Uncovering Patterns in HandRelated Injuries in Manufacturing

Team Innovate Y Team members: Debasmita Ray (dxr230049@utdallas.edu) Soundariyan Venkatachalam (sxv240043@utdallas.edu) Avantika Patrina Ananth (axa230177@utdallas.edu) Harish Kumar Sarathi (hxs230100@utdallas.edu)

Inspiration

The idea for our project was sparked by a recurring challenge in industrial safety: hand-related injuries. As one of the most common—and most preventable—injuries in manufacturing, we wanted to understand why they happen and how we could help prevent them using data. The detailed narratives in OSHA’s Severe Injury Reports presented a rich, untapped resource that we felt could be unlocked using AI and NLP. That became our mission.

What it does

The objective of this study is to analyze OSHA's Severe Injury Report dataset to identify prevailing patterns and root causes of hand-related injuries within the manufacturing industry. By uncovering the most common scenarios and mechanisms of these injuries, this research aims to provide actionable insights that support S.E.G.'s ongoing mission of developing and refining targeted safety interventions, thereby effectively reducing workplace hand injuries.

How we built it

1. Dashboards in PowerBI
2. We cleaned and processed over 23,000 records, filtered for manufacturing sector injuries, and trained an LDA model with 5 topics. Each topic was interpreted and linked to both injury type (e.g., crushing, laceration) and source (e.g., press machine, rotating equipment). We also developed a prediction function to classify new narratives instantly. ## Challenges we ran into

Accomplishments that we're proud of

NLP prediction model and our dashboards

What we learned

We deepened our knowledge of:

1. Natural Language Processing (NLP) with unstructured safety data
2. Latent Dirichlet Allocation (LDA) to uncover hidden injury themes
3. Categorical data engineering for injury types, sources, and severities
4. Reverse geocoding and time-based analysis to enrich missing location and date features
5. Dashboarding for data-driven storytelling Most importantly, we learned how domain expertise (safety, OSHA standards) and modern AI can come together to solve real-world problems.

What's next for Uncovering Patterns in HandRelated Injuries in Manufacturing

Dynamic LLM models

Built With

• natural-language-processing
• powerbi
• r