport to sus-tainable mobility and to the movement of goods and passengers, the demand for intelligent and efficient maintenance strategies continues to grow. However, a comprehensive predictive maintenance framework tailored to the railway sector remains lacking. This study presents an in-depth review of the principal methods employed in recent years, combining a PRISMA-guided systematic search of three major scientific databases Scopus, Web of Science, and ScienceDirect. A total of 61 full-text articles were retained and analyzed in detail. The paper synthesizes current research trends, frameworks, and algorithms, categorized into data-driven, model-based, and hybrid approaches, and examines their distribution across railway subsystems, revealing a con-centration of research effort on rolling-stock components such as wheelsets and switch machines relative to less-instrumented, comparably safety-critical assets such as traction, signaling, and door systems. Building on these insights, a conceptual hybrid predictive maintenance framework is proposed as a synthesis of the reviewed approaches. Finally, the paper discusses the key challenges identified in the literature and outlines open research questions with potential direc-tions for future work.

Abstract: Fire evacuation in complex multi-storey buildings is a dynamic task in which route safety changes depending on fire development, smoke propagation, and the spatial distribution of evacuees. Contemporary research increasingly applies artificial-intelligence methods for adaptive evacuation planning, but most of these approaches achieve adaptivity at the expense of interpretability and traceability. This is a limitation that is especially critical for systems with direct relevance to human safety. The present paper introduces a fully deterministic approach to intelligent fire evacuation that extends the hierarchical building graph model proposed by Ivanov [1] with continuous sensor-based risk assessment (temperature, smoke, CO₂, crowd density), unified through weighted fusion with hysteresis. Route evaluation uses a calibrated composite edge-cost model, complemented by a threshold-based table for adaptive node priority and multi-agent coordination through virtual load, while evacuee movement is modeled by a cellular automaton. All components of the proposed system are configurable and calibrated rather than trainable, which ensures full traceability, auditability, and compliance with fire-safety regulatory requirements. Evaluation across thirteen scenarios in four real buildings shows that the system's adaptivity stems mainly from multi-agent coordination and dynamic route recomputation, rather than from offline calibration of the graph weights. Coordination reduces the standard deviation of the maximum evacuation time by a factor of 2.7 to 4.9 relative to an uncoordinated baseline algorithm, at a mean evacuation time that is practically equivalent (within 1%) or, in the worst case, about 9% higher. The system achieves complete load balancing across exits (Cliff's delta up to 1.00) and guaranteed avoidance of fire and smoke nodes in all test scenarios, with the only exception involving boundary cases in which fire spreads faster than the sensor-classification interval, a physical detection-latency limit rather than a routing failure. These results indicate that deterministic, calibrated coordination can achieve adaptivity comparable to learning-based methods while preserving the traceability and auditability required for regulatory-compliant fire-safety deployment.

Abstract:

Construction workers in Riyadh often wear personal protective equipment (PPE) in hot conditions. This exploratory study examined whether heat-related discomfort, supervisor or leadership influence, and language accessibility were associated with self-reported routine PPE use. Anonymous survey responses from 85 workers at seven construction sites were analysed alongside unlinked site-observation summaries from February to April 2026. Routine use was coded from whether PPE was part of daily work or worn when reminded. Four associations were assessed using Pearson chi-square tests, Cramer’s V, odds ratios, 95% confidence intervals and Holm-adjusted p-values. Heat-related discomfort was associated with routine use, χ²(1, N = 84) = 5.05, adjusted p = 0.025, V = 0.245; the positive direction may indicate that routine users experienced more discomfort rather than that heat improved compliance. Supervisor or leadership influence showed the strongest association, χ²(1, N = 82) = 39.03, adjusted p < 0.001, V = 0.690. Training-language and safety-instruction language accessibility were also associated with routine use (adjusted p = 0.002 and 0.004). Observation summaries recorded lower correct PPE use and more heat-attributed adjustment, but counts, measured heat exposure and worker linkage were unavailable. The findings support prospective studies using validated compliance measures, measured exposure and linked observations.

Abstract: Methane emissions from offshore installations are routinely monitored using gas detection systems that form an important component of operational safety management. Detector outputs are commonly used to support safety-critical decisions during helicopter approach, vessel transfer, inspection activities, and routine offshore operations. However, the extent to which atmospheric conditions influence the effectiveness of these safety barriers remains poorly quantified. This study presents a probabilistic modelling framework to assess methane detectability as a function of emission rate, atmospheric regime, receptor height, and sensor sensitivity, with particular focus on the implications of detection failure for offshore safety. A Gaussian plume dispersion model combined with Monte Carlo simulation was used to account for variability in wind speed and wind direction, allowing estimation of detection probability under realistic offshore conditions. Detection was evaluated at two representative receptor locations: helideck height (25 m), representative of helicopter operations, and near-surface level (~2 m), representative of vessel-based detection. Emissions were varied over a wide range (0.01–1000 g s⁻¹; approximately 0.04–3600 kg h⁻¹), while detection thresholds of 0.1, 1, and 10 ppm were used to represent different classes of methane monitoring systems. Results show that methane detectability is frequently controlled by plume geometry and atmospheric structure rather than emission magnitude alone. In the near field (< ~200 m), limited vertical dispersion results in plume–sensor misalignment, producing false-negative detection outcomes in which substantial methane releases remain undetected despite the presence of monitoring systems. Detection improves at intermediate distances (~400–800 m), where plume spreading increases the likelihood of plume interception, before decreasing again at larger distances due to dilution. Increasing detection thresholds significantly reduces detectability, with 10 ppm sensors failing to detect large emissions across wide operating conditions, particularly under stratified atmospheric regimes. These findings demonstrate that gas detection systems should be regarded as conditional safety barriers whose effectiveness depends on atmospheric transport processes as well as sensor performance. The absence of methane detection should therefore not be interpreted as evidence that a hazardous emission is absent. This has direct implications for offshore risk management, situational awareness, and safe-ty-critical operational decision-making.

Abstract: Fall from height (FFH) remain the predominant cause of fatalities and injuries in the construction in construction and industrial safety worldwide. Traditional accident cause analysis approaches often focus on single-factor identification or static statistical analysis, ignoring the hierarchical coupling and correlational relationship of multiple factors. To address this research gap, this study proposes a novel method based on overlay network chain in quotient space to identify overlapping causal factor clusters in FFH accidents. First, a causal factor interaction network is constructed based on the similarity of causes. On this basis, an overlay network chain in quotient space is established to identify overlapping causal factor clusters. The proposed method is validated using real FFH accident data. Results demonstrate that this approach effectively identifies hidden high-risk overlapping clusters that are undetectable by traditional methods. Notably, Cluster C5—comprising PPE non-compliance, inadequate supervision, unverified qualifications, and insufficient training—exhibits the highest occurrence rate (0.072) and structural vulnerability (CVI = 9.5), identifying it as the most critical risk pattern. The identified clusters are highly consistent with practical engineering experience. This study provides a systematic, data-driven tool for FFH accident causation analysis, supporting targeted safety prevention and control strategies.

Abstract: Artificial intelligence (AI) is increasingly used for tool-wear prediction in computer numerical control (CNC) machining, yet a high average accuracy can conceal systematic failure in the safety-critical high-wear region, where under-prediction makes a worn tool appear healthier than it is. This paper presents a trustworthy multimodal framework that audits, mitigates, and ultimately characterises this bias on the public MATWI dataset. A fusion model combining 512 image features with 84 force, acceleration, and acoustic features attains a strong test mean absolute error (MAE) of 30.9 µm, outperforming image-only (51.4 µm) and sensor-only (68.3 µm) baselines. A signed, stratified bias audit nevertheless reveals a pronounced directional error: the model predicts the healthy region to within 21.3 µm but under-predicts the worn region by approximately 85 µm (worn-region MAE 85.5 µm), with the error deepening monotonically as wear increases. A three-stage mitigation pipeline (data-level resampling, cost-sensitive and relevance-based training, and post-hoc recalibration) reduces the worn-region error by up to 18% but cannot eliminate it, and only at a measurable cost to the well-sampled region. The residual bias is shown to be both model-independent, persisting across six regressor families with a 69–127 µm residual gap, and optimisation-independent, surviving rigorous hyperparameter search and bias-aware loss functions behind an ≈ 85 µm error floor. These results establish the bias as data-intrinsic: it arises from the sparse sampling of the high-wear regime rather than from model or training choices, and the effective remedy lies in balanced, unbiased data collection rather than further algorithmic correction. The contribution is a reusable protocol for distinguishing model-reducible from data-intrinsic bias in safety-critical regression.

Abstract: Structural durability and reliability are fundamental imperatives for the safe operation of engineering infrastructure, profoundly influencing both lifecycle asset economics and socio-environmental safety. A critical component of structural integrity engineering involves the precise spatial localization of defects, the continuous monitoring of their propagation kinetics, and the high-fidelity quantification of their impact on global structural health. To achieve this, advanced diagnostic methodologies are required to detect the earliest indicators of material degradation and track its evolution throughout the operational lifespan of the asset. Crucially, these non-destructive techniques must transcend reliance on subjective, localized visual inspections or unverified numerical models. The Acoustic Emission (AE) method represents a highly effective Non-Destructive Testing (NDT) paradigm that satisfies these requirements by performing real-time analysis of active degradation mechanisms coupled with specialized reference signal databases. This paper investigates the application of an AE-based framework integrated with a fuzzy pattern recognition database—the Identifying Gas Network Anomalies (IGNA) methodology—to localize, monitor, and classify destructive processes induced by mechanical loads. The proposed approach was deployed to evaluate the structural condition of critical gas infrastructure components fabricated from steel, cast iron, and polymeric materials. Comprehensive laboratory campaigns and long-term in-situ experimental trials validated the efficacy of the AE framework for automated and semi-automated diagnostics of gas networks under real-time load conditions. The outcomes of this research led to the successful deployment of a pattern recognition-based system that facilitates proactive failure prevention and ensures the operational security of gas network infrastructure.