The degradation of bearings plays a key role in the failures of industrial machinery. Prognosis of bearings is critical in adopting an optimal maintenance strategy to reduce the overall cost and to avoid unwanted downtime or even casualties by estimating the remaining useful life (RUL) of the bearings. Traditional data-driven approaches of RUL prediction rely heavily on manual feature extraction and selection using human expertise. This paper presents an innovative two-stage automated approach to estimate the RUL of bearings using deep neural networks (DNNs). A denoising autoencoder-based DNN is used to classify the acquired signals of the monitored bearings into different degradation stages. Representative features are extracted directly from the raw signal by training the DNN. Then, regression models based on shallow neural networks are constructed for each health stage. The final RUL result is obtained by smoothing the regression results from different models. The proposed approach has achieved satisfactory prediction performance for a real bearing degradation dataset with different working conditions.
24 September 2018 in Neural Computing and Applications
In unstructured environments, terrain perception is essential for stability and environmental awareness of Quadruped robot locomotion. Existing approaches primarily rely on visual or proprioceptive signals, but their effectiveness is limited under conditions of visual occlusion or ambiguous terrain features. To address this, this study proposes a multimodal terrain perception method that integrates acoustic features with proprioceptive signals. This terrain perception method collects environmental acoustic information through an externally mounted sound sensor, and combines the sound signal with proprioceptive sensor data from IMU and joint encoder of the quadruped robot. The method was deployed on the quadruped robot Lite2 platform developed by Deep Robotics, and experiments were conducted on four representative terrain types: concrete, gravel, sand, and carpet. Mel-spectrogram features are extracted from the acoustic signals and concatenated with the IMU and joint encoder to form feature vectors, which are subsequently fed into a support vector machine for terrain classification. For each terrain type, 400 s of data were collected. Experimental results show that the terrain classification accuracy reaches 78.28% without using acoustic signals, while increasing to 82.52% when acoustic features are incorporated. To further enhance the classification performance, this study performs a combined exploration of the SVM hyperparameters and as well as the time-window length . The final results demonstrate that the classification accuracy can be improved to as high as 99.53% across all four terrains.
Hyperspectral imagery (HSI) provides rich spectral information for land-cover classification, while Light Detection and Ranging (LiDAR) data provide complementary elevation and structural cues. Existing HSI–LiDAR fusion methods can achieve strong performance, but many rely on complex cross-modal interaction modules with substantial computational cost. This paper proposes the HSI-Primary Residual Adaptation with Consistency Regularization Network (SRAC-Net) for lightweight HSI–LiDAR classification. The method treats HSI as the primary spectral–spatial modality and introduces LiDAR features as an adapted residual correction. A learnable channel-wise residual scaling vector controls the contribution of the LiDAR residual in each feature channel. In addition, the HSI-primary branch is explicitly supervised and provides a stop-gradient reference distribution for consistency regularization of the fused prediction. Experiments on Houston2013, MUUFL, and Trento show that SRAC-Net achieves the highest mean OA, AA, and Kappa values among the evaluated internal baselines and selected representative fusion methods under the adopted protocol. The ablation results show that the complete configuration obtains the best mean performance among the evaluated variants. LiDAR perturbation experiments on Houston2013 further show smaller mean OA reductions than direct residual fusion under the tested Gaussian-noise, random-dropout, and block-occlusion settings. The method also maintains a compact parameter scale and low measured inference latency relative to several heavier multimodal architectures. These results suggest that HSI-primary residual adaptation with consistency regularization is an effective lightweight fusion alternative for the evaluated HSI–LiDAR classification settings.
Robotic visual perception and terrain recognition are critical for autonomous locomotion and adaptive control in complex environments. However, existing models often extract weak features, confuse classes, and deliver unstable recognition. Most prior studies use end-to-end convolutional networks or single-stream feature extraction, which limits the balance between fine-grained visual representation and adaptive discrimination of confusing samples. To solve this problem, this paper proposes a vision model that blends attention mechanisms with a confusion augmentation strategy. Using an improved ResNet50 backbone, we add a local feature sharpening module and a channel–spatial attention module to strengthen edge texture and global context representation. We also design a confusion augmentation strategy based on the similarity of hard samples. It generates mixed samples through cross-perturbation in feature space, thereby improving the discrimination of highly similar terrains. Experiments show that our model achieves an accuracy of over 98.19% on various terrains, including cement, asphalt, sand, and snow. t-SNE visualization and Grad-CAM analysis demonstrate clear class separability and good interpretability, confirming the effectiveness and robustness of the approach for robotic terrain recognition.
1 January 2025 in Journal of Mechanical Engineering