Low-rank adaptation (LoRA) and its variants have recently gained much interest due to their ability to avoid excessive inference costs. However, LoRA still encounters the following challenges: (1) Limitation of low-rank assumption; and (2) Its initialization method may be suboptimal. To this end, we propose PMSS(Pre-trained Matrices Skeleton Selection), which enables high-rank updates with low costs while leveraging semantic and linguistic information inherent in pre-trained weight. It achieves this by selecting skeletons from the pre-trained weight matrix and only learning a small matrix instead. Experiments demonstrate that PMSS outperforms LoRA and other fine-tuning methods across tasks with much less trainable parameters. We demonstrate its effectiveness, especially in handling complex tasks such as DROP benchmark(+3.4%/+5.9% on LLaMA2-7B/13B) and math reasoning (+12.89%/+5.61%/+3.11% on LLaMA2-7B, Mistral-7B and Gemma-7B of GSM8K).The code and model will be released soon.

The federated fault diagnosis approach has achieved remarkable results in recent years, which enables multiple clients with similar mechanical devices to collaboratively construct global intelligent diagnostic models while protecting data privacy. However, in practice, the statistical heterogeneity of data collected from different clients, as well as the model heterogeneity due to local model personalization, pose great challenges to federated learning (FL). Meanwhile, using a central server as an information management center to build global models increases additional model parameters and the risk of data privacy leakage. To address these issues, this article proposes a heterogeneous FL framework based on peer-to-peer communication (P2PCHF) for rotating machinery fault diagnosis. To achieve heterogeneous client communication without relying on a central server, the sharing unlabeled dataset is utilized in the collaborative updating phase to achieve peer-to-peer communication between clients and to align instance dimensions and clustering dimensions between heterogeneous clients by constructing intercorrelation matrices to achieve feature-level and semantic-level knowledge exchange for better interdomain generalization capabilities. Joint knowledge distillation based on class labels and class relations is introduced in the local update phase to mitigate forgetting effect in the local update phase of private models and effectively balance multidomain category knowledge. It is verified in three cases that the proposed P2PCHF can effectively address model heterogeneity and data statistics heterogeneity among clients, and enable locally-privatized models to gain interdomain generalization capability. The code framework is available athttps://github.com/JC952/P2PCHF.

The construction of a digital twin model for the full life cycle of rolling bearings is of great significance for analyzing their degradation performance and health management. However, existing researches primarily concentrate on the degradation of the outer ring of bearings. The cage, as an important component of bearings, lacks extensive research. Therefore, this article proposes a digital twin assisted assessment method for the degradation of bearing cages. First, a dynamic model including bearing cage fracture is established to generate simulation degradation signals. Second, the simulation signal is modified based on the squeeze and excitation cycle generative adversarial network (SECycleGAN) to minimize the characteristic distribution differences between the simulation and real signals. Finally, the corrected high-fidelity signal is used to train the proposed selective kernel transformer (SKformer) model to assess the degradation stage of the bearing cage. This model can simultaneously capture the long-range temporal correlation features and local mutation multiscale features of the input signals, thus improving the model's recognition ability and generalization performance. The effectiveness of the proposed method is demonstrated through signals collected on real and open-source bearing cage degradation test rigs. The results indicate that the proposed method can produce high-fidelity bearing cage degradation signals and achieve better classification accuracy with limited data.

Single-sided Magnetic Particle Imaging (MPI) devices enable easy imaging of areas outside the MPI device, allowing objects of any size to be imaged and improving clinical applicability. However, current single-sided MPI devices face challenges in generating high-gradient selection fields and experience a decrease in gradient strength with increasing detection depth, which limits the detection depth and resolution. We introduce a novel spatial encoding method. This method combines high-frequency alternating excitation fields with variable offset fields, leveraging the inherent characteristic of single-sided MPI devices where the magnetic field strength attenuates with distance. Consequently, the harmonic signals of particle responses at different spatial positions vary. By manipulating multiple offset fields, we correlate the nonlinear harmonic responses of magnetic particles with spatial position data. In this work, we employed an image reconstruction using a system matrix approach, which takes into account the spatial distribution of the magnetic field during the movement of the device within the field of view. Our proposed encoding approach eliminates the need for the classical selection field and directly links the spatial resolution to the strength and spatial distribution of the magnetic field, thus reducing the dependency of resolution on selection field gradients strength. We have demonstrated the feasibility of the proposed method through simulations and phantom measurements.