Hangzhou Wang

Dynamic occupational risk model for offshore operations in harsh environments

Abstract The expansion of offshore oil exploitation into remote areas (e.g., Arctic) with harsh environments has significantly increased occupational risks. Among occupational accidents, slips, trips and falls from height (STFs) account for a significant portion. Thus, a dynamic risk assessment of the three main occupational accidents is meaningful to decrease offshore occupational risks. Bow-tie Models (BTs) were established in this study for the risk analysis of STFs considering extreme environmental factors. To relax the limitations of BTs, Bayesian networks (BNs) were developed based on BTs to dynamically assess risks of STFs. The occurrence and consequence probabilities of STFs were respectively calculated using BTs and BNs, and the obtained probabilities verified BNs׳ rationality and advantage. Furthermore, the probability adaptation for STFs was accomplished in a specific scenario with BNs. Finally, posterior probabilities of basic events were achieved through diagnostic analysis, and critical basic events were analyzed based on their posterior likelihood to cause occupational accidents. The highlight is systematically analyzing STF accidents for offshore operations and dynamically assessing their risks considering the harsh environmental factors. This study can guide the allocation of prevention resources and benefit the safety management of offshore operations.

Bayesian treed Gaussian process method for process monitoring

Abstract The Bayesian treed Gaussian method is introduced in this paper to implement process monitoring based on historical data. This method can cover the disturbances in a process and discover differences among individually monitored variables before and after an abnormal situation occurs. The analysis results from the historical values of each variable help to differentiate abnormal from normal states in the process. Here, the Tennessee Eastman process is studied to show the effectiveness of this method for process monitoring.

An Approximate Modelling Method for Industrial l-lysine Fermentation Process

l-lysine is an important chemical, usually produced by fed-batch fermentation process. Usually, feed stock compositions, reactant or product concentrations, and operating conditions vary with different fed-batches in this process. It is difficult to establish a kinetics-based model for an industrial fed-batch fermentation process. In this paper, we proposed a data-based approximate graphical modelling method to model this process. Variables values are treated as correlated Gaussian process. The methodology comprises of two important steps: i) the missing-data imputation within records, and ii) the dynamic Bayesian network learning, including structure learning, using the low order conditional independence method, and parameters learning, using the multivariate auto regressive method. The l-lysine fed-batch fermentation process is studied to demonstrate the effectiveness of this approximate modelling method.