Meng Yang

In recent years, explainable artificial intelligence has received increasing attention in the field of bearing fault diagnosis. However, existing interpretability methods, such as Shapley Additive Explanations (SHAP), often rely on the quality of input features. To achieve high diagnostic accuracy, researchers often extract a large number of features from vibration signals across multiple domains, leading to feature redundancy. This redundancy not only increases the computational cost and risk of overfitting in diagnostic models but also dilutes the contributions of core features during interpretability analysis, resulting in biased explanations. To address this challenge, we propose a two-stage interpretable fault diagnosis approach. In the first stage, the Explainable Boosting Machine (EBM) selects core features to reduce redundancy. In the second stage, EBM is enhanced by Random Forest (RF) through residual learning to form the RF-EBM diagnostic model. EBM and SHAP are further used for dual interpretability analysis. Experimental results on public laboratory benchmark datasets demonstrate that the proposed approach achieves good diagnostic performance and outperforms traditional EBM. Overall, the approach reduces redundancy through feature selection, improves diagnostic performance, and makes the decision-making process more transparent, providing a useful methodological reference for trustworthy fault diagnosis in industrial applications.

Bearing signal denoising is a pivotal task in predictive maintenance and condition monitoring of industrial machinery. However, conventional denoising methods often face difficulties in simultaneously suppressing noise and preserving essential physical features. To address this challenge, we propose a novel denoising framework that incorporates physical feature priors derived from low-dimensional manifold-based simulated data. Specifically, a regression branch—constructed using convolutional and residual neural networks—is integrated into the main denoising model to exploit the intrinsic structure of bearing signals. By embedding manifold-informed priors, the regression branch enhances denoising performance and ensures the retention of critical physical features. Experimental results demonstrate that the proposed approach surpasses traditional methods in both signal denoising and bearing fault diagnosis. Notably, the incorporation of manifold-derived priors improves the model’s capability to capture the underlying physical characteristics of the signals, indicating that the network has learned their inherent features rather than simply minimizing the loss function. Overall, this study introduces a robust denoising paradigm for complex industrial environments where bearing signals exhibit significant variability.

Transient Electromagnetic (TEM) method is widely used in various geophysical applications, providing valuable insights into subsurface properties. However, time-domain TEM signals are often submerged in various types of noise. While recent deep learning-based denoising models have shown strong performance, these models are mostly trained on simulated or single real-world scenario data, overlooking the significant differences in noise characteristics from different geographical regions. Intuitively, models trained in one environment often struggle to perform well in new settings due to differences in geological conditions, equipment, and external interference, leading to reduced denoising performance. To this end, we propose the Dictionary-driven Prior Regularization Test-time Adaptation (DP-TTA). Our key insight is that TEM signals possess intrinsic physical characteristics, such as exponential decay and smoothness, which remain consistent across different regions regardless of external conditions. These intrinsic characteristics serve as ideal prior knowledge for guiding the TTA strategy, which helps the pre-trained model dynamically adjust parameters by utilizing self-supervised losses, improving denoising performance in new scenarios. To implement this, we customized a network, named DTEMDNet. Specifically, we first use dictionary learning to encode these intrinsic characteristics as a dictionary-driven prior, which is integrated into the model during training. At the testing stage, this prior guides the model to adapt dynamically to new environments by minimizing self-supervised losses derived from the dictionary-driven consistency and the signal one-order variation. Extensive experimental results demonstrate that the proposed method achieves much better performance than existing TEM denoising methods and TTA methods.

Grouting technology is widely used in construction and civil engineering, where evaluating grouting effectiveness is crucial due to the uncertainty of subsurface conditions. Existing methods face drawbacks such as destructiveness, high cost, poor durability, and limited data collection. To address these issues, this paper proposes a novel wireless real-time monitoring system based on a ZigBee sensor network framework. The sensor system integrates a direct current method in geophysics with apparent resistivity measurement to assess grouting effectiveness in real time. It consists of multichannel data acquisition units with electrodes for sensing underground currents and a user control unit for centralized management and data processing. A system acquisition performance test confirmed that the differential input channel’s equivalent input noise of the ADC was only 175 μV and 188 μV, and the average error of the captured sine wave data was 4.51 mV and 4.19 mV, ensuring the voltage measurement accuracy of the data acquisition units. Stability testing of the equipment in road and construction environments showed an average RSD of 2.86% and 2.92%, respectively, indicating good stability of the measurements. ZigBee network performance tests in three simulated environments and a field test showed that the packet loss rate (PLR) was less than 2% from 0 to 50 m, ensuring network communication in grouting project scenarios. On-site experiments demonstrate that the system can simultaneously monitor multiple profiles and perform inversions in the grouting area, which can be assembled into 3D inversion images for evaluating grout diffusion, offering valuable insights for optimizing construction operations, and enhancing grouting efficiency.

Currently, deep learning applications in tomato disease recognition are becoming widespread. However, in greenhouse tomato cultivation environments, many techniques neglect the foggy image caused by wet conditions, adversely affecting the performance of disease recognition models, which still remains unresolved. To alleviate adverse environmental impacts, such as fog, and enhance the precision and robustness of the model, this study introduces an improved dark channel prior (DCP) dehazing algorithm and proposes the TDR-Model based on the MobileNetV3. Employing the quad-tree spatial hierarchy search strategy, guided filtering, tolerance mechansim, and multi-scale color restoration (MSRCR) algorithm, an improved version of DCP dehazing algorithm is realized. Image preprocessing with the improved DCP dehazing algorithm obtains more authentic dehazed images. Moreover, MobileNetV3 is improved by the incorporation of the convolutional block attention module (CBAM) and Omni-Dimensional Dynamic Convolution(ODC), improving the accuracy of disease feature recognition. The experiment results demonstrate that, with the parameter size remaining nearly unchanged (a slight reduction of 0.04M), an increase of 16.88% is achieved in accuracy compared to the original version for foggy images.