To address the issue of low diagnostic accuracy caused by distribution differences between the source and target domains in rolling bearing fault diagnosis, this study proposes a method combining balanced distribution adaptation (BDA) and support vector machines (SVMs). The approach utilizes BDA to simultaneously minimize discrepancies in both the marginal and conditional distributions between domains, enabling effective feature alignment and enhancing the model's cross-domain generalization in small-sample scenarios. After extracting time-and frequency-domain features, BDA adaptively adjusts the feature distributions, and SVMs are employed for fault classification. Experimental results demonstrate that the BDA-SVM method achieves over 94 % diagnostic accuracy, showcasing strong performance and robustness in bearing fault diagnosis. Compared with traditional SVMs and other methods without transfer learning, the proposed approach shows a significant improvement in diagnostic accuracy under cross-domain conditions.
With the continuous improvement of intelligent manufacturing and robot perception capabilities, traditional robotic arm grasping methods still face problems such as inaccurate posture prediction and poor task adaptability in dealing with dynamic scenes, complex objects, and functional action execution. Therefore, the research develops an optimized contact grasping network model that integrates scene constraints and task constraints. It combines UR5 six degree of freedom robotic arm, visual input, and Contact-GraspNet architecture based on point cloud, and introduces PointNet++ local feature enhancement mechanism and lightweight encoder design to effectively improve the spatial perception and action planning capabilities of grasping points. Experimental results show that on the GraspNet and YCB Dataset, the model achieves F1 scores of 92.54% and 91.82% respectively, with average execution time reduced to 0.61 seconds. In functional operation scenarios involving door handles, kettle handles, and drawer pulls, grasping accuracy remained above 0.89, with task completion rates significantly outperforming mainstream baseline models. Under visual interference conditions with up to 75% occlusion rate, the average inference latency was controlled within 0.82 seconds. Under varying light intensities, the pose angle error remained within the range of 1.21° to 1.87°. Therefore, this model exhibits comprehensive advantages in grasping precision, latency control, and deployment efficiency, and has the potential to be largely applied in industrial, service, and special task environments.
The alkali recovery boiler is an important piece of equipment in chemical pulp production. It uses papermaking black liquor as fuel. Due to the large number of impurities, the heating surface is prone to severe ash deposition. To improve the heat transfer coefficient, the traditional method is to carry out timed soot blowing, which has problems such as untimely ash cleaning or excessive steam consumption caused by overly frequent soot blowing. In this paper, an experimental device for the relationship between the heat transfer coefficient and the weight of ash deposition was developed. The functional relationship between the two was obtained. There is an inflection point when the amount of ash deposition reaches 294.77 g. A control strategy of soot blowing at the inflection point was further proposed. A load-strain experimental testing device was developed. Experiments show that when the weight of ash deposition reaches 3503 N, the deformation of the suspender is 9 με, and starting the soot blowing device is the most efficient when the heat transfer coefficient is 725.15 W/(m²·K). Therefore, the degree of decline of the heat transfer coefficient can be indirectly reflected by measuring the strain. A method of soot blowing by detecting the degree of suspender strain was further proposed. This method is more scientific than the original timed soot blowing method and provides a technical reference for the energy-saving and consumption-reduction theories and engineering applications of alkali recovery boilers.
This study investigates the impact of incorporating clay nanoplatelets and waste eggshell particles into polymethyl methacrylate (PMMA) to develop hybrid nanocomposites with improved mechanical and structural properties. Waste eggshells, rich in biogenic calcium carbonate, were utilized as a sustainable, low-cost filler, while clay nanoplatelets provided nanoscale reinforcement and improved interfacial bonding. Hybrid nanocomposites were prepared with 5 wt. % of filler contents using the solution casting method. Scanning electron microscopy (SEM) was employed to examine the dispersion and morphology of the fillers within the structure of the PMMA matrix. At the same time, Fourier-transform infrared spectroscopy (FTIR) was utilized to assess the powder chemical interactions and potential bonding between the fillers (clay nanoplatelets and eggshell particles). Mechanical behavior was also evaluated through tensile testing of PMMA and its produced hybrid nanocomposites. The results indicate a synergistic enhancement due to the combination of organic biowaste and inorganic nanoclay, offering a promising route for the development of sustainable, high-performance polymer nanocomposites.
Although radiographic inspection is one of the oldest techniques for non-destructive testing, it is still considered vital in many industrial fields to ensure the quality of welds and meet the demands of work conditions and design, as well as safety and reliability requirements. This paper presents an algorithm that identifies and categorizes welding defects in radiographic images using machine learning techniques. For this aim, two supervised classifiers are proposed and performed, which are 1) K-nearest Neighbour (KNN), which is a nonparametric classifier, and 2) a multiclass classifier based on Support Vector Machine (SVM) as an intensive learning-based classifier (enthusiastically learns). SVM is commonly used in binary classification, but it can be adapted for multi-classification using various common methods, such as one-versus-one and one-versus-all. The texture features are adopted in this paper as inputs to the classifiers, where two groups of them are used: the feature of the Local Binary Pattern (LBP) and the grey-level co-occurrence extraction matrix (GLCM), to obtain the feature vector. To avoid the risk of overfitting, four k-fold cross-validations are applied. The experimental results are reported for two different classifiers, achieving an accuracy of 91.66% when combining GLCM and SVM.
