Fault Diagnosis of Chillers Based on Multi-scale Domain Generative Networks

doi:10.3969/j.issn.1004-132X.2026.03.018

Abstract

Abstract:

To address the issues that domain generalization methods relied on data from multiple source domains for model training， while obtaining multi-operating condition data for chiller units was challenging， a fault diagnosis method was proposed for chiller units based on multi-scale domain generative network （MSDGN）. First， a multi-scale encoder-decoder convolutional neural network was used to extract multi-scale features from source domain data， and learnable weight parameters were introduced to dynamically adjust the importance of features at each scale to enhance the diversity of the extended domain. Then， focal loss was applied to strengthen the penalty for semantically inconsistent samples， improving the semantic consistency of the extended domain. A combination of reverse metric learning strategies and a domain classifier was used to maximize the distribution difference between sources and extended domains， thereby achieving diversity in the training data. Finally， a domain adversarial strategy was employed to extract domain-invariant features from both the source and extended domains， and a triplet loss was introduced to minimize the distribution difference across multiple domains， enabling fault diagnosis for unknown operating conditions. By generating the extended domain， the model’s fault diagnosis performance was improved under unknown conditions. The proposed method was experimentally validated using ASHRAE 1043-RP dataset and a metro dataset from a certain city. The results on ASHRAE 1043-RP dataset demonstrate that the proposed method effectively identifies faults even when target operating conditions are unseen， achieving a maximum diagnosis accuracy of 98.19%. Results on the metro dataset indicate that the proposed method exhibits practical applicability in real-world scenarios. Compared with existing methods， the proposed approach achieves superior fault diagnosis performance.

Key words: chiller, single domain generalization, fault diagnosis, domain adversarial, metric learning

摘要：

针对领域泛化方法依赖于多个源域数据进行模型训练，但是冷水机组的多工况运行数据获取困难的问题，提出了一种基于多尺度域生成网络（MSDGN）的冷水机组故障诊断方法。利用多尺度编解码卷积神经网络提取源域数据的多尺度特征，引入可学习权重参数动态调整各尺度特征的重要性以增强扩展域的多样性；利用焦损失加强对语义不一致样本的惩罚力度，提高扩展域的语义一致性。结合反向度量学习策略和域分类器最大化源域与扩展域的分布差异，实现训练数据的多样性；采用域对抗策略提取源域和扩展域的域不变特征，引入三元组损失最小化多域之间的分布差异，实现对未知工况的故障诊断。通过生成扩展域来提高模型在未知工况下的故障诊断性能。利用ASHRAE 1043-RP数据集和某城市地铁数据集对所提方法进行了实验验证。ASHRAE 1043-RP数据集实验结果表明，该方法能够在目标工况不可见的情况下有效识别故障，最高诊断正确率高达98.19%。地铁数据集实验结果表明，该方法在真实场景中有一定的实用性。与现有方法相比，所提方法具有更好的故障诊断效果。

关键词: 冷水机组, 单域泛化, 故障诊断, 域对抗, 度量学习

CLC Number:

TP277

GAO Xuejin, WANG Xuan, JIANG Kexin, GAO Huihui, QI Yongsheng. Fault Diagnosis of Chillers Based on Multi-scale Domain Generative Networks[J]. China Mechanical Engineering, 2026, 37(3): 688-696.

高学金, 王璇, 姜渴鑫, 高慧慧, 齐咏生. 基于多尺度域生成网络的冷水机组故障诊断[J]. 中国机械工程, 2026, 37(3): 688-696.

Figures/Tables 15

References 26

[1]	AFSHARI A， FRIEDRICH L. A Proposal to Introduce Tradable Energy Savings Certificates in the Emirate of Abu Dhabi［J］. Renewable and Sustainable Energy Reviews， 2016， 55： 1342-1351.
[2]	KATIPAMULA S， BRAMBLEY M R. Review Article： Methods for Fault Detection， Diagnostics， and Prognostics for Building Systems—a Review， Part I［J］. HVAC&R Research， 2005， 11（1）： 3-25.
[3]	KIM W， KATIPAMULA S. A Review of Fault Detection and Diagnostics Methods for Building Systems［J］. Science and Technology for the Built Environment， 2018， 24（1）： 3-21.
[4]	NIU Jide， ZHOU Ruoyu， TIAN Zhe， et al. Energy-saving Potential Analysis for a 24-hour Operating Chiller Plant Using the Model-based Global Optimization Method［J］. Journal of Building Engineering， 2023， 69： 106213.
[5]	ZHANG Hanyuan， YANG Wenxin， YI Weilin， et al. Imbalanced Data Based Fault Diagnosis of the Chiller via Integrating a New Resampling Technique with an Improved Ensemble Extreme Learning Machine［J］. Journal of Building Engineering， 2023， 70： 106338.
[6]	ZHANG Boyan， WANG Peng， LIU Guangyu， et al. Diagnosis of Single and Multiple-source Faults of Chiller Sensors Using EWEEMD-ICKNN by Time Sequence Denoising and Non-Gaussian Distribution Feature Extraction［J］. Energy and Buildings， 2023， 298： 113572.
[7]	温楷儒，陈祝云，黄如意，等. 基于可解释时空图卷积网络的多传感数据融合诊断方法［J］. 机械工程学报， 2024， 60（12）： 158-167.
	WEN Kairu， CHEN Zhuyun， HUANG Ruyi， et al. Multi-sensor Data Fusion Diagnosis Method Based on Interpretable Spatial-Temporal Graph Convolutional Network［J］. Journal of Mechanical Engineering， 2024， 60（12）： 158-167.
[8]	SCHÖLKOPF B. Causality for Machine Learning［M］∥Probabilistic and Causal Inference. New York： ACM， 2022： 765-804.
[9]	Van de SAND R， CORASANITI S， REIFF-STEPHAN J. Data-driven Fault Diagnosis for Heterogeneous Chillers Using Domain Adaptation Techniques［J］. Control Engineering Practice， 2021， 112： 104815.
[10]	ZHU Xu， CHEN Kang， ANDUV B， et al. Transfer Learning Based Methodology for Migration and Application of Fault Detection and Diagnosis between Building Chillers for Improving Energy Efficiency［J］. Building and Environment， 2021， 200： 107957.
[11]	WANG Jindong， LAN Cuiling， LIU Chang， et al. Generalizing to Unseen Domains： a Survey on Domain Generalization［J］. IEEE Transactions on Knowledge and Data Engineering， 2023， 35（8）： 8052-8072.
[12]	JIN Xin， LAN Cuiling， ZENG Wenjun， et al. Style Normalization and Restitution for Domain Generalization and Adaptation［J］. IEEE Transactions on Multimedia， 2021， 24： 3636-3651.
[13]	YUE Xiangyu， ZHANG Yang， ZHAO Sicheng， et al. Domain Randomization and Pyramid Consistency： Simulation-to-Real Generalization without Accessing Target Domain Data［C］∥2019 IEEE/CVF International Conference on Computer Vision （ICCV）. Seoul， 2019： 2100-2110.
[14]	WU Guile， GONG Shaogang. Collaborative Optimization and Aggregation for Decentralized Domain Generalization and Adaptation［C］∥2021 IEEE/CVF International Conference on Computer Vision （ICCV）. IEEE， 2021： 6464-6473.
[15]	BLANCHARD G， DESHMUKH A A， DOGAN U， et al. Domain Generalization by Marginal Transfer Learning［J］. Journal of Machine Learning Research， 2021， 22（2）：1-55.
[16]	ZHAO S， GONG M， LIU T， et al. Domain Generalization via Entropy Regularization［J］. Advances in Neural Information Processing Systems， 2020， 33： 16096-16107.
[17]	ZHU Ronghang， LI Sheng. Self-supervised Universal Domain Adaptation with Adaptive Memory Separation［C］∥2021 IEEE International Conference on Data Mining （ICDM）. Auckland， 2021： 1547-1552.
[18]	YU Xi， TSENG H H， YOO S， et al. INSURE： an Information Theory iNspired diSentanglement and pURification modEl for Domain Generalization［J］. IEEE Transactions on Image Processing， 2024， 33： 3508-3519.
[19]	LI Qi， CHEN Liang， KONG Lin， et al. Cross-domain Augmentation Diagnosis： an Adversarial Domain-augmented Generalization Method for Fault Diagnosis under Unseen Working Conditions［J］. Reliability Engineering & System Safety， 2023， 234： 109171.
[20]	ZHOU Fan， JIANG Zhuqing， SHUI Changjian， et al. Domain Generalization via Optimal Transport with Metric Similarity Learning［J］. Neurocomputing， 2021， 456： 469-480.
[21]	JIANG Kexin， GAO Xuejin， GAO Huihui， et al. VIT-GADG： a Generative Domain-generalized Framework for Chillers Fault Diagnosis under Unseen Working Conditions［J］. IEEE Transactions on Instrumentation and Measurement， 2023， 72： 3527413.
[22]	GANIN Y， USTINOVA E， AJAKAN H， et al. Domain-adversarial Training of Neural Networks［M］∥Domain Adaptation in Computer Vision Applications. Cham： Springer International Publishing， 2017： 189-209.
[23]	LIN T Y， GOYAL P， GIRSHICK R， et al. Focal Loss for DENSE Object Detection［J］.IEEE Transactions on Pattern Analysis & Machine Intelligence， 2020，42（2）：318-327.
[24]	COMSTOCK M C. Development of Analysis Tools for the Evaluation of Fault Detection and Diagnostics in Chillers［D］. West Lafayette： Purdue University， 1999.
[25]	ZHAO Chao， SHEN Weiming. Adversarial Mutual Information-guided Single Domain Generalization Network for Intelligent Fault Diagnosis［J］. IEEE Transactions on Industrial Informatics， 2023， 19（3）： 2909-2918.
[26]	PU Yuanyue， TANG Jian， LI Xuegang， et al. Single-domain Incremental Generation Network for Machinery Intelligent Fault Diagnosis under Unknown Working Speeds［J］. Advanced Engineering Informatics， 2024， 60： 102400.

诊断任务	源域数量	数据分布	目标域可见
深度学习	1	一致	是
领域自适应	≥1	不一致	是
多源域泛化	＞1	不一致	否
单源域泛化	1	不一致	否

诊断任务	源域数量	数据分布	目标域可见
深度学习	1	一致	是
领域自适应	≥1	不一致	是
多源域泛化	＞1	不一致	否
单源域泛化	1	不一致	否

工况	1043RP数据集		地铁数据集
工况	控制变量	数值/℃	日期
A	冷却水温度	40	9月19日
B	冷却水温度	45	9月21日
C	冷却水温度	50	9月22日

工况	1043RP数据集		地铁数据集
工况	控制变量	数值/℃	日期
A	冷却水温度	40	9月19日
B	冷却水温度	45	9月21日
C	冷却水温度	50	9月22日

诊断任务	源域	目标域
T₀	A	B
T₁	A	C
T₂	B	A
T₃	B	C
T₄	C	A
T₅	C	B