{"id":{"repo_id":"houston","oai_identifier":"oai:uh-ir.tdl.org:10657/21533"},"canonical_url":"https://search.dev.ndltd.org/etd/houston/oai:uh-ir.tdl.org:10657/21533","repository":{"repo_id":"houston","name":"University of Houston","base_url":"https://uh-ir.tdl.org/server/oai/request"},"display":{"title":"Improving Offshore Safety Through Data Analysis with Large Language Models and Accident Review with VR Simulation","abstract":"Offshore platforms represent high-risk environments where variable environmental conditions and equipment malfunctions require extensive safety training systems. Local LLMs (Large Language Models) offer advanced tools for data analysis to develop training protocols and enable users to maintain their data privacy. However, these models frequently pose significant barriers, requiring powerful computers and extensive, specialized datasets to be effective. Information from these proprietary sources is essential for developing corrective actions, learning from lessons, refining procedures, and training personnel to help prevent further incidents. This dissertation develops a scalable customizable system by leveraging proprietary datasets from offshore incidents, Local LLMs, and a custom simulation application. The proposed system uses the LLM to manipulate preset elements in Unity to create realistic simulation scenes and scenarios. The research was conducted in three phases. First, the capabilities of Unity as a real-world simulation package and for integration with other software were evaluated. These include integration with applications for complex computations and packages for online multi-user capabilities. Second, the feasibility of a local multi-LLM system was assessed, utilizing a local decoder-only and encoder-only model to optimize, enhance, and classify safety incident data from the proprietary dataset. Both the Unity simulator and LLM-classification frameworks were evaluated for their computational resource usage to ensure proper operation within the limited resources of a local system. Finally, these frameworks were combined into a cohesive system in which LLMs parse natural language input to create simulation scenes and generate actions within them. Once development was complete, the system was validated through a human factor experiment. Evaluation focused on the technical performance for the LLM-simulation system and on its ease of use. The human factor study was limited to a focus group with detailed feedback from each participant. Evaluation results demonstrate that the LLM-driven simulation system effectively communicated the safety training objectives, although the User Interface requires further refinement to enhance user experience, especially for those with limited simulation experience. The present dissertation contributes a viable framework for deploying privacy-centric, AI-enhanced training solutions in resource-constrained industrial environments.","abstract_html":"Offshore platforms represent high-risk environments where variable environmental conditions and equipment malfunctions require extensive safety training systems. Local LLMs (Large Language Models) offer advanced tools for data analysis to develop training protocols and enable users to maintain their data privacy. However, these models frequently pose significant barriers, requiring powerful computers and extensive, specialized datasets to be effective. Information from these proprietary sources is essential for developing corrective actions, learning from lessons, refining procedures, and training personnel to help prevent further incidents. This dissertation develops a scalable customizable system by leveraging proprietary datasets from offshore incidents, Local LLMs, and a custom simulation application. The proposed system uses the LLM to manipulate preset elements in Unity to create realistic simulation scenes and scenarios. The research was conducted in three phases. First, the capabilities of Unity as a real-world simulation package and for integration with other software were evaluated. These include integration with applications for complex computations and packages for online multi-user capabilities. Second, the feasibility of a local multi-LLM system was assessed, utilizing a local decoder-only and encoder-only model to optimize, enhance, and classify safety incident data from the proprietary dataset. Both the Unity simulator and LLM-classification frameworks were evaluated for their computational resource usage to ensure proper operation within the limited resources of a local system. Finally, these frameworks were combined into a cohesive system in which LLMs parse natural language input to create simulation scenes and generate actions within them. Once development was complete, the system was validated through a human factor experiment. Evaluation focused on the technical performance for the LLM-simulation system and on its ease of use. The human factor study was limited to a focus group with detailed feedback from each participant. Evaluation results demonstrate that the LLM-driven simulation system effectively communicated the safety training objectives, although the User Interface requires further refinement to enhance user experience, especially for those with limited simulation experience. The present dissertation contributes a viable framework for deploying privacy-centric, AI-enhanced training solutions in resource-constrained industrial environments.","abstract_has_math":false,"creators":["Haces-Garcia, Arturo 1998-"],"institution":"University of Houston","degree_name":"Doctor of Philosophy","degree_level":null,"degree_discipline":"Electrical Engineering","degree_department":null,"school":null,"contributors":[],"advisors":["Zhu, Weihang","Pan, Miao"],"committee_chairs":[],"committee_members":["Hu, Bin","Fu, Xin","Becker, Aaron"],"year":2026,"date_issued":"2026-05","date_published":"2026-05","updated_at":"2026-07-24T02:32:47Z","subjects":["LLM","Unity","Robotics","Mistral","Classification","Simulation","Virtual Reality","Large Language Model","Offshore platform","Safety","Training"],"languages":["English"],"rights":[],"rights_urls":[],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/10657/21533","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Zhu, Weihang","Pan, Miao"]},{"key":"dc:contributor.committeemember","label":"Committee Member","values":["Hu, Bin","Fu, Xin","Becker, Aaron"]},{"key":"dc:creator","label":"Author","values":["Haces-Garcia, Arturo 1998-"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2026-07-14T19:47:43Z"]},{"key":"dc:date.issued","label":"Date","values":["2026-05"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_discipline","label":"Discipline","values":["Electrical Engineering"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Doctor of Philosophy"]},{"key":"thesis:institution_name","label":"Thesis Institution Name","values":["University of Houston"]}]},{"id":"subjects_keywords","label":"Subjects and Keywords","entries":[{"key":"dc:subject","label":"Dc Subject","values":["LLM","Unity","Robotics","Mistral","Classification","Simulation","Virtual Reality","Large Language Model","Offshore platform","Safety","Training"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:language.iso","label":"Language (ISO)","values":["English"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/10657/21533"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["Offshore platforms represent high-risk environments where variable environmental conditions and equipment malfunctions require extensive safety training systems. Local LLMs (Large Language Models) offer advanced tools for data analysis to develop training protocols and enable users to maintain their data privacy. However, these models frequently pose significant barriers, requiring powerful computers and extensive, specialized datasets to be effective. Information from these proprietary sources is essential for developing corrective actions, learning from lessons, refining procedures, and training personnel to help prevent further incidents. This dissertation develops a scalable customizable system by leveraging proprietary datasets from offshore incidents, Local LLMs, and a custom simulation application. The proposed system uses the LLM to manipulate preset elements in Unity to create realistic simulation scenes and scenarios. The research was conducted in three phases. First, the capabilities of Unity as a real-world simulation package and for integration with other software were evaluated. These include integration with applications for complex computations and packages for online multi-user capabilities. Second, the feasibility of a local multi-LLM system was assessed, utilizing a local decoder-only and encoder-only model to optimize, enhance, and classify safety incident data from the proprietary dataset. Both the Unity simulator and LLM-classification frameworks were evaluated for their computational resource usage to ensure proper operation within the limited resources of a local system. Finally, these frameworks were combined into a cohesive system in which LLMs parse natural language input to create simulation scenes and generate actions within them. Once development was complete, the system was validated through a human factor experiment. Evaluation focused on the technical performance for the LLM-simulation system and on its ease of use. The human factor study was limited to a focus group with detailed feedback from each participant. Evaluation results demonstrate that the LLM-driven simulation system effectively communicated the safety training objectives, although the User Interface requires further refinement to enhance user experience, especially for those with limited simulation experience. The present dissertation contributes a viable framework for deploying privacy-centric, AI-enhanced training solutions in resource-constrained industrial environments."]},{"key":"dc:format.mimetype","label":"Dc Format Mimetype","values":["application/pdf"]},{"key":"dc:title","label":"Title","values":["Improving Offshore Safety Through Data Analysis with Large Language Models and Accident Review with VR Simulation"]}]}],"canonical_facts":{"dc:contributor.advisor":["Zhu, Weihang","Pan, Miao"],"dc:contributor.committeemember":["Hu, Bin","Fu, Xin","Becker, Aaron"],"dc:creator":["Haces-Garcia, Arturo 1998-"],"dc:date.accessioned":["2026-07-14T19:47:43Z"],"dc:date.issued":["2026-05"],"dc:description.abstract":["Offshore platforms represent high-risk environments where variable environmental conditions and equipment malfunctions require extensive safety training systems. Local LLMs (Large Language Models) offer advanced tools for data analysis to develop training protocols and enable users to maintain their data privacy. However, these models frequently pose significant barriers, requiring powerful computers and extensive, specialized datasets to be effective. Information from these proprietary sources is essential for developing corrective actions, learning from lessons, refining procedures, and training personnel to help prevent further incidents. This dissertation develops a scalable customizable system by leveraging proprietary datasets from offshore incidents, Local LLMs, and a custom simulation application. The proposed system uses the LLM to manipulate preset elements in Unity to create realistic simulation scenes and scenarios. The research was conducted in three phases. First, the capabilities of Unity as a real-world simulation package and for integration with other software were evaluated. These include integration with applications for complex computations and packages for online multi-user capabilities. Second, the feasibility of a local multi-LLM system was assessed, utilizing a local decoder-only and encoder-only model to optimize, enhance, and classify safety incident data from the proprietary dataset. Both the Unity simulator and LLM-classification frameworks were evaluated for their computational resource usage to ensure proper operation within the limited resources of a local system. Finally, these frameworks were combined into a cohesive system in which LLMs parse natural language input to create simulation scenes and generate actions within them. Once development was complete, the system was validated through a human factor experiment. Evaluation focused on the technical performance for the LLM-simulation system and on its ease of use. The human factor study was limited to a focus group with detailed feedback from each participant. Evaluation results demonstrate that the LLM-driven simulation system effectively communicated the safety training objectives, although the User Interface requires further refinement to enhance user experience, especially for those with limited simulation experience. The present dissertation contributes a viable framework for deploying privacy-centric, AI-enhanced training solutions in resource-constrained industrial environments."],"dc:format.mimetype":["application/pdf"],"dc:identifier.uri":["https://hdl.handle.net/10657/21533"],"dc:language.iso":["English"],"dc:subject":["LLM","Unity","Robotics","Mistral","Classification","Simulation","Virtual Reality","Large Language Model","Offshore platform","Safety","Training"],"dc:title":["Improving Offshore Safety Through Data Analysis with Large Language Models and Accident Review with VR Simulation"],"dc:type":["Thesis"],"thesis:degree_discipline":["Electrical Engineering"],"thesis:degree_name":["Doctor of Philosophy"],"thesis:institution_name":["University of Houston"]},"updated_at":"2026-07-24T02:32:47Z"}