Search

Anomaly detection in dynamic graphs presents a significant challenge due to the temporal evolution of graph structures and attributes. The conventional approaches that tackle this problem typically employ an unsupervised learning framework, capturing normality patterns with exclusive normal data during training and identifying deviations as anomalies during testing. However, these methods face critical drawbacks: they either only depend on proxy tasks for general representation without directly pinpointing normal patterns, or they neglect to differentiate between spatial and temporal normality patterns, leading to diminished efficacy in anomaly detection. To address these challenges, we introduce a novel Spatial-Temporal memories-enhanced graph autoencoder (STRIPE). Initially, STRIPE employs Graph Neural Networks (GNNs) and gated temporal convolution layers to extract spatial features and temporal features, respectively. Then STRIPE incorporates separate spatial and temporal memory networks, which capture and store prototypes of normal patterns, thereby preserving the uniqueness of spatial and temporal normality. After that, through a mutual attention mechanism, these stored patterns are then retrieved and integrated with encoded graph embeddings. Finally, the integrated features are fed into the decoder to reconstruct the graph streams which serve as the proxy task for anomaly detection. This comprehensive approach not only minimizes reconstruction errors but also refines the model by emphasizing the compactness and distinctiveness of the embeddings in relation to the nearest memory prototypes. Through extensive testing, STRIPE has demonstrated a superior capability to discern anomalies by effectively leveraging the distinct spatial and temporal dynamics of dynamic graphs, significantly outperforming existing methodologies, with an average improvement of 15.39% on AUC values.

Knowledge graph (KG) embedding methods are at the basis of many KG-based data mining tasks, such as link prediction and node clustering. However, graphs may contain confidential information about people or organizations, which may be leaked via embeddings. Research recently studied how to apply differential privacy to a number of graphs (and KG) analyses, but embedding methods have not been considered so far. This study moves a step towards filling such a gap, by proposing the Differential Private Knowledge Graph Embedding (DPKGE) framework. DPKGE extends existing KG embedding methods (e.g., TransE, TransM, RESCAL, and DistMult) and processes KGs containing both confidential and unrestricted statements. The resulting embeddings protect the presence of any of the former statements in the embedding space using differential privacy. Our experiments identify the cases where DPKGE produces useful embeddings, by analyzing the training process and tasks executed on top of the resulting embeddings.

Varieties of structural damage identification methods have been proposed in the past years. Most of the methods adopt modal information including modal frequencies and mode shapes. These methods provide excellent references for researchers and engineers in this field. However, there is still an unsolved critical problem regarding the number of modes that should be used in the identification. To circumvent this problem, a two-step method for determination of mode order is proposed. Modal effective mass analysis is used to estimate the range of mode orders in the first step. Stabilization diagram analysis is then employed in the second step to decide the reasonable mode order. In the numerical simulation, the proposed method is used to determine the mode order in the damage localization of a truss structure. Results demonstrate that all of the damaged elements are correctly identified when adopting reasonable mode order determined by the proposed method.

Finite element model updating is an effective way to build accurate analytical models for structures. Most of the available updating methods employ information from modal testing. However, in astronautics engineering, information provided by vibration table testing is more valuable than those from modal testing. Therefore, it is necessary to study updating methods which can adopt information from vibration table testing. This paper presents the study on such issue. The base excitation response function is analyzed with the assumption that the vibration table gives the structure a single direction motion excitation. Model updating method which adopts the response function is then proposed. In the numerical simulation, several case studies are constructed for a truss structure with small or significant modeling errors respectively. Data selection, which has great influence on the success of updating, is carefully studied. A novel adaptive data selection approach is suggested. Simulation results show that model updating converge with good accuracy when the adaptive data selection approach is used.

Finite element model updating is an important research field in structural dynamics. Though a variety of updating methods have been proposed in the past decades, all the methods could be effective only on the assumption that the initial finite element model is updatable. The assumption has led to the fact that many researchers study on how to update the model while little attention is paid to studies on whether the model is updatable. This has become inevitable obstacle between research and engineering applications because the assumption is not a tenable hypothesis in practice. To circumvent this problem, the evaluation of model updatability is studied in this paper. Firstly, two conditional statements about mapping are proved as a theoretical basis. Then, two criteria for evaluation of initial models are deduced. A beam is employed in the numerical simulations. Two different initial models for the beam are constructed with different boundary conditions. The models are evaluated using the proposed criteria. The results indicate that the criteria are able to distinguish the model updatability.

Runyang suspension bridge is the longest suspension bridge in China with a main span of 1490 meters. During the construction of the bridge, a structural health monitoring system was installed, which was designed by the Southeast University. Since the bridge was open to traffic, quantities of structural ambient vibration responses have been recorded by the monitoring system. It is important to extract dynamic characteristics from these responses for structural assessment and maintenance. This paper presents the study on extraction of modal parameters from the Runyang suspension bridge structural health monitoring system records using wavelet analysis. Time-frequency domain modal identification using wavelet analysis is studied with an emphasis on the efficient approach for determination of the dilatation parameter. Then the wavelet analysis based method was adopted to identify dynamic properties including modal frequencies, mode shapes and damping ratios from the ambient vibration responses recorded by the monitoring system. Identified results were compared with those from Enhanced Frequency Domain Decomposition method and Stochastic Subspace Identification method. The differences between results are analyzed. Suggestions on future study are also given.