The Stacking-RF model is proposed to improve the low accuracy of a single classifier in bearing fault diagnosis tasks by using a stacking ensemble learning strategy. The model utilizes KNN, LR, SVM, BP, and RF as base classifiers, with RF employed as meta classifiers. Subsequently, we evaluate the performance between the single and stacking models, while investigating the most effective stacking combination to identify various failure modes of rolling bearings accurately. The experimental results of the XJTU-SY bearing dataset show that the diagnosis accuracy of stacking models (98.10%-99.55%) is significantly improved, compared with each member classifier (94.60%-98.98%). It can be demonstrated that the proposed Stacking-RF model can effectively integrate the valuable information of different classifiers, which ultimately leads to a higher accuracy (99.55%). This study shows that the stacking ensemble learning method has a good application prospect in rolling bearing fault diagnosis.
KEYWORDS: Deep learning, Data modeling, Machine learning, Performance modeling, Neural networks, Education and training, Transformers, Statistical modeling, Signal processing, Random forests
We systematically carried out a comparative study of 12 kinds of tree-based models for the task of rolling bearing fault diagnosis, using the publicly available XJTU-SY bearing dataset as an example. The results show that the ensemble tree models including random forest (RF), extremely randomized trees (ETs), and deep learning tree model (multi-Grained Cascade Forest, i.e. gcForest) are high-precision and strong robust models suiting industrial application of this task, which have better-performing detecting accuracy and stability than conventional machine learning and single tree models (decision tree and extremely randomized tree). gcForest achieves 99.37% test accuracy using only 3% of the training samples, while RF and ETs also exceed 98%, which outperform eXtreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGBM), categorical boosting (CatBoost), and neural tree models, i.e. Neural Network with Random Forest (NNRF) and TabNet. RF and ETs are better suited for real-time industrial detection tasks in terms of time consumption. This study provides a scientific basis for the rational selection of rolling bearing fault diagnosis methods.
Continuously improving the accuracy is a hot topic in hyperspectral image (HSI) classification with small-scale samples, due to the high label noise of traditional labeling systems and the high cost of expert labeling systems. We focus on constructing a smaller and more informative training sample set, so an iterative sample selection method guided by uncertainty measurement (ISS-Un) is proposed. The method learns shallow and deep features in the spectral and spatial domains via a convolutional neural network (CNN), where an uncertainty measurement algorithm such as least confidence (LC), marginal sampling (MS) or entropy (Ent) is used to iteratively select high-quality samples for the training set. In addition, we propose a more efficient uncertainty measurement algorithm named margin-entropy fusion (MEF) algorithm to integrate multiple-criteria information. The proposed method is compared with the conventional random sampling method. Experimental results on three HSI datasets show that the proposed ISS-Un method can significantly alleviate the redundancy of training samples and form a more compact and efficient training set, thus improving the classification performance of pixel-oriented HSI. Meanwhile, training sets constructed based on different uncertainty measurement algorithms are applied to five popular CNN models to verify the quality and generalizability of the selected samples. The results show that these training sets work better than random training sampling. Moreover, the proposed MEF algorithm outperforms the LC, MS, and Ent algorithms in selecting samples and is the main recommended scheme.
Volcanic rock formations, as an important oil and gas resource reservoir, have received the focus of the energy industry in recent years. Shear wave logging is essential geophysical data for the exploration and evaluation of volcanic rock oil and gas reservoirs. Due to the strong nonlinear relationship between reservoir logging parameters and S-wave velocity, the conventional point-to-point machine learning methods can not effectively construct the feature space. Deep learning adds neighborhood information to learn the depth features relationship, and builds the mapping of S-wave velocity and wireline logs with its powerful nonlinear solving capability, achieves S-wave velocity prediction. Taking the volcanic reservoir in Xujiaweizi area of Songliao Basin in Northeast China as an example, thirteen logging parameters sensitive to S-wave velocity are selected, and the S-wave velocity prediction models are based on deep learning methods (represented by CNN, ViT, and MLP-Mixer) are proposed. The research demonstrates that the proposed deep learning models are able to predict S-wave velocity with more precision, and the modeling method can give great significance for the exploration of the volcanic reservoir.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.