The current-based fault diagnosis method is a feasible way to replace the conventional vibration-based method, as it is more economical, implemental, and reliable. With the deep learning (DL) method applied, the current-based methods have achieved satisfactory diagnosis accuracy. DL methods, however, demand large quantities of training samples, which are difficult to implement in real industrial sites. To tackle this problem, this paper proposes a novel lightweight fault diagnosis method based on Convolutional Neural Network (CNN), called CombFilterNet (CF-Net). The first convolutional layer of CF-Net is called CF-layer, where the convolution kernel is the comb filter kernel (CF-kernel). Each CF-kernel only has three parameters to be updated, achieving a lightweight design that makes CF-Net suitable for limited training sample conditions. The effectiveness and generalization ability of the proposed method are validated by a laboratory-acquired current dataset and an open-source vibration dataset. The results demonstrate that the proposed method is superior to the comparative methods under limited training sample conditions.

Cloud-induced atmospheric extinction and occlusion significantly affect the effectiveness and quality of telescope observations. Real-time cloud-cover distribution and long-term statistical data are essential for astronomical siting and telescope operations. Visual inspection is currently the primary approach for analyzing cloud distribution at ground-based astronomical sites. However, the main disadvantages of manual observation methods are human subjectivity, heavy workloads, and poor real-time performance. Therefore, a real-time automatic cloud image classification method is desperately needed. This paper presents a novel cloud identification method named the PSO+XGBoost model, which combines eXtreme Gradient Boosting (XGBoost) with particle-swarm optimization (PSO). The entire cloud image is divided into 37 sub-regions to identify the distribution of the clouds more precisely. Nineteen features, including the sky background, star density, lighting conditions, and subregion grayscale values, are extracted. The experimental results have shown that the overall classification accuracy is 96.91%, and our model can outperform several state-of-the-art baseline methods. Our approach achieves high accuracy in comparison with the manual observation methods. Moreover, this method meets telescope real-time scheduling requirements.

Modal noise is a key error source limiting the precision of radial velocity (RV) measurements in fiber-fed high-resolution spectrographs. Mechanical scrambling is one of the effective methods for mitigating modal noise. However, existing mechanical scramblers lack a clear explanation of their operating mechanism and suffer from limited scrambling efficiency. To address these issues, this paper investigates the physical mechanism of dynamic scrambling, revealing that scrambling efficiency is closely related to the spatial mounting configuration and motion trajectory of the fiber. Based on this theoretical insight, we propose an optimized scrambling scheme utilizing a spatially separated dual-axis orthogonal architecture combined with a non-harmonic driving strategy to achieve chaotic modulation of the internal stress field within the fiber. Experimental results demonstrate that this method significantly enhances the performance of the dynamic scrambler. The method improves the signal-to-noise ratio (SNR) of the multimode fiber output from 4.35 to 60.55 and suppresses the RV error induced by modal noise to 5.3 cm s ⁻¹ , achieving sub-meter-per-second precision. With its advantages of compactness, cost-effectiveness, and high efficiency, this scheme provides a robust solution for modal noise mitigation in high-precision astronomical spectroscopy.

Cloud-induced atmospheric extinction and occlusion have a major effect on the effectiveness and quality of telescope observations. Real-time cloud cover distribution and long-term statistical data are essential for astronomical siting and telescope operations. At ground-based astronomical telescope sites, cloud cover distribution is currently analyzed using manual observation methods. However, the main disadvantages of manual observation methods are human subjective, heavy workloads and poor real-time performance. Therefore, a real-time automatic cloud images classification method is desperately needed. This paper presents a new cloud identification method, which is named PSO+XGBoost model by combining eXtreme Gradient Boosting (XGBoost) and Particle Swarm Optimization (PSO). The entire cloud image is divided into 37 sub-regions in order to more precisely identify the distribution of the clouds. 19 features are then extracted, including the sky background, star density, lighting conditions, and subregion grayscale values. The experimental results have shown that the overall classification accuracy of 96.91% and our model is able to outperform several state-of-the-art baseline methods. Our approach achieves high accuracy in comparison with the manual observation methods. Moreover, this method meets telescope real-time scheduling requirements.

The increasing aperture of next-generation telescopes demands higher-resolution wavefront sensing. The sensing resolution of the Shack-Hartmann wavefront sensor (SHWFS), a key component in adaptive optics, is fundamentally limited by its sub-aperture density. To overcome this limitation, we introduce a frequency-domain filter-based neural network (FF-Net) for super-resolution wavefront reconstruction (SRWR). Inspired by the physical imaging principles of the SHWFS and the convolution theorem, FF-Net utilizes learnable Gabor filters to extract aberration features across various frequency bands. This approach enables more effective feature decoupling from aliased sub-aperture spots compared to standard convolutional kernels, leading to high-precision reconstruction. Numerical simulations demonstrate that FF-Net achieves state-of-the-art SRWR performance. Notably, with an under-sampled SHWFS, our method outperforms the low-order reconstruction accuracy of the conventional method employing four times the sub-aperture density, while simultaneously retrieving higher-order modes. Furthermore, its GPU-accelerated inference time of under 1 ms meets the real-time requirements of most astronomical adaptive optics systems. FF-Net demonstrates that physics-informed network design is a powerful strategy for developing more accurate, robust, and interpretable deep learning solutions in adaptive optics.