Wei E. I. Sha

Jixuan Cui, Jun Li, Zhen Mei et al.

Deep learning-based fault diagnosis (FD) approaches require a large amount of training data, which are difficult to obtain since they are located across different entities. Federated learning (FL) enables multiple clients to collaboratively train a shared model with data privacy guaranteed. However, the domain discrepancy and data scarcity problems among clients deteriorate the performance of the global FL model. To tackle these issues, we propose a novel framework called representation encoding-based federated meta-learning (REFML) for few-shot FD. First, a novel training strategy based on representation encoding and meta-learning is developed. It harnesses the inherent heterogeneity among training clients, effectively transforming it into an advantage for out-of-distribution generalization on unseen working conditions or equipment types. Additionally, an adaptive interpolation method that calculates the optimal combination of local and global models as the initialization of local training is proposed. This helps to further utilize local information to mitigate the negative effects of domain discrepancy. As a result, high diagnostic accuracy can be achieved on unseen working conditions or equipment types with limited training data. Compared with the state-of-the-art methods, such as FedProx, the proposed REFML framework achieves an increase in accuracy by 2.17%-6.50% when tested on unseen working conditions of the same equipment type and 13.44%-18.33% when tested on totally unseen equipment types, respectively.

Qi I. Dai, Jun Wei Wu, Ling Ling Meng et al.

Characteristic mode (CM) analysis poses challenges in computational electromagnetics (CEM) as it calls for efficient solutions of dense generalized eigenvalue problems (GEP). Multilevel fast multipole algorithm (MLFMA) can greatly reduce the computational complexity and memory cost for matrix-vector product operations, which is powerful in iteratively solving large scattering problems. In this article, we demonstrate that MLFMA can be easily incorporated into the implicit restarted Arnoldi (IRA) method for the calculation of CMs, where MLFMA with the sparse approximate inverse (SAI) preconditioning technique is employed to accelerate the construction of Arnoldi vectors. This work paves the way of CM analysis for large-scale and complicated three-dimensional ($3$-D) objects with limited computational resources.

Xiao Z. Wang, Wei E. I. Sha

This report mainly focused on the basic concepts and the recovery methods for the random sampling. The recovery methods involve the orthogonal matching pursuit algorithm and the gradient-based total variation strategy. In particular, a fast and efficient observation matrix filling technique was implemented by the classic Shannon interpolation and Poisson summation formulae. The numerical results for the trigonometric signal, the Gaussian-modulated sinusoidal pulse, and the square wave were demonstrated and discussed. The work may give some help for future work in theoretical study and practical implementation of the random sampling.