[1]Temporal--Spatial Attention Network: A Novel Axial Piston Pump Coupled Fault Diagnosis Method
问题域
多源耦合故障的诊断问题
方法
（1）multihead attention-based temporal feature extraction model is constructed to extract periodic fault-sensitive features of each sensor on the temporal scale
(2) multihead attention-based spatial feature extraction model is constructed to fuse the extracted temporal features of each sensor and extract fault-sensitive features between different sensors on a spatial scale,
[2] MPINet: Multiscale Physics-Informed Network for Bearing Fault Diagnosis With Small Samples
问题域
故障样本不足
方法
（1）a physicsinformed block (PIB) is developed to extract fault features, which is customized for each failure mode.
（2）multiple independently trained PIBs encode the physical knowledge of their corresponding failure mode into the model, and thus yield multiscale fault features.
[3] Feature Generating Network With Attribute-Consistency for Zero-Shot Fault Diagnosis
问题域
zero-shot fault diagnosis
方法
(1)unseen fault class generation: The attribute consistency constraint adopted in data generation can make the generated data represent their attribute well
(2) discriminative feature transformation:a concatenation operation is used to transforming the generated samples into more discriminative representations.
[4] Physical Graph-Based Spatiotemporal Fusion Approach forProcess Fault Diagnosis
问题域
数据驱动的方法往往忽略了系统内的内在物理相关性，并且缺少为故障诊断提供可信解释的可靠机制。
方法
（1）a graph convolution network (GCN)-based node spatial encoding module is integrated.
（2）to capture temporal dependencies in fault data,a spatio temporal feature fusion module based on the long short-term memory network (LSTM) is employed.
（3）a fault diagnosis analysis method based on node masking is designed to enhance the interpretability of the model
[5] Generalized Out-of-Distribution Fault Diagnosis (GOOFD) via Internal Contrastive Learning
问题域
开发集成探测、分类和开集故障诊断体系 （多任务学习）
方法
(1)a unified fault diagnosis method based on internal contrastive learning and Mahalanobis distance is put forward to underpin the proposed generalized framework.
(2)The method involves feature extraction through internal contrastive learning and outlier recognition based on the Mahalanobis distance
[6] Multidomain Class-Imbalance Generalization With Fault Relationship-Induced Augmentation for Intelligent Fault Diagnosis
问题域
多源域域自适应的类别不平衡问题
方法
（1）a class-unbiased weak diagnostic model is trained using an under-sampling strategy to regularize the semantic information of generated samples in the subsequent stages
（2）the relationship between fault and normal classes in source domains is explored, which helps mappers convert normal-class samples to specific fault-class samples.
（3）a class-unbiased domain-agnostic diagnostic model is obtained by training on generated and real datasets.
[7] Prior Knowledge-Augmented Meta-Learning for Fine-Grained Fault Diagnosis
问题域
few-shot learning and new fault emerging issue
方法
(1) In the training phase, pretraining is used to incorporate prior knowledge of fault mechanism into the model that PKAML(train-test) can effectively handle low discriminative features among new fine-grained fault categories.
(2) In the test phase, Grad-CAM is embedded into the model to generate heatmaps as prior knowledge.
[8] A Multiattribute Learning Model for Zero-Sample Mechanical Fault Diagnosis
问题域
zero-sample learning
方法
（1）A multiattribute learning model with multiclass attribute classifiers is proposed for zero-sample mechanical fault diagnosis, making the fault diagnosis possible without the target fault samples.
（2）A Resnet-based CNN is constructed to synchronously achieve feature extraction and attribute classifiers construction, allowing for the automatic extraction of highquality features and direct prediction of attribute values.
[9] A Graph Attention Based Multichannel Transfer Learning Network for Wheelset Bearing Fault Diagnosis With Nonshared Fault Classes
问题域
source domain has more working conditions, fault types, and samples than the target domain
方法
（1）The recurrence graph attention residual network (ResGANet) is introduced for feature extraction, which can explicitly fit the relationship between samples.
（2）A multichannel diagnosis decision strategy is adopted to make fuse the diagnosis results of multiple ResGANet subnets. By jointly training multiple source domains, negative impact between different distribution spaces can be reduced.
（3）A nonshared class filtering mechanism is proposed, which enables the model to effectively fit additional fault features in the source domain. The spatial distribution weighting strategy is used to filter negative transfer of samples.
[10] Knowledge correlation graph-guided multi-source interaction domain adaptation network for rotating machinery fault diagnosis
问题域
multi-source domain adaptation
方法
（1） Proposing a multi-source interaction domain adaptation network by considering the interactions of data categories in
different domains for rotating machinery fault diagnosis.
（2） Applying random mini-batch to update learned comprehensive feature representations, to improve the knowledge
interaction between the current and the following epochs.
（3）Constructing a knowledge correlation graph to realize the knowledge interaction in various domains, thus fully lever
aging the knowledge from multi-source domains to participate in model construction.
[11] Self-Supervised-Enabled Open-Set Cross-Domain Fault Diagnosis Method for Rotating Machinery
问题域
open-set fault diagnosis (new and unknown faults)
方法
(1) self-supervised contrastive learning is used to extract the structural information and obtain robust discriminative features
(2)a novel class identification method based on the designed squeeze confidence rule is proposed to detect unknown- and known-class faults
(3)a pseudolabel consistency self-training method is used to reduce domain discrepancy, and the selected high-reliable unknown fault samples are used in the training process as effective supervised information to improve classification performance
[12] Variable-Wise Stacked Temporal Autoencoder for Intelligent Fault Diagnosis of Industrial Systems
问题背景
fault diagnosis in dynamic industrial systems
方法
（1）A variable-wise strategy is proposed to identify the optimal fault-relevant variables with DF and fault label information.
（2）A model named TAE is designed to capture the temporal and the spatial feature synchronously and model the complex dependencies of dynamic samples during reconstruction.
（3）A comprehensive fault diagnosis scheme is constructed with a bank of VW-STAEs and a fully connected layer.
[13] A Unified Framework With Incremental Learning Capacity for Industrial Fault Detection and Classification
问题域
增量学习（new data and fault types）故障探测和分类集成框架体系
方法
（1）Industrial fault detection and classification tasks can be synchronously achieved on new data samples in an incremental manner
（2）A unified framework where fault diagnosis can be performed directly on newly collected data without dimensionality reduction or any distributional assumptions is proposed.
（3）A decision tree-based ISVDD with an adaptive threshold is proposed to perform fault diagnosis tasks in kernel space.
[14]Multivariate Fusion Covariance Matrix Network and Its Application in Multichannel Fault Diagnosis With Fewer Training Samples
问题域
insufficient training samples
方法
a novel MFCMN is presented for multichannel fault feature learning, which preserves the linkage of data from different channels
[15] From Coarse to Fine: Hierarchical Zero-Shot Fault Diagnosis With Multigrained Attributes
问题域
zero-shot learning considering fault attribute
方法
（1）a PDKT strategy is designed to facilitate the knowledge transfer from coarse-grained attributes to fine-grained attributes, thereby enhancing prediction accuracy.
（2）An MFFI strategy is developed to integrate differentgrained attributes with the help of the tree structure of fault classes.
[16]Gaussian Mixture Variational-Based Transformer Domain Adaptation Fault Diagnosis Method and Its Application in Bearing Fault Diagnosis
问题域
the issue of loss oscillation and slow convergence
方法
（1）A feature extractor based on transformer layers is designed to capture long-term dependence information and local features.
（2）a domain alignment term is proposed to project the features learned from both working conditions into the common assistance distribution and make them follow the same distribution after the alignment process
（3）a Gaussian mixture is utilized to build the common assistance distribution.
[17] Distribution Character-Guided Projection Replay Network for Class-Incremental Fault Diagnosis of Rotating Machinery
问题域
th issue of class incremental leanring
方法
（1）a distribution projection replay (DPR) module is designed to store the distribution information of former fault classes and replay the general fault knowledge in the next incremental training stage
（2）a prototype adaptive update (PAU) module is further developed to avoid biased prediction of frequent parameter updates faced by existing fault incremental diagnosis methods.
[18]A novel triage-based fault diagnosis method for chemical process
问题背景
当前的故障诊断方法使用单一模型和相同的特征输入来训练和检测所有故障，导致忽略了故障之间的相关性和差异性，并且故障诊断性能较差
方法
(1) a novel fault diagnosis method named triage-based convolutional neural network (TrCNN) for fault diagnosis is proposed. Initially, the fault set is partitioned into distinct triage types.
(2) distinct models are formulated and applied to their respective triage types in the sub networks layer, while a triage network is developed in the triage layer.
(3) the models from the triage layer and sub networks layer come together to constitute the triage fault diagnosis system.
[19] Fault diagnosis based oncounterfactual inference for the batch fermentation process
问题背景
现有的故障诊断方法存在无法评估故障幅度或效率低下的缺点（偏故障溯源）
方法
（1） quality-related process variables are selected using mutual information (MI)
（2） a two-dimensional moving window is used to obtain input sequences from the process data
（3） two statistics from the latent and residual domains of the CNN-VAE model are constructed for fault detection， FDCI is used to locate the root cause of a fault
[20] A novel label-aware global graph construction method and spiking-codedgraph neural network for intelligent process fault diagnosis
问题背景
traditional techniques often face difficulties in handling large-scale data characterized by complex structures and relationships.
方法
（1）Using the label-aware method to transform data into graph could capture the intrinsic correlations within the industrial process data.
（2）By integrating graph-based data representation with spiking neural mechanisms, a novel theoretical framework was developed to learn data feature representation, focusing on balancing high accuracy and efficiency
（3）a weighted loss function was introduced to address data imbalance issues, enhancing the framework's robustness.
