Domain adaptation in fault diagnosis can efficiently handle different data distributions by co-training source and target domain data. However, the source domain data may not be accessible due to privacy or memory issues. An effective solution is to perform unsupervised parameter tuning of the source model using unlabeled target domain data. The focus of this article is to perform unsupervised domain adaptation using the knowledge of multiple pretrained source models in Industrial Internet of Things (IIoT) without accessing source data. Almost all of the current researches in this setting requires tuning the entire backbone network, increasing computing costs as model size or source domains grow. To this end, this article proposes an efficient multidomain knowledge fusion adaptation (EMDKFA) method, applied to source free cross-domain fault diagnosis. Unlike previous methods, this method greatly reduces the number of trainable parameters and saves computational overhead via low-rank reparameterization strategy. To better fuse and transfer multiple pretrained models, we propose an unsupervised domain weight initialization method based on nuclear norm, initially focusing credible source models. Furthermore, the multidomain weighted entropy minimization penalty and noisy label learning are designed to promote correct allocation of target samples. Comprehensive cross-condition and cross-device experiments demonstrate the method's effectiveness.

Traditional collaborative filtering techniques deal with no time-effect data. For recommender systems, as user preferences change over time, it can-make the preferences of recommendation results change. In addition, the data explosion makes the model process more and more data over time, which greatly increases the computational cost. To solve the above problems, a similarity index time-effect collaborative filtering algorithm based on Attentional Double BP Network (SITCF-ADBP) is proposed. Firstly, three time-effect functions are proposed for processing evaluation metrics of user preferences changed over time. Secondly, after clustering the training samples using K-Nearest Neighbor (KNN), a set of similarity index screening principles is specified, and only one sample with high similarity is retained to reduce the input of the model. Finally, a Double BP Network incorporating a channel attention mechanism is designed to process the adequately processed data. Experiments were conducted on the MovieLens datasets (ml-1m) and (ml-100k), wherein our proposed approach was benchmarked against several state-of-the-art (SOTA) deep learning methods. The obtained results indicate that the model can effectively cope with the time-effect and data explosion of the recommender system.

Bearing is a common rotating component, the health status of bearing affects the operation and maintenance of equipment. Thus, the prediction of bearing remaining useful life is of great significance. The remaining useful life prediction based on neural network has black box property, which makes the prediction result may be contrary to the actual physical law. In this paper, a physics guided long short-term memory (LSTM) network is proposed based on the change trend of the time-frequency domain feature indicators of bearings in the process of degradation. Specifically, indexes such as monotonicity are used to select feature indicators that are highly trendy in the process of bearings degradation. On this basis, a regularization term based on the consistent variation of the feature indicators and the remaining useful life (RUL) in the process of bearing degradation is constructed to make the result of the network more consistent with the actual physical law. Meanwhile, the variation of feature indicators is used as dynamic weight to enhance the potential physical consistency. The experimental comparison results show that the prediction results of the network are more accurate and consistent with the actual physical laws with the guidance of physical prior knowledge.

Failure of cooling fan will have a great impact on the operation of the air-conditioning cooling system. Traditional anomaly detection methods for fans rely on manual feature extraction, which is easy to lose effective information. Deep learning-based methods have potential to overcome such a problem, however, it is difficult to train the network without enough monitoring data. In operation, cooling usually worked at high speed, and its monitoring data contain much high-speed rotational noise. This will confuse the networks to focus on high-energy noise while lose the weak but useful information. To solve these problems, we propose a semi-supervised multi-scale deep adversarial model (SMDAM) for fan anomaly detection. In SMDAM, we propose the Linear Spectral Line Elimination (LSLE) technique to reduce the interference of rotational noise spectral lines on network training. Subsequently, a multi-scale feature extraction module is introduced to enrich diagnostic information, which extracts features from different scales without increasing the depth of the convolutional layer. Finally, two engineering data sets are used to verify the superiority of the proposed method. The results show that SMDAM is superior to other methods.

Recently, with the large-scale application of solid-state drives (SSDs), the failure of SSDs has become the main reliability issue in data centers. SSD manufacturers developed self-monitoring, analysis, and reporting technology (SMART) to indicate the health status of SSDs. However, for the newly enabled and new types of SSDs in the data centers, lack of enough labeled data will be a critical problem, making failure prediction for SSDs a difficult task. To this end, this article proposes a multi-instance adversarial learning domain adaptation network (MALDAN) for coping with this task. A multi-instance learning method with an attention mechanism is designed to solve the problem of features extraction from unlabeled data by assigning weights to features over the lifespan. Moreover, the distribution differences between different SSD models prevent the knowledge of the labeled information from being used for failure prediction of the unlabeled data, and an adversarial domain adaptation (DA) method is used to align the distributions. Finally, the proposed method is verified on Alibaba’s dataset and shows much better performance than other methods.

Recently, domain adaptation methods are widely applied to intelligent failure prediction to solve the problem of lack of labeled data for newly designed equipment. Generally, domain adaptation assumes that the target to be predicted is consistent with the label space of the source domain. However, a more practical scenario may be that the distribution of the source domain is a subset of the target's since the new types of equipment may have different failure characteristics. Learning from a single source domain is insufficient to support the prediction task on the target domain. Multiple-source domain adaptation becomes a desperately needed solution. To this end, this paper proposes a locally weighted multi-domain collaborative adaptation method (LWMDCA). A locally weighted technique is introduced to construct the multi-source collaborative joint domain based on the similarity weighted regular coefficient to provide complete diagnostic knowledge. Moreover, to further utilize the knowledge, a new feature extractor based on multi-instance learning with attention mechanism is designed to focus on the important failure characteristics. Finally, comparative analysis with many state-of-the-art failure prediction methods using Alibaba's SSDs datasets is presented to demonstrate the effectiveness of the proposed method.

Recently, domain adaptation has been widely investigated for solving domain shift problems in mechanical fault diagnosis. Generally, domain adaptation-based diagnosis methods assume that the source and target domains have identical label space. However, a more realistic scenario is that the label space of the target domain is a subset of the source domain, which may introduce two problems: mismatching caused by the occurrence of outlier classes and misalignment caused by overweighting of the uncertain samples near the classification boundary. To address the above problems, a balanced adversarial domain adaptation network (BADAN) is proposed for intelligent fault diagnosis tasks under partial transfer scenarios. A balanced strategy is introduced to augment classes in the target domain using source samples. On this basis, an adversarial domain adaptation method with class-level weight is designed to avoid negative transfer by filtering outlier classes and promote positive transfer by mitigating the distribution discrepancy of shared classes. Moreover, to alleviate the misalignment problem, a complement objective function for ensuring alignment direction toward the support of the source classes is derived by minimizing their predicted scores of the incorrect classes rather than ground-truth classes. Extensive partial transfer diagnosis tasks constructed on two machines are used to demonstrate the robust and superior performance of BADAN.