A Comparative Study on AI/ML-Based Transient Temperature Predictions and Real-Time Operational Transient Temperature Data of Coke Drum

Abstract

Abstract Uncertainties in real-time operational practice result in the deviation of actual as-operated time-varying parameters from planned operational time-varying parameters. Coke drums experience severe transient thermal-induced cyclic stresses resulting in unanticipated, low cycle, fatigue damage. The health of coke drums is of great importance as it is associated with the increased need for inspection frequencies and unwelcome shutdowns. As a result, the need for online monitoring and prediction of the integrity of components of coke drums is growing among plant operators. In setting up an asset health integrity and monitoring system, operators must decide among timeliness, precision, and reliability. The time-varying transient temperature is the dominant parameter that has a stimulating effect on the fatigue life of coke drums. Establishing a methodology to predict the future transient temperature of typical operational sequences is of great significance to predict the health status of coke drums. In this work, Artificial Intelligence and Machine learning (AI/ML) are used to develop a predictive model. Using the data of transient thermal temperature of a coke drum recorded during its operation, the AI/ML-based prediction model is trained. The case study of an effort to generate transient temperature predictions using AI/ML algorithms is presented in this paper. The results obtained from this model are compared to the realtime as-operated transient temperature data set. Further, this AI/ML-based predictive model is planned to be used in a Digital Twin for preventive maintenance of coke drums.