基于CNN与ResNet的离心式冷水机组故障诊断
Fault diagnosis of centrifugal chillers based on CNN and ResNet
摘要：
卷积神经网络（convolutional neural network，CNN）因其具有自主学习并从大量数据中提取特征的能力被广泛应用于各种领域，但在制冷系统故障诊断领域中应用较少。本文提出了一种基于卷积神经网络构建、适用于离心式冷水机组的故障诊断模型，并采用残差结构对模型进行了优化。对ASHRAE RP-1043项目离心式冷水机组7种典型故障的诊断结果显示：包含21个卷积层的ResNet_21模型的整体故障诊断正确率达到了99.40%，较浅层网络CNN_3提升7.48%；系统级故障中的制冷剂泄漏故障诊断正确率较CNN_14提升1.24%，达到98.55%；对正常工况的识别更准确，达到98.77%，虚警率降低1.43%；局部故障的诊断正确率均达到99.7%以上。
Abstract：
Convolutional neural network (CNN) is widely used in various fields due to its ability of autonomous learning and extracting features from a large amount of data, but it is rarely used in the field of refrigeration system fault diagnosis. In this paper, a fault diagnosis model for centrifugal chillers based on CNN is proposed, and the residual structure is used to optimize the model. The diagnosis results of seven typical faults of centrifugal chiller of ASHRAE RP-1043 project show that the overall fault diagnosis accuracy rate of ResNet_21 model with 21 convolutional layers is 99.40%, 7.48% higher than that of shallow network CNN_3. The refrigerant leakage fault diagnosis accuracy rate of system-level faults is 1.24% higher than that of CNN_14, reaching 98.55%. The identification of normal working conditions is more accurate, reaching 98.77%, and the false alarm rate is reduced by 1.43%. The diagnostic accuracy rate of local faults is above 99.7%.
Keywords:refrigerationsystem;centrifugalchiller;faultdetectionanddiagnosis;convolutionalneuralnetwork;residualneuralnetwork