An intelligent quantitative risk assessment method for ammonia synthesis process

Abstract

• An intelligent risk assessment method is proposed to capture potential risks. • LSTM trained to provide predictions without mechanistic understanding. • The average MAPE of risk variables predicted by LSTM is only 0.6356%. • Cooling water failure and controller failure have a higher risk level. • Level controller failure has more serious harm in a short time. Safety assessment is a prerequisite for the normal operation of chemical process. Although it can reduce the frequency of accidents, it relies heavily on the experience of experts and is difficult to assess potential accident risks caused by dynamical chemical conditions. To address this problem, a novel intelligent quantitative risk assessment method (DYN-LSTM-QRA) is proposed based on dynamic mechanism model for ammonia synthesis process in this paper. The process under study is first simulated by Aspen Dynamics (DYN) to obtain dynamic data set under abnormal conditions. Then, the potential relationships between variables are deeply learnt by long short-term memory (LSTM) to represent complex mechanism relationship among predicted variables that have a direct impact on the hazard severity of accident. Finally, a quantitative risk assessment (QRA) method based on variable prediction is given to assess the risk of multiple conditions and their dynamic hazard range. The reliable control plans are also proposed to ensure the safe running of the process. The proposed method was applied to the leakage accident of the ammonia synthesis process, which effectiveness and reliability are proven by dynamic simulation.