The random forest (RF) algorithm is a typical representative of ensemble learning, which is widely used in rolling bearing fault diagnosis. In order to solve the problems of slower diagnosis speed and repeated voting of traditional RF algorithm in rolling bearing fault diagnosis under the big data environment, an efficient rolling bearing fault diagnosis method based on Spark and improved random forest (IRF) algorithm is proposed. By eliminating the decision trees with low classification accuracy and those prone to repeated voting in the original RF, an improved RF with faster diagnosis speed and higher classification accuracy is constructed. For the massive rolling bearing vibration data, in order to improve the training speed and diagnosis speed of the rolling bearing fault diagnosis model, the IRF algorithm is parallelized on the Spark platform. First, an original RF model is obtained by training multiple decision trees in parallel. Second, the decision trees with low classification accuracy in the original RF model are filtered. Third, all path information of the reserved decision trees is obtained in parallel. Fourth, a decision tree similarity matrix is constructed in parallel to eliminate the decision trees which are prone to repeated voting. Finally, an IRF model which can diagnose rolling bearing faults quickly and effectively is obtained. A series of experiments are carried out to evaluate the effectiveness of the proposed rolling bearing fault diagnosis method based on Spark and IRF algorithm. The results show that the proposed method can not only achieve good fault diagnosis accuracy, but also have fast model training speed and fault diagnosis speed for large-scale rolling bearing datasets.

To address the problems of low recognition accuracy, slow convergence speed and weak generalization ability of traditional LeNet-5 network used in rolling-element bearing fault diagnosis, a rolling-element bearing fault diagnosis method using improved 2D LeNet-5 network is put forward. The following improvements to the traditional LeNet-5 network are made: the convolution and pooling layers are reasonably designed and the size and number of convolution kernels are carefully adjusted to improve fault classification capability; the batch normalization (BN) is adopted after each convolution layer to improve convergence speed; the dropout operation is performed after each full-connection layer except the last layer to enhance generalization ability. To further improve the efficiency and effectiveness of fault diagnosis, on the basis of improved 2D LeNet-5 network, an end-to-end rolling-element bearing fault diagnosis method based on the improved 1D LeNet-5 network is proposed, which can directly perform 1D convolution and pooling operations on raw vibration signals without any preprocessing. The results show that the improved 2D LeNet-5 network and improved 1D LeNet-5 network achieve a significant performance improvement than traditional LeNet-5 network, the improved 1D LeNet-5 network provides a higher fault diagnosis accuracy with a less training time in most cases, and the improved 2D LeNet-5 network performs better than improved 1D LeNet-5 network under small training samples and strong noise environment.

Industrial equipment failure diagnosis is a crucial issue that impacts the national industrial manufacturing level, economic cycle development, and sustainable technological advancement. A multimodal knowledge graph (MMKG) based intelligent diagnostic model for industrial equipment fault is proposed to address the issues of insufficient and inadequate fault data samples encountered when using a single-mode model for fault diagnosis in existing industrial equipment. This model does not require extensive data learning for equipment fault diagnosis in complex industrial scenarios. The model utilizes an improved Faster Regions with CNN features (Faster RCNN) object detection module to extract visual information feature vectors of semi-ordered main and non-main objects. These feature vectors are then mapped to entity, attribute, and relationship vectors in a knowledge graph using cosine similarity for feature correspondence mapping. The semantic matching inference is performed based on this mapping, resulting in a set of fault triplets. Finally, the Bidirectional and Auto-Regressive Transformers (BART) text generation model processes this triplet set to generate fault diagnosis texts. Experimental results demonstrate that the improved Faster RCNN object detection model achieves a 1.2% increase in confidence when trained with small training data sets. The accuracy of generated fault description texts reaches approximately 98% compared to standard texts. The model presented in this paper addresses the challenge of diagnosing faults in industrial equipment, particularly in complex scenarios with limited data, such as substations. It enhances the target detection model to effectively extract visual features even when data is scarce. Additionally, it utilizes an MMKG to enable interpretable intelligent decision-making.

Sparse matrix–vector multiplication (SpMV) is of singular importance in sparse linear algebra, which is an important issue in scientific computing and engineering practice. Much effort has been put into accelerating SpMV, and a few parallel solutions have been proposed. This paper focuses on a special type of SpMV, namely sparse quasi-diagonal matrix–vector multiplication (SQDMV). The sparse quasi-diagonal matrix is the key to solving many differential equations, and very little research has been done in this field. This paper discusses data structures and algorithms for SQDMV that are efficiently implemented on the compute unified device architecture (CUDA) platform for the fine-grained parallel architecture of the graphics processing unit (GPU). A new diagonal storage format, a hybrid of the diagonal format (DLA) and the compressed sparse row format (CSR) (HDC) will be presented, which overcomes the inefficiency of DLA in storing irregular matrices and the imbalances of CSR in storing non-zero elements. Furthermore, HDC can adjust the storage bandwidth of the diagonal to adapt to different discrete degrees of sparse matrix, so as to get a higher compression ratio than DLA and CSR, and reduce the computational complexity. Our implementation in a GPU shows that the performance of HDC is better than that of other formats, especially for matrices with some discrete points outside the main diagonal. In addition, we combine the different parts of HDC to make a unified kernel to get a better compression ratio and a higher speedup ratio in the GPU.