When Brillouin distributed sensing systems are applied to concrete crack monitoring, weak Brillouin gain spectra (BGS) changes caused by microcracks are easily submerged by various factors, such as concrete deformation, residual stress, and system noise, leading to low detection accuracy and poor real-time performance. Existing demodulation algorithms struggle to distinguish the slight changes in the BGS caused by microcracks. To address this limitation, we propose a novel demodulation algorithm integrating baseline differential preprocessing with frequency domain sliding window accumulation. The baseline differencing eliminates global systematic interferences and highlights crack-induced spectral changes, while the frequency domain sliding window accumulation suppresses both coherent fading noise and random noise based on their zero-mean statistical characteristics. Experimental results show that the proposed algorithm improves the signal-to-noise ratio (SNR) by approximately 10 dB, reduces the Brillouin frequency shift (BFS) uncertainty from ±4 MHz to ±1 MHz, and reduces the required data acquisition averages from 5000 to 500. To verify algorithm performance, we cast C30 concrete beams and conducted bending tests. Results show the algorithm can accurately identify millimeter-level cracks. With simple structure and high computational efficiency, this algorithm provides a practical solution for high-precision real-time crack monitoring of bridges, tunnels, and other infrastructures.
Accurate compensation of magnetic interference generated by moving platforms is essential for reliable geomagnetic navigation. Conventional Tolles–Lawson (T‑L) methods require dedicated high‑altitude maneuvers under a clean magnetic background and bandpass filtering, which introduce systematic errors and are impractical for platforms lacking maneuverability. To address these limitations, this paper proposes a hierarchical compensation framework that synergistically combines physical model calibration with data‑driven refinement. First, a Helmholtz coil generates a known true field, enabling direct extraction of platform interference without filtering. An Adaptive Transient‑Aliasing‑Outlier (ATAO) suppression algorithm automatically removes coil transient effects, performs anti‑aliasing decimation, and eliminates outliers. A Truth‑Guided Ridge Regression (TGRR) estimator exploits the known true field to select the optimal regularization parameter deterministically, yielding unbiased and stable T‑L parameters. The remaining nonlinear residuals—which the linear T‑L model cannot capture—are then compensated by a deep neural network with wavelet decomposition and attention mechanism. Experimental results on an unmanned ground vehicle demonstrate that coil‑based calibration improves parameter accuracy by approximately 48.0% compared to conventional methods. The complete hierarchical method achieves final compensation errors of 16.55 ± 1.69 nT (Route 1, training), and 20.24 ± 1.12 nT, 3.62 ± 0.29 nT, and 40.13 ± 3.13 nT (Routes 2–4, testing), respectively. These results verify the performance and generalization ability of the proposed method.
Cuffless blood pressure (BP) estimation from photoplethysmography (PPG) signals using deep learning has attracted significant attention for continuous, non-invasive health monitoring. However, traditional machine learning relies on handcrafted features that capture only limited aspects of the signal. While deep learning can learn richer representations from data, it often requires large-scale training to achieve stable generalization across subjects and conditions and produces black-box models with limited interpretability, hindering clinical adoption in safety-critical applications. To address these issues, we propose a framework that integrates subject-specific topological priors derived from persistent homology of PPG morphology. We construct a six-dimensional topological prior vector (TPV) that summarizes per-beat global morphology and inject it into a dual-branch CNN via TPV-conditioned temporal gating, emphasizing waveform segments that are informative for BP. We evaluate our model on the MIMIC-III PPG dataset. Even when trained from scratch using only a small number of subject-specific segments, it achieves mean absolute errors of 3.82 ± 0.53 mmHg for systolic BP and 2.63 ± 0.41 mmHg for diastolic BP, demonstrating strong performance under limited-data conditions. Our approach embeds morphological priors tailored to each subject. This design significantly improves both data efficiency and personalization, enabling more practical BP estimation in real-world scenarios.
To address performance degradation caused by limited fault samples in PMSMs, a few-shot fault diagnosis method based on multi-source information fusion and irrelevant information suppression is proposed. Initially, one-dimensional (1D) signals from current and magnetic field sensors are reconstructed into 2D images and fused via a multi-scale residual pyramid attention mechanism to enrich feature representations. Furthermore, an Irrelevant Information Suppression Matching Network (IISMN) is developed to improve the traditional matching network. The IISMN utilizes a dual-branch structure to analyze local variations and global trends of fault features, employing a self-subtraction mechanism to eliminate irrelevant information and extract highly fault-correlated generalized features for robust diagnosis. Experimental results demonstrate that the proposed method significantly outperforms baseline models and effectively adapts to cross-condition scenarios, including steady-state, variable-speed, and variable-load operations, providing a highly reliable solution for PMSM fault diagnosis.
To meet the demand for safe oil and gas exploitation, this paper designs and numerically analyzes a three-side polished coreless fiber (CF) sensor based on the surface plasmon resonance (SPR) principle, which can realize the simultaneous detection of methane (CH4), hydrogen (H2) and hydrogen sulfide (H2S). The sensor uses a gold film as the plasmon excitation layer, and adopts zinc oxide (ZnO), ultraviolet-cured fluorosiloxane complex (UVCFS/Crypt-A), and palladium-tungsten trioxide (Pd-WO3) as the specific sensitive materials for the three target gases, respectively. The three-side polished structure effectively shortens the optical field coupling distance significantly improves the sensing sensitivity. Numerical analysis results show that, in the concentration ranges of 0%–2% for CH4, 0%–2% for H2, and 0–80 ppm for H2S, the average sensitivities of the sensor to the three gases are −183 nm/%, −245 nm/% and 16.1 nm/ppm, respectively. This sensor is the first to achieve comprehensive simultaneous detection of three types of hazardous gases in oil and gas fields, featuring zero inter-gas crosstalk. It effectively eliminates temporal and spatial detection deviations, significantly enhances detection accuracy, and is applicable to complex downhole operating environments characterized by strong multi-component coupling.
