Hasil untuk "Industrial safety. Industrial accident prevention"

Ayan Das, Dhaval Patel

Large language model (LLM) agents are increasingly deployed for complex tool-orchestration tasks, yet existing benchmarks fail to capture the rigorous demands of industrial domains where incorrect decisions carry significant safety and financial consequences. To address this critical gap, we introduce PHMForge, the first comprehensive benchmark specifically designed to evaluate LLM agents on Prognostics and Health Management (PHM) tasks through realistic interactions with domain-specific MCP servers. Our benchmark encompasses 75 expert-curated scenarios spanning 7 industrial asset classes (turbofan engines, bearings, electric motors, gearboxes, aero-engines) across 5 core task categories: Remaining Useful Life (RUL) Prediction, Fault Classification, Engine Health Analysis, Cost-Benefit Analysis, and Safety/Policy Evaluation. To enable rigorous evaluation, we construct 65 specialized tools across two MCP servers and implement execution-based evaluators with task-commensurate metrics: MAE/RMSE for regression, F1-score for classification, and categorical matching for health assessments. Through extensive evaluation of leading frameworks (ReAct, Cursor Agent, Claude Code) paired with frontier LLMs (Claude Sonnet 4.0, GPT-4o, Granite-3.0-8B), we find that even top-performing configurations achieve only 68\% task completion, with systematic failures in tool orchestration (23\% incorrect sequencing), multi-asset reasoning (14.9 percentage point degradation), and cross-equipment generalization (42.7\% on held-out datasets). We open-source our complete benchmark, including scenario specifications, ground truth templates, tool implementations, and evaluation scripts, to catalyze research in agentic industrial AI.

Prof. Vishal Tiwari, V. Singh, Dr. Sandeep Kumar Yadav et al.

Abstract Industrial risk zones such as manufacturing plants, construction sites, chemical processing units, and mining areas are highly prone to accidents due to unsafe practices, human error, and environmental hazards. Traditional safety monitoring methods rely heavily on manual supervision and CCTV observation, which are often inefficient, reactive, and prone to oversight. This research presents the development of a Machine Learning powered Safety Observation System that enables automated, real-time detection of unsafe acts, hazardous conditions, and non-compliance with safety protocols in industrial environments. The proposed system leverages computer vision, deep learning algorithms, and real-time video analytics to monitor worker behavior, personal protective equipment (PPE) compliance, proximity to dangerous zones, and abnormal activities. The system architecture, methodology, algorithm design, and implementation framework are discussed in detail. Analytical evaluation shows significant improvement in hazard detection accuracy, response time, and reduction in accident probability. The study demonstrates how AI-based surveillance can transform industrial safety management from reactive monitoring to proactive risk prevention. Keywords: Machine Learning, Industrial Safety, Computer Vision, PPE Detection, Hazard Monitoring, AI Surveillance.

Yuan Cao, Yue Liu, Yongkui Sun et al.

The rapid detection of internal rail defects is critical to maintaining railway safety, but this task faces a significant challenge due to the limited computational resources of onboard detection systems. This paper presents YOLOv8n-LiteCBAM, an advanced network designed to enhance the efficiency of rail defect detection. The network designs a lightweight DepthStackNet backbone to replace the existing CSPDarkNet backbone. Further optimization is achieved through model pruning techniques and the incorporation of a novel Bidirectional Convolutional Block Attention Module (BiCBAM). Additionally, inference acceleration is realized via ONNX Runtime. Experimental results on the rail defect dataset demonstrate that our model achieves 92.9% mAP with inference speeds of 136.79 FPS on the GPU and 38.36 FPS on the CPU. The model's inference speed outperforms that of other lightweight models and ensures that it meets the real-time detection requirements of Rail Flaw Detection (RFD) vehicles traveling at 80 km/h. Consequently, the YOLOv8n-LiteCBAM network is with some potential for industrial application in the expedited detection of internal rail defects.

Patrycja Kabiesz, Grażyna Płaza, Tayyaba Jamil

Modern technologies are transforming occupational health and safety training by enhancing education, innovation, fire prevention, and promoting sustainability conditions in various sectors of industries. Digital tools such as virtual reality, artificial intelligence, and interactive simulations improve learning efficiency, engagement, and risk awareness. By integrating the technologies, companies can better prepare employees for hazardous situations, reduce workplace accidents, and ensure compliance with safety regulations. Fire courses on fire prevention and control are an essential element in health and safety trainings, and a crucial aspect of safety management. In any business, employees should be prepared for emergency situations, including fires by using modern tools like artificial intelligence. This article aimed to assess the implementation of modern technologies in Polish occupational health and safety training across various industrial sectors. Additionally, this research considered variations in training program development based on company size and financial capacity, highlighting the importance of integrating training, education, and innovative technologies into the company’s overall development strategy. The relationships between safety training programs, education, and innovation in 597 industrial companies were evaluated. The research findings suggest that integrating innovative technologies into training can improve working conditions in a more sustainable way and enhance the market competitiveness of enterprises.

A. Sheeba, C. A. Subasini, Charan Shiv Sps et al.

This study offers an Arduino-based LPG gas leakage detection system with an automatic shut-off regulator to offset gas leak-related accidents. The proposed system uses an MQ-2 gas sensor to sense leaks and automatically activates a solenoid valve to close gas flow. According to experimental findings, the system has a rapid response time of 3 seconds and a detection rate of 95% in controlled environments, largely mitigating fire hazard risk. Moreover, an IoT-based warning system offers real-time warnings, enhancing user consciousness and responsiveness. The research identifies a cost-effective and reliable solution to maximize gas safety in domestic and industrial application. This system will activate an effective alarm immediately it detects leakage in the gas and provides critical early warnings to occupants. In addition, it automatically shuts off the main gas supply, thus preventing possible production of more hazards. A system like this is well needed because detecting LPG leaks that are flammable is often very challenging, which can easily lead to the damage of property or physical harm to individuals. Besides, leaks, especially those that go unreported, and improper covering of hoses and regulators point out the importance of automating safety solutions.

Eslam Ali Khodabandehlou, Amir Hooman Hemmasi, Akramolmoluk Lahijanian et al.

Background and objective The present study aims to analyze the status of physical and infrastructural resilience to fire accidents in District 20 of Tehran, Iran based on the geographic information system. Method First, the criteria to evaluate the physical and infrastructural resilience were collected. Then, researcher-made questionnaires were completed by 15 experts from fire departments in Tehran, to rate the importance of these criteria. The weight of these factors was calculated using the Expert Choice 11 software. By layering the criteria using the weights obtained from the analytical hierarchical process (AHP) in ArcGIS software, version 10.6 and overlaying the layers, the final resilience map of District 20 was prepared. Results There were three main criteria, seven sub-criteria, and 21 indicators to evaluate the resilience. Based on the software output, the main criteria were prioritized as infrastructure resilience with a weight of 0.731, physical resilience with a weight of 0.188, and environmental resilience with a weight of 0.081. Based on the zoning map of resilience in the context of health, safety, and environment, it was found that 3.3% of the area (729,718 m2) had low resilience, 35.38% (7,802,578 m2) moderate resilience, 29.54% (6,513,646 m2) high resilience, 30.1% (6,639,824 m2) very high resilience, and 1.65% (364,196 m2) extremely high resilience. Conclusion According to the results, the District 20 of Tehran has appropriate physical and infrastructural resilience to fire and accidents.

Damir Kralj

Aman Verma, Keshav Samdani, Mohd. Samiuddin Shafi

This paper presents the design, implementation, and evolution of a comprehensive multimodal room-monitoring system that integrates synchronized video and audio processing for real-time activity recognition and anomaly detection. We describe two iterations of the system: an initial lightweight implementation using YOLOv8, ByteTrack, and the Audio Spectrogram Transformer (AST), and an advanced version that incorporates multi-model audio ensembles, hybrid object detection, bidirectional cross-modal attention, and multi-method anomaly detection. The evolution demonstrates significant improvements in accuracy, robustness, and industrial applicability. The advanced system combines three audio models (AST, Wav2Vec2, and HuBERT) for comprehensive audio understanding, dual object detectors (YOLO and DETR) for improved accuracy, and sophisticated fusion mechanisms for enhanced cross-modal learning. Experimental evaluation shows the system's effectiveness in general monitoring scenarios as well as specialized industrial safety applications, achieving real-time performance on standard hardware while maintaining high accuracy.

Leila Ismail, Abdelmoneim Abdelmoti, Arkaprabha Basu et al.

With the increasing complexity of industrial systems, there is a pressing need for predictive maintenance to avoid costly downtime and disastrous outcomes that could be life-threatening in certain domains. With the growing popularity of the Internet of Things, Artificial Intelligence, machine learning, and real-time big data analytics, there is a unique opportunity for efficient predictive maintenance to forecast equipment failures for real-time intervention and optimize maintenance actions, as traditional reactive and preventive maintenance practices are often inadequate to meet the requirements for the industry to provide quality-of-services of operations. Central to this evolution is digital twin technology, an adaptive virtual replica that continuously monitors and integrates sensor data to simulate and improve asset performance. Despite remarkable progress in digital twin implementations, such as considering DT in predictive maintenance for industrial engineering. This paper aims to address this void. We perform a retrospective analysis of the temporal evolution of the digital twin in predictive maintenance for industrial engineering to capture the applications, middleware, and technological requirements that led to the development of the digital twin from its inception to the AI-enabled digital twin and its self-learning models. We provide a layered architecture of the digital twin technology, as well as a taxonomy of the technology-enabled industrial engineering applications systems, middleware, and the used Artificial Intelligence algorithms. We provide insights into these systems for the realization of a trustworthy and efficient smart digital-twin industrial engineering ecosystem. We discuss future research directions in digital twin for predictive maintenance in industrial engineering.

Wenbo Pan, Zhichao Liu, Qiguang Chen et al.

Large Language Models' safety-aligned behaviors, such as refusing harmful queries, can be represented by linear directions in activation space. Previous research modeled safety behavior with a single direction, limiting mechanistic understanding to an isolated safety feature. In this work, we discover that safety-aligned behavior is jointly controlled by multi-dimensional directions. Namely, we study the vector space of representation shifts during safety fine-tuning on Llama 3 8B for refusing jailbreaks. By studying orthogonal directions in the space, we first find that a dominant direction governs the model's refusal behavior, while multiple smaller directions represent distinct and interpretable features like hypothetical narrative and role-playing. We then measure how different directions promote or suppress the dominant direction, showing the important role of secondary directions in shaping the model's refusal representation. Finally, we demonstrate that removing certain trigger tokens in harmful queries can mitigate these directions to bypass the learned safety capability, providing new insights on understanding safety alignment vulnerability from a multi-dimensional perspective. Code and artifacts are available at https://github.com/BMPixel/safety-residual-space.

Emily L. Tucker, Mohammadhossein Mohammadisiahroudi

Quantum computing is rapidly emerging as a new computing paradigm with the potential to improve decision-making, optimization, and simulation across industries. For industrial engineering (IE) and operations research (OR), this shift introduces both unprecedented opportunities and substantial challenges. The learning curve is high, and to help researchers navigate the emerging field of quantum operations research, we provide a road map of the current field of quantum operations research. We introduce the foundational principles of quantum computing, outline the current hardware and software landscape, and survey major algorithmic advances relevant to IE/OR, including quantum approaches to linear algebra, optimization, machine learning, and stochastic simulation. We then highlight applied research directions, including the importance of problem domains for driving long-term value of quantum computers and how existing classical OR models can be reformulated for quantum hardware. Recognizing the steep learning curve, we propose pathways for IE/OR researchers to develop technical fluency and engage in this interdisciplinary domain. By bridging theory with application, and emphasizing the interplay between hardware and research development, we argue that industrial engineers are uniquely positioned to shape the trajectory of quantum computing for practical problem-solving. Ultimately, we aim to lower the barrier to entry into quantum computing, motivate new collaborations, and chart future directions where quantum technologies may deliver tangible impact for industry and academia.

Alice Di Bella, Toni Seibold, Tom Brown et al.

This study analyzes how Europe can decarbonize its industrial sector while remaining competitive. Using the open-source model PyPSA-Eur, it examines key energy- and emission-intensive industries, including steel, cement, methanol, ammonia, and high-value chemicals. Two development paths are explored: a continued decline in industrial activity and a reindustrialization driven by competitiveness policies. The analysis assesses cost gaps between European green products and lower-cost imports, and evaluates strategies such as intra-European relocation, selective imports of green intermediates, and targeted subsidies. Results show that deep industrial decarbonization is technically feasible, led by electrification, but competitiveness depends strongly on policy choices. Imports of green intermediates can lower costs while preserving jobs and production, whereas broad subsidies are economically unsustainable. Effective policy should focus support on sectors like ammonia and steel finishing while maintaining current production levels.

Sergey Chigvincev, I.Yu. Chidunchi, Diyar Särsen et al.

One of the principles of the labor legislation of the Republic of Kazakhstan is the priority of the life and health of the employee, as well as ensuring the right to working conditions that meet the requirements of safety and hygiene [1]. One of the most important mechanisms designed to ensure compliance with this principle is the internal control mechanism for ensuring occupational safety and health. Internal control over occupational safety and health is carried out by the employer in order to comply with the established requirements for occupational safety and health at the workplace and take immediate measures to eliminate the detected violations. An integral part of the internal control mechanism is public control by employee representatives represented by technical inspectors on labor protection and trade unions. The main purpose of the study is to consider the role and influence of public control on ensuring industrial safety and prevention of industrial injuries, therefore, the authors analyzed the statistics of industrial injuries in the LLP "Ekibastuz GRES-1 named after Bulat Nurzhanov". The authors of the work noted the importance of the participation of employee representatives in monitoring the provision of safe working conditions aimed at reducing the level of occupational injuries and accidents, improving the culture of occupational safety

Kyung-Woo Kim, Kwangsu Moon, Ji Dong Lee

Background This study explored the applicability of the Trust Leading Indicator (TLI) and Proactive Leading Indicator (PLI), developed as part of the Vision Zero, a global campaign for the dissemination of prevention culture, in Korean industries. The relationship between these indicators and safety culture-related variables were compared, such as safety climate, safety behavior, risk perception, and accident experience. Methods The study sample comprised 630 workers from 12 subcontractors affiliated with the Republic of Korea's large manufacturing plant. Correlations among the main variables were examined, including group differences in TLI and PLI based on subjective accident experience. Results The TLI and PLI had significant positive correlations with the sub-factors of safety climate and safety behavior and negative correlations with risk perception, indicating their potential utility as extensions of existing safety culture indicators. A significant difference in TLI and PLI was observed across accident experience levels. Conclusion Despite limitations, such as the predominance of male workers in the study owing to the nature of the industry and use of subjective accident experience rather than official industrial accident data, this study is significant as it explores the applicability of the two leading indicators of prevention culture in Korean industries, confirming the potential utility of these indicators across various cultural contexts and contributing to global efforts to disseminate a prevention culture.

Kun Woo Lee

Traditional construction safety education has often been limited in effectiveness due to its theory-oriented delivery and low learner engagement, resulting in insufficient impact on accident prevention. Recent studies have highlighted the potential of Virtual Reality (VR) technology to transform safety training into an immersive, learner-centered experience by enabling repetitive practice and high-risk simulations. Since 2021, experiential training using VR has been increasingly adopted in statutory safety education for construction professionals in Korea. This study compares VR-based safety education with traditional lecture-based approaches, analyzing their effectiveness and limitations. From March to May 2025, a VR safety training program was conducted with 45 construction professionals at the Kyungbok University Construction Education Center, focusing on realism, immersion, learning outcomes, and applicability to the field. Survey results showed that over 89% of participants rated VR training as effective (4 or higher on a 5-point scale), with particularly high satisfaction regarding immersion. However, technical issues such as dizziness, differences from real sites, and discomfort with equipment were also noted. The study suggests that for VR safety education to further develop as an effective tool for industrial accident prevention, collaborative content development reflecting learner needs, improvement of training equipment usability, and systematic post-training effectiveness management are essential.

V. V. Egelsky, N. N. Nikolaev, E. Egelskaya et al.

Introduction. The scientific literature discusses the potential of artificial intelligence (AI) for ensuring industrial safety. Risk control methods are considered and recommendations for incident prevention are given. The relationship between the competencies of lifting crane operators and the probability of accidents has been studied. Examples of using neural networks for determining the reliability of removable lifting devices are presented. Remote monitoring of operational safety is described. However, the use of AI to manage risks in a car repair station has not been sufficiently studied. This research aims to address this gap. The aim of this work is to demonstrate the potential of neural networks in creating a safety monitoring system for an automobile repair facility.Materials and Methods. The design materials of the service station at the equipment repair and maintenance center served as basic information. This enterprise was created by specialists of the Department of Operation of Transport Systems and Logistics at the Don State Technical University (DSTU). Risks were classified according to GOST ISO 12 100 and GOST R 58 771. Neural networks were trained using open-source libraries for the Python programming language. The digital monitoring system model with visualization was implemented using the AnyLogic simulation system.Results. The authors of this work trained 20 neural networks and selected five with the lowest error function values (from 74% to 78%). Out of five networks that worked most correctly, one was chosen that predicted the output parameter more accurately — 74%. The neural network with the best performance was a multilayer perceptron with 30 neurons in the input layer, 15 in the hidden layer, and 3 in the output layer. It was used to create a digital twin that warned in real time about potentially dangerous events: the movement of a car, a crane, and the opening of an inspection pit. Additionally, it identified workers without personal protective equipment or access to the work area.Discussion and Conclusion. The use of a digital safety monitoring system model will make it possible to identify high-risk work areas in advance, and reduce accidents and industrial injuries. The introduction of this model in auto repair facilities involves the installation of sensors and warning systems. In the future, we plan to explore the possibility of integrating algorithms with the risk monitoring system to help personnel repair specific types of machines.

Chanyakarn Kokaphan, Phongthon Saengchut, Nipaporn Khamhlom et al.

The occupational health and safety management system (OHSMS) of industry is a key issue in reducing risks and accidents in the workplace. A study to evaluate the OHSMS factors in the perception of workers in Nakhon Ratchasima province using a questionnaire of 937 people from 13 types of industries. The results showed confirmed of questionnaire met the recommended criteria and confirmatory factor analysis (CFA) between observable with variables of safety management found safety management’s Goodness of fit indicators. The perception of safety management in industry found that most safety officers were aware of occupational health and safety policy focuses on loss prevention and control and compliance with the law (99.51%) that had the highest influence on safety management (β=0.432) while most employees were aware of communication back to management to jointly promote safe working practices and appointment of safety committee and safety officer (91.13%) that had the highest influence on safety management (β=0.327). The study on OHSMS in industrial sectors found that safety officers focus on policy compliance and risk prevention, while employees emphasize communication and safety committees. Discrepancies in safety perceptions highlight the need for better alignment between officers and employees. The study suggests combining compliance-focused programs with proactive employee engagement to strengthen safety culture. Future research should explore the impact of these approaches on accident rates and employee satisfaction across sectors.

S. Kudryavtsev, P. Yemelin, Aigerim Rakhimberlina et al.

Despite significant progress in labor safety, improvements in process safety performance and the prevention of major accidents at hazardous industrial sites remain less apparent. Enhancing the efficiency of occupational and industrial safety management systems plays a crucial role in addressing this challenge, particularly by advancing risk assessment methods. However, a reliable evaluation of risks associated with hazardous industrial operations requires extensive statistical data. This article aims to critically assess current approaches to building databases in the fields of fire and industrial safety and to identify key criteria for their application in fire risk assessment methodologies using expert evaluation methods. It also addresses challenges related to the content and structure of such databases. The authors propose clusters of criteria parameters that can serve as the foundation for a methodological approach to fire safety risk assessment at industrial enterprises, leveraging expert evaluation techniques.

Spyridon Damilos, Stratos Saliakas, Dimitris Karasavvas et al.

Airborne pollutants pose a significant threat in the occupational workplace resulting in adverse health effects. Within the Industry 4.0 environment, new systems and technologies have been investigated for risk management and as health and safety smart tools. The use of predictive algorithms via artificial intelligence (AI) and machine learning (ML) tools, real-time data exchange via the Internet of Things (IoT), cloud computing, and digital twin (DT) simulation provide innovative solutions for accident prevention and risk mitigation. Additionally, the use of smart sensors, wearable devices and virtual (VR) and augmented reality (AR) platforms can support the training of employees in safety practices and signal the alarming concentrations of airborne hazards, providing support in designing safety strategies and hazard control options. Current reviews outline the drawbacks and challenges of these technologies, including the elevated stress levels of employees, cyber-security, data handling, and privacy concerns, while highlighting limitations. Future research should focus on the ethics, policies, and regulatory aspects of these technologies. This perspective puts together the advances and challenges of Industry 4.0 innovations in terms of occupational safety and exposure assessment, aiding in understanding the full potential of these technologies and supporting their application in industrial manufacturing environments.

Hasanika Samarasinghe, Shadi Heenatigala

The study delves into 425 industrial incidents documented on Kaggle [1], all of which occurred in 12 separate plants in the South American region. By meticulously examining this extensive dataset, we aim to uncover valuable insights into the occurrence of accidents, identify recurring trends, and illuminate underlying causes. The implications of this analysis extend beyond mere statistical observation, offering organizations an opportunity to enhance safety and health management practices. Our findings underscore the importance of addressing specific areas for improvement, empowering organizations to fortify safety measures, mitigate risks, and cultivate a secure working environment. We advocate for strategically applying statistical analysis and data visualization techniques to leverage this wealth of information effectively. This approach facilitates the extraction of meaningful insights and empowers decision-makers to implement targeted improvements, fostering a preventive mindset, and promoting a safety culture within organizations. This research is a crucial resource for organizations committed to transforming data into actionable strategies for accident prevention and creating a safer workplace.