Parisa Kavianpour

This paper introduces a hybrid deep learning model combining a Convolutional Neural Network (CNN) with a Bidirectional Long Short-Term Memory (BiLSTM) network, enhanced by an attention mechanism, for earthquake prediction. The model leverages the CNN to extract spatial features from seismic data and the BiLSTM to capture temporal dependencies, while the attention mechanism focuses on the most informative parts of the time series. Experimental results show superior performance in predicting seismic events compared to other baseline models.

Bearing fault diagnosis under varying working conditions faces challenges, including lack of labeled data, distribution discrepancies, and resource constraints. To address these issues, we propose a progressive knowledge distillation framework that transfers knowledge from a complex teacher model, utilizing a Graph Convolutional Network (GCN) with Autoregressive moving average (ARMA) filters, to a compact and efficient student model. To mitigate distribution discrepancies and labeling uncertainty, we introduce Enhanced Local Maximum Mean Square Discrepancy (ELMMSD), which leverages mean and variance statistics in the Reproducing Kernel Hilbert Space (RKHS) and incorporates a priori probability distributions between labels. This approach increases the distance between clustering centers, bridges subdomain gaps, and enhances subdomain alignment reliability. Experimental results on benchmark datasets (CWRU and JNU) demonstrate that the proposed method achieves superior diagnostic accuracy while significantly reducing computational costs. Comprehensive ablation studies validate the effectiveness of each component, highlighting the robustness and adaptability of the approach across diverse working conditions.

Real-world bearing fault diagnosis faces challenges due to lack of access to labeled data and missing or incomplete samples for certain faulty classes, requiring generated data and methods to mitigate imbalanced class distributions and partial-set problems. To address these challenges, we propose a novel physics-informed deep learning framework using a physics-informed synthetic data generation method with an augmented weighted hybrid domain adaptation approach that combines adversarial training with a novel Multi-Kernel Maximum Mean Squared Discrepancy (MK-MMSD) metric. Unlike traditional Domain adaptation method, MK MMSD considers both the mean and variance of data distributions to effectively bridge the domain gap between synthetic and real unlabeled data. Furthermore, we propose a self-adaptive multi-weight block to address the partial-set problem and a novel re-balanced feature Mixup (RF-Mixup) technique to mitigate imbalanced class distributions. Experimental results on the CWRU and JNU datasets demonstrate the effectiveness of our approach in overcoming these real-world challenges in bearing fault diagnosis.

This conference paper presents a deep learning model for classifying earthquake magnitudes. We utilize multi-scale dilated convolutions to capture features at different temporal resolutions from seismic waveforms. An attention mechanism is incorporated to weigh the importance of different time steps, leading to improved classification accuracy.