The federated learning (FL) method based on model aggregation can balance data and protect data privacy, but the existing method is difficult to achieve the same effectiveness as the centralized learning method under data sharing. In addition, it is difficult for the existing federated model to realize the real-time update of the clients' network parameters, because it inhibits the optimal performance of the client. Therefore, this article proposes an asynchronous update paradigm of FL with real-time identification of the client's network parameters to tackle the shortcomings. First, we adopt the linear fusion method based on sequential filtering and fuse the parameters of federated center asynchronously considering communication delay, which can approach the diagnostic accuracy based on the centralized learning. Second, we establish the real-time identification method for the clients based on linear filtering with the new labeled samples obtained at nonequal intervals, which expects the client to acquire better performance. Finally, we test the fault classification ability of the proposed method based on the actual collected fault dataset and the test platform of bearing fault dataset.

Efficient and accurate fault diagnosis of rotating machinery is extremely important. Fault diagnosis methods using vibration signals based on convolutional neural networks (CNNs) have become increasingly mature. They often struggle with capturing the temporal dynamics of vibration signals. To overcome this, the application of Transformer-based Vision Transformer (ViT) methods to fault diagnosis is gaining attraction. Nonetheless, these methods typically require extensive preprocessing, which increases computational complexity, potentially reducing the efficiency of the diagnosis process. Addressing this gap, this paper presents the Time Series Vision Transformer (TSViT), tailored for effective fault diagnosis. The TSViT incorporates a convolutional layer to extract local features from vibration signals alongside a transformer encoder to discern long-term temporal patterns. A thorough experimental comparison of three diverse datasets demonstrates the TSViT’s effectiveness and adaptability. Moreover, the paper delves into the influence of hyperparameter tuning on the model’s performance, computational demand, and parameter count. Remarkably, the TSViT achieves an unprecedented 100% average accuracy on two of the test sets and 99.99% on the other, showcasing its exceptional fault diagnosis capabilities for rotating machinery. The implementation of this model will bring significant economic benefits.

Although the federated learning method has the ability to balance data and protect data privacy by means of model aggregation, while the existing methods are difficult to achieve the effectiveness of centralized learning under data sharing. The existing federated structure only has a certain degree of confidentiality for data privacy, that is to say, each client can reconstruct a part of the information of other clients based on the model parameters shared between the server and the clients under certain conditions. In order to make the federated learning mechanism more confidential, we breaks the existing mechanism that the parameters between the federated model and the client model are completely shared, and establishes a new asynchronous quasi-cloud/edge/client collaborative federated learning mechanism. We construct a hierarchical multi-level confidential communication network, where the network parameters are shared in a way of quasi-cloud/edge/client coordination without data communication. The cloud and the edges respectively use the sequential Kalman filter algorithm to perform an asynchronous fusion of the network parameters uploaded in their respective fusion centers for the next round of updates; The effectiveness of the proposed algorithm is verified on the data of a type of rotating machinery

Deep learning has been widely applied in the field of rotating machinery fault diagnosis. Existing deep learning-based fault diagnosis methods commonly utilize deep neural networks to extract features from raw data and employ fully connected layers for classification of the extracted features. However, in existing methods, fault type-related information is coupled with irrelevant information (e.g., load variations, fault sizes, and system intrinsic properties) at the fully connected layer. The lack of decoupling may result in fault type-related information being overwhelmed by irrelevant factors, thereby hindering accurate fault type determination. We propose that there exists an intrinsic feature in the data, where information is interrelated and closely tied to the fault type. This feature remains highly stable under changes in irrelevant factors, such as load and fault size, and provides an accurate representation of the fault type. By extracting and analyzing fault-type intrinsic features, the interference caused by irrelevant factors on fault type determination can be effectively eliminated. To address this, this article proposes a novel decoupling generative adversarial network for decoupling the intrinsic feature of fault type. The proposed method effectively extracts the intrinsic features of fault types that are unaffected by irrelevant factors, significantly improving fault diagnosis accuracy and robustness, and, to some extent, endowing the fault-type intrinsic features with interpretability and physical rationality.

The existing federated structure protects data privacy with only a certain level of confidentiality, and it is difficult to resist the reconstruction of other clients’ data by malicious participants inside the federated and the illegal manipulation by external attackers or interceptors on the shared information. Besides, the average fusion algorithm used in the cloud center is difficult to eliminate the negative impact of outliers on model updates, and it cannot handle and fuse the time delay or even packet loss that occurs in the information obtained from each local client promptly. Therefore, to make the federated learning (FL) mechanism with stronger privacy protection ability and security, while effectively avoiding the negative impact of outliers on the aggregation of model parameters. We innovatively establish multi-Level FL based on cloud-edge-client collaboration and outlier-tolerance for fault diagnosis. At first, we build a multi-level FL network framework based on the cloud-edge-client collaborative approach for restricted sharing of network parameters level by level without data communication. Then, the edge-side performs Euclidean metrics on the restricted shared model parameters uploaded to the primary edge by each client, and uses them to identify outliers to evaluate and weight them for outlier-tolerance; Then, an outlier-tolerance mechanism is designed based on a centralized Kalman filtering algorithm that is to adjust the modeling error weights adaptively; Lastly, the cloud center performs asynchronous aggregation on the model parameters uploaded asynchronously by the highest-level edge based on a sequential Kalman filtering algorithm and transmitted the optimal model parameters back along the original path. Finally, the effectiveness of the proposed method is verified on the collected dataset.

As an emerging artificial intelligence technology, federated learning plays a significant role in privacy preservation in machine learning, although its main objective is to prevent peers from peeping data. However, attackers from the outside can steal metadata in transit and through data reconstruction or other techniques to obtain the original data, which poses a great threat to the security of the federated learning system. In this paper, we propose a differential privacy strategy including encryption and decryption methods based on local features of non-Gaussian noise, which aggregates the noisy metadata through a sequential Kalman filter in federated learning scenarios to increase the reliability of the federated learning method. We name the local features of non-Gaussian noise as the non-Gaussian noise fragments. Compared with the traditional methods, the proposed method shows stronger security performance for two reasons. Firstly, non-Gaussian noise fragments contain more complex statistics, making them more difficult for attackers to identify. Secondly, in order to obtain accurate statistical features, attackers must aggregate all of the noise fragments, which is very difficult due to the increasing number of clients. We conduct experiments that demonstrate that the proposed method can greatly enhanced the system’s security.