Applications of AI in Occupational Health and Safety: A Systematic Review

This article is a systematic review following the PRISMA guidelines and registered on PROSPERO (CRD42024568795), covering 43 studies. It analyzes the application effects of AI models such as CNN, LSTM, and YOLO in hazard detection, prediction, and prevention in the field of Occupational Health and Safety (OHS), while also discussing the current challenges and future development directions.

Occupational Health and Safety (OHS) is a core issue in industrial production. Traditional management relies on manual inspections and experience-based judgments, which have problems such as limited coverage and delayed response. AI technology shows potential in hazard detection, risk prediction, and safety prevention. This review retrieved studies from 2020 onwards through databases such as Google Scholar, PubMed, and Scopus, and included 43 relevant studies for analysis.

This study adopts strict methods to ensure reliability:

1. Search strategy: Covers databases such as Google Scholar, PubMed, and Scopus, focusing on studies since 2020;
2. Inclusion criteria: Focuses on AI applications in OHS, uses models like CNN/LSTM/YOLO, peer-reviewed English literature, and includes experimental/observational/comparative studies;
3. Quality assessment: Independent double-blind review with Cohen's kappa coefficient of 0.87, ensuring reliable evaluation results.

Main AI Technologies

• CNN: Visual hazard detection (e.g., unprotected PPE, equipment abnormalities) with accuracy over 80%;
• LSTM: Time-series analysis for predicting safety risks to enable proactive prevention;
• YOLO: Real-time object detection to support industrial monitoring scenarios.

Application Areas

Covers hazard detection, ergonomic analysis, predictive maintenance, safety training, and emergency response optimization.

Effects

Hazard detection accuracy exceeds 80%, accident prediction capability is improved, monitoring real-time performance and coverage are enhanced, and manual inspection efficiency is increased.

Current AI applications in OHS face the following challenges:

1. Data quality: Inconsistent labeling and sample imbalance affect model generalization;
2. Ethical privacy: Monitoring involves employee privacy, requiring a balance between safety and privacy;
3. Implementation integration: Enterprises lack technical capabilities and change management experience;
4. Insufficient standardization: Lack of unified norms makes results difficult to compare and reproduce;
5. Interpretability: The decision-making process of deep learning models is opaque, affecting trust and usage.

AI applications in OHS have industry-specific differences:

• Construction: High-altitude operation safety, heavy machinery monitoring, hazard identification at construction sites;
• Manufacturing: Predictive maintenance of equipment, production line monitoring, ergonomic optimization;
• Mining: Geological risk prediction, underground monitoring, emergency response optimization;
• Chemical industry: Chemical leakage detection, process safety monitoring, environmental risk assessment.

Future Development Directions

1. Multimodal fusion: Integrate visual, auditory, and sensor data;
2. Edge computing: Deploy edge devices to achieve low-latency real-time monitoring;
3. Explainable AI: Improve model decision transparency;
4. Cross-domain transfer: Explore knowledge transfer between industries;
5. Human-machine collaboration: AI as an intelligent assistant rather than a replacement tool.

Practical Recommendations

1. Pilot first: Start with small-scale scenarios;
2. Data construction: Invest in data collection and labeling;
3. Talent cultivation: Train compound talents or cross-departmental teams;
4. Ethical compliance: Focus on privacy protection and algorithm fairness;
5. Continuous evaluation: Regularly test model performance and optimize.

This review shows that AI models such as CNN, LSTM, and YOLO have achieved significant results in the OHS field, but still face challenges such as data, ethics, and integration. With technological progress and experience accumulation, AI is expected to become an important enabling tool for OHS, driving safety management from passive response to proactive prevention and achieving a safer and healthier working environment. Safety managers, enterprise decision-makers, and policy-makers need to understand its potential and limitations and formulate reasonable implementation strategies.