Gaussian process flow fusion physical model for fatigue evaluation of petrochemical equipment considering residual analysis

Abstract

Petrochemical equipment is subjected to frequent discontinuous operation and underwent alternate loads, leading to fatigue failure of metal structure. Due to the fact that petrochemical equipment operates in a harsh environment, it’s necessary to assess the fatigue state of metal materials to ensure production safety. The residual stress of metal materials is closely related to the fatigue evolution. Physical model is unable to express the possibility of the evolution of residual stress. Data-driven model requires large-scale data to optimize the model and may not conform to the laws of physical evolution. Therefore, this paper proposes a probabilistic model integrating physical model and Gaussian Process Flow (GPF) for fatigue assessment of metal structure of petrochemical equipment under scarce data. Firstly, different evolutionary trajectories with stochastic parameters are obtained by applying stochastic differential equation (SDE). The uncertainty of SDE is assessed by Gaussian process (GP). Then, from the perspective of a particle filter, the physical model is combined to develop a fusion-driven model to infer the importance of the residual analysis. An example of an irreplaceable metal material in petrochemical equipment is studied to demonstrate the effectiveness of this model. The model proposed could provide a promising solution for preventing fatigue failure at petrochemical processing facilities.