With the development of artificial intelligence technology and the popularity of intelligent production projects, intelligent inspection systems have gradually become a hot topic in the industrial field. As a fundamental problem in the field of computer vision, how to achieve object detection in the industry while taking into account the accuracy and real-time detection is an important challenge in the development of intelligent detection systems. The detection of defects on steel surfaces is an important application of object detection in the industry. Correct and fast detection of surface defects can greatly improve productivity and product quality. To this end, this paper introduces the MSFT-YOLO model, which is improved based on the one-stage detector. The MSFT-YOLO model is proposed for the industrial scenario in which the image background interference is great, the defect category is easily confused, the defect scale changes a great deal, and the detection results of small defects are poor. By adding the TRANS module, which is designed based on Transformer, to the backbone and detection headers, the features can be combined with global information. The fusion of features at different scales by combining multi-scale feature fusion structures enhances the dynamic adjustment of the detector to objects at different scales. To further improve the performance of MSFT-YOLO, we also introduce plenty of effective strategies, such as data augmentation and multi-step training methods. The test results on the NEU-DET dataset show that MSPF-YOLO can achieve real-time detection, and the average detection accuracy of MSFT-YOLO is 75.2, improving about 7% compared to the baseline model (YOLOv5) and 18% compared to Faster R-CNN, which is advantageous and inspiring.
Feature selection and dimensionality reduction are important for the performance of wind turbine condition monitoring models using supervisory control and data acquisition (SCADA) data. In this paper, an improved random forest algorithm, namely Feature Simplification Random Forest (FS_RF), is proposed, which is capable of identifying features closely correlated with wind turbine working conditions. The Euclidian distances are employed to distinguish the weight of the same feature among different samples, and its importance is measured by means of the random forest algorithm. The selected features are finally verified by a two-layer gated recurrent unit (GRU) neural network facilitating condition monitoring. The experimental results demonstrate the capacity and effectiveness of the proposed method for wind turbine condition monitoring.
Tool wear is a key factor in the machining process, which affects the tool life and quality of the machined work piece. Therefore, it is crucial to monitor and diagnose the tool condition. An improved CaAt-ResNet-1d model for multi-sensor tool wear diagnosis was proposed. The ResNet18 structure based on a one-dimensional convolutional neural network is adopted to make the basic model architecture. The one-dimensional convolutional neural network is more suitable for feature extraction of time series data. Add the channel attention mechanism of CaAt1 to the residual network block and the channel attention mechanism of CaAt5 automatically learns the features of different channels. The proposed method is validated on the PHM2010 dataset. Validation results show that CaAt-ResNet-1d can reach 89.27% accuracy, improving by about 7% compared to Gated-Transformer and 3% compared to Resnet18. The experimental results demonstrate the capacity and effectiveness of the proposed method for tool wear monitor.
8 June 2024 in Mechanical Systems and Signal Processing
One of the methods for detecting transmission line damage is image defect detection of transmission line insulators. The industrialization of image defect detection has made the construction of detection algorithm models with quick detection times, high accuracy, and compact models into a fundamental issue. In order to efficiently address the issues of accuracy and speed of insulator defect detection, this work offers an enhanced version of the INS-YOLOX model based on a one-stage detector. The history of the insulator defect is complicated, as are its potential targets. The target is tiny, and there is little distinction between the images of a faulty insulator and a lossless insulator. The first step is the proposal of a new loss function, L-SIoU, which not only considers the vector Angle between necessary regressions, redefines the penalty index, but also considers the influence of the aspect ratio between the prediction box and regression box, which increases reasoning accuracy and speed. The detection precision of the imbalanced data set was then increased by switching from BCE Loss to Varifocal Loss. In order to improve the feature point extraction, the CBAM attention mechanism was lastly integrated to the Decoupled Head structure. Additionally, INS-YOLOX adds the SGD optimizer and successfully optimizes the model's training parameters. A total of 6227 photos were gathered as a data set to evaluate the model's validity. The test results demonstrate the real-time detection capabilities of INS-YOLOX. The final map, at 83.2%, is 11.1% higher than the benchmark model of YOLOX-tiny and 6% higher than the most recent Yolov7. This represents a major accomplishment.
In the training of the neural network model, the large number of features in the data set will lead to the complexity of the network model and high time cost. Therefore, the feature selection operation of the original data set is carried out to select the feature subset conducive to model training to improve the model's performance. The traditional feature selection algorithm has the problems of a thin process and needs help to eliminate the features with small discrimination. Therefore, this paper proposes the discrimination filtering formula and DI-CFS feature selection algorithm. The discrimination filtering formula can filter out invalid features and inefficient features with small discrimination. DI-CFS algorithm combines discrimination filtering formula, isolation forest algorithm, and improved CFS algorithm. On a set of wind turbine data, the DI-CFS algorithm and other traditional feature selection algorithms are used to select features from the data, respectively, and the obtained feature subsets are input into the same neural network model for training and performing classification tasks. The experimental results show that the discrimination filtering formula has a positive effect, and the DI-CFS algorithm has a better feature selection effect.