Soft Evidence-Enhanced Object-Oriented Bayesian Network for Process Fault Diagnosis: A Case Study in Oilfield Transfer Station System

Summary With the development of digitalization and intelligence in the oil and gas industry, fast and accurate fault diagnosis is getting more attention day by day and the Bayesian network is widely used as an effective method in this field. However, oilfield transfer stations are highly complex, making it difficult to capture the dynamic response symptoms of fault modes. Meanwhile, existing Bayesian networks fail to consider the locational characteristics of equipment parameters during the modeling process, weakening the interpretability of fault propagation. In addition, the inadequate quantification of risk fluctuations between thresholds by hard evidence leads to low model adaptability. To overcome these limitations, we propose a soft evidence-enhanced object-oriented Bayesian network (OOBN). First, a simulation model is developed to generate accurate and reliable dynamic fault characteristics. Second, we incorporate the locations of equipment parameters into the Bayesian network, enhancing the interpretability of fault propagation. Then, soft evidence is used to quantify risk changes through probabilistic mapping, thereby reducing the misdiagnosis in the Bayesian network. Finally, the proposed Bayesian network is compared with Bayesian methods based on hard evidence and deep learning models and validated through multiple case studies, fully demonstrating the accuracy and robustness of the proposed model.