Combining computer fluid dynamics, virtual reality, machine and deep learning for risk and reliability modeling in the process industry

• Grant number: 22/06726-2
• Support Opportunities: Regular Research Grants
• Duration: November 01, 2022 - October 31, 2024
• Field of knowledge: Engineering - Chemical Engineering
• Convênio/Acordo: CONFAP - National Council of State Research Support Foundations
• Principal Investigator: Savio Souza Venancio Vianna
• Grantee: Savio Souza Venancio Vianna
• Host Institution: Faculdade de Engenharia Química (FEQ). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Associated researchers: Flávio Vasconcelos da Silva ; Luz Adriana Alvarez Toro

Abstract

This project proposes a comprehensive methodology to tackle undesirable situations in the process industry (e.g., leakage of hazardous chemical substances) to avoid severe consequencesto humans, the environment, and technical systems. The methodology development relies on machine and deep learning models for hardware and human anomaly detection, Computational Fuid Dynamics (CFD), and virtual reality. The learning process involves the interaction between these techniques as follows: many accidental scenarios with various inputs (e.g., leakage location, wind speed, wind direction) feed CFD models, which simulategas dispersion in the surrounding environment. The generated CFD data is used in two ways. Firstly, gas concentrations time series on several points of the processing unit (e.g., sensor locations) will be inputted into the hardware anomaly detection model. Secondly, 3Dmesh-grids feed a serious game involving a virtual reality approaching the real processing unit, which will simulate the occurrence of the undesired event. The serious game objectiveis that the player (e.g., emergency operator) needs to stop or mitigate the leakage considering the dispersion behavior. During the game, the player will carry wearables to monitor biological signals that will enter the human anomaly detection model. It has the potential to turn into a technological product to assist in training emergency teams. Their immersion, with wearables, in virtual scenarios that mimic rare stressful situations is an opportunityto gather data to enhance human reliability models, training programs, and emergency responses, which are of utmost importance in the process industry. Finally, whenever the CFD machine and deep learning models are adjusted and trained, they can be part of the processing unit safety barriers as additional tools to support rapid decision-making toward human, environmental, and system integrity. (AU)