A method for assessing the engineering quality of petrochemical devices based on Leaky-Bayesian inference

Engineering quality directly affects and restricts the “safe, stable, long-term, full and excellent” operation of large-scale petrochemical units. However, the accuracy of the existing evaluation technology is low, and it is difficult to ensure continuous and stable production and maximum benefit output. Therefore, in this work, we propose a quantitative method of evaluating petrochemical plant engineering quality based on Leaky-Bayesian inference. Firstly, the engineering quality indexes of each level of the evaluated device were screened, and a systematic analysis of the index layer and the criterion layer was carried out to obtain the corresponding index level. Secondly, the evaluation index system was transformed into a Bayesian network (BN) model, and the Leaky Noisy-or Gate (LNoG) theory was subsequently introduced to construct a Leaky-Bayesian inference model. Then, according to the index analysis and weight distribution results of the criterion layer, the nodes and conditional probabilities of the inference model were determined. Finally, the Bayesian formula was used to evaluate the quality of the petrochemical plant engineering quantitatively. The resulting Leaky-Bayesian inference model was verified for three real cases: a leakage explosion accident in a propylene oxide device, an explosion accident in an ethylene glycol device and an explosion accident in a raw oil buffer tank. The results using our model are consistent with the accident assessment results for these dangerous events. The comprehensive evaluation method, grey correlation degree evaluation method, traditional Bayesian inference model and our Leaky-Bayesian inference model were compared for the leakage and explosion accident of a propylene oxide device. Compared with the other methods, the engineering quality evaluation method based on Leaky-Bayesian inference has higher efficacy, rationality and accuracy.