基于融入决策边界贝叶斯网络的冷水机组故障诊断

Fault Diagnosis Based on A Bayesian Network with A Decision Boundary for Chillers

虚警率可调、单发与并发故障诊断、新故障识别是冷水机组故障诊断模型所期望具备的属性。本文提出了1种基于融入决策边界贝叶斯网络的故障诊断方法,通过调节决策边界的显著性水平,实现了对虚警率的可调节;通过构建故障与特征之间的关联矩阵,设置基于决策边界的诊断规则,实现了单发与并发故障的诊断和新故障的识别。为了验证所构建方法的有效性,使用ASHRAE RP-1043项目的实验数据对模型进行了评价。结果显示,所构建方法实现了在同一个诊断体系中的虚警率可调,对4种单发故障的诊断正确率为99.79%,对并发故障的诊断正确率为88.62%,以及对新故障识别的正确率不低于93.49%;与其他常规方法的比较结果显示,所提出方法表现了更好的诊断性能。

An adjustable false alarm rate,single and concurrent fault diagnosis,and new fault identification are the expected attributes of the fault diagnosis model for chillers.In this paper,a fault diagnosis method based on a Bayesian network incorporating a decision boundary was proposed.By adjusting the significance level of the decision boundary,the false alarm rate can be adjusted;the diagnosis of single and concurrent faults and the identification of new faults can be implemented by constructing a correlation matrix between faults and features and setting the diagnosis rules based on the decision boundary.To verify the effectiveness of the method,the model was evaluated by using experimental data from the ASHRAE RP-1043 project.The results showed that the method achieved adjustable false alarm rates within the same diagnostic system,with a diagnostic accuracy of 99.79%for four types of single engine faults,88.62%for concurrent faults,and not lower than 93.49%for new fault recognition;the results of comparison with other conventional methods showed that the proposed method exhibited better diagnostic performance.