Search

In the realm of rotary machine maintenance, rolling bearings emerge as crucial yet frequently vulnerable components. Ensuring their operational integrity is pivotal for the overall safety and performance of the machinery. Traditional diagnostic methods have utilized a variety of signal processing techniques, ranging from blind source separation to wavelet analysis, and more recently, deep neural networks (DNNs). However, a significant challenge in this domain is the scarcity of labeled failure data in real-world applications, leading to a reliance on data augmentation (DA) techniques to enhance the quality and quantity of the training data. Compounding this problem is the use of DA techniques tailored for images, which are not suitable for the time–frequency representations. This study proposes an innovative Data Augmentation with Analytic Wavelets (DAAW) approach, tailored specifically for time-series (TS) data classification in bearing fault diagnosis. The essence of DAAW lies in its ability to generate synthetic scalogram samples that closely mirror the properties of original samples. The proposed DAAW is directly applied at the input stage of a learning model by successively adjusting predefined parameters to generate scalograms, without the need for auxiliary DA algorithms. This technique is versatile in generating synthetic time–frequency data samples, and the parameterization enables it to be tailored to different time series datasets. Through experimentation with publicly available datasets, we show that combining a simple Convolutional Neural Network (CNN) with the proposed approach results in improved performance. It outperforms existing methods, achieving a mean classification accuracy of 99.49%. The critical distance for average accuracy, as determined by the Nemenyi post-hoc test, is 0.4619. Our results also show enhanced performance with accuracy rates of 90.14%, 78.94%, and 62.86% even under low signal-to-noise ratio (SNR) conditions, using 80%, 60%, and 50% of the dataset, respectively, for model training. Notably, its efficacy remains robust even under challenging low SNR conditions. The proposed DAAW significantly improves classification accuracy in bearing fault diagnosis, simultaneously obviating the need for proxy DA tasks. • Time-series data augmentation enhances fault diagnosis using deep learning. • Data augmentation with analytics wavelets to enhance fault prediction accuracy. • Time-series data augmentation through time–frequency representation. • Generating scalograms by adjusting parameters of generalized morse wavelets. [ABSTRACT FROM AUTHOR]

The thermal error stands as a pivotal determinant, significantly impacting both the precision and its stability of gear hobbing processes. Thus far, conspicuous delays in compensation and insufficiencies in precision have been duly noted. To achieve a proficient and exceedingly precise thermal error compensation, an intricate mathematical correlation is expounded, delineating the connection between the spatial pose error of the hob and the inherent thermal error in the gear hobbing machine. In addressing the thermal error linked to the B-axis, a bidirectional-minimal gated unit network is harnessed to characterize long-term memory attributes. An advanced crystal structure algorithm is introduced, encompassing components such as the population initialization mechanism, the nonlinear convergence factor, the Gaussian perturbation mechanism, and an enhanced optimization equation. The weights of the bidirectional-minimal gated unit network are judiciously assigned via the self-attention mechanism. This innovation, recognized as the improved crystal structure algorithm-self-attention-bidirectional-minimal gated unit network, is introduced for the training of the thermal error model for the B-axis. Subsequently, thermal analysis models are formulated, yielding the thermal errors of the X- and Z- axes. An error control model and an error decoupling model are established, employing the thermal errors of the B-, X-, and Z- axes as their input. Ultimately, an intelligent error control system is crafted, in which the error control model, error decoupling model, and thermal error model are seamlessly integrated. The aim is to mitigate tooth flank errors. With the integration of the intelligent error control system, the levels of precision pertaining to gear helix deviation, maximum deviation of a single pitch, and cumulative total pitch deviation are elevated by one, one, and two levels, respectively. More significantly, the geometric error reduction of the tooth flank achieved by the meticulously designed intelligent error control system surpasses that of traditional pitch compensation by a substantial margin. Furthermore, the execution efficiency of the intelligent error control system, when operating within the machine-edge-fog-cloud framework, exhibits a remarkable improvement of 24.18%, 16.20%, and 57.84% when compared to the intelligent error control system operating within the machine-edge-cloud, machine-fog-cloud, and machine-cloud frameworks, respectively. [ABSTRACT FROM AUTHOR]

Authorship verification is the task of determining if a candidate is the real author of the provided sequences of texts. Stylometric learning is the study of the writing style by analyzing the textual data through computational approaches for many linguistic tasks such as authorship verification, authorship identification, etc. In recent years, Deep learning has been used as an effective solution to stylometric learning for authorship verification. However, most of the existing studies do not consider the semantic biases captured by the model, and the learned stylometric features include a strong indicator of the topical biases. During the extraction of stylometric features, we found that specific words related to the latent topics impact the performance of machine learning algorithms and, therefore, are not generalizable toward out-of-sample authors. In this paper, we propose a latent topic score dictionary with a Topic-Debiasing Representation Learning Model (TDRLM) for stylometric representation learning on the problem of authorship verification. The model applies position-specific topic scores on a topic-debiasing attention mechanism in order to adjust the tokenized texts based on the topical bias. The experimental results show that our proposed approach achieves the best performance with the highest Area Under Curve (AUC) of 92.47% for the Twitter-Foursquare dataset and 93.11% for the (ICWSM) Twitter dataset, both better than the state-of-the-art stylometric learning models and the up-trained language models. [ABSTRACT FROM AUTHOR]

Although recent developments in deep learning have brought several tasks closer to human performance, there is still a significant gap between human and machine performance in certain image captioning tasks. Image captioning is the process of creating a textual description of an image. Image captioning focuses on recognizing the main regions of an image, their attributes, and their relationships. It aims to generate textual descriptions that are syntactically and semantically correct. For simple image descriptions, deep learning-based techniques perform well in terms of intricacies and constraints. However, it is challenging to construct sentences when faced with complicated scenes with many entities and relationships, such as how to concurrently solve diversity, anchoring, and controllability—a seemingly simple ability for humans. Scene graphs can significantly alleviate this problem by fully mining spatial and semantic information. However, despite these promising findings, they are fragmented and do not form a systematic comparative overview. We provide a comprehensive overview of the available scene-graph-based image captioning methods in this survey. The foundations of these techniques are discussed to examine their performance, strengths, and constraints. Furthermore, we discuss the comparisons of the state-of-the-art methods, datasets, and commonly utilized evaluation measures. Finally, we conclude the survey with an in-depth discussion of the present and future research challenges. This study will assist readers in comprehending how scene graphs can be applied to image captioning. • Identifying the main regions of an image, describing their attributes, and understanding their relationships. • Alleviating inductive bias by fully exploring spatial and semantic heterogeneity of perceptual information. • Using scene graphs to enhance the understanding of spatial and semantic information of images. • Discussing the comparisons of deep learning-based and scene graph-based methods, datasets, and commonly utilized evaluation measures. [ABSTRACT FROM AUTHOR]

This study presents a comprehensive and rigorous process to develop the most appropriate machine learning (ML) model for predicting the shear strength of RC deep beams (RCDBs). The process consists of the crucial stages and state-of-the-art techniques of ML, including the development of ML models, selection of input features using Shapley Additive explanations, optimisation of the training process, assessment of data randomness, comparisons to the conventional practice codes, and development of novel web-based design platform based on the proposed ML model. For this purpose, seven machine learning models, i.e., linear regression, artificial neural networks (ANN), support vector machines, decision trees, ensemble of trees (EoT), extreme gradient boosting (XGBoost), and Gaussian process regression (GPR) were developed to predict the shear strength of RC deep beams based on a database of 518 samples with 15 input features. The four best models (i.e., ANN, EoT, XGBoost, and GPR) were then considered to assess the influence of varying the number of input features on the prediction performance. The results proved that GPR is the most reliable and accurate ML model. In addition, a set of nine optimal input features is proposed for predicting the shear strength of RCDBs. It was observed that randomly dividing the dataset into training and testing sets can significantly impact the predicted results. In some cases, the R2 value dropped to under 0.78, highlighting the importance of carefully considering the methodology for dividing the dataset when conducting machine learning experiments. The shear strength predicted by ML models was then compared with the three most prominent practice codes (i.e., ACI318, EC2, CSA 23.3-04), which indicated ML approach is highly reliable and accurate over conventional methods. In addition, the study used the Monte Carlo method to evaluate the robustness of the machine learning models and developed a user-interface platform to facilitate the practical application of the proposed machine learning model. [ABSTRACT FROM AUTHOR]

• Developing machine learning based control charts (MLCC) for network surveillance. • Monitoring the counts of communications between the nodes of a given network. • Defining novel input features and scaling methods for the proposed MLCC. • Using the ensemble learning idea to improve the detection ability of the MLCC. • Extensive performance comparisons between MLCC and statistical control charts. Network surveillance, i.e., the detection of anomalous behaviour in communications in a network, has become an important issue in recent years. In this field, techniques of statistical process monitoring, especially control charts, have been frequently applied to monitor the probability of existing connections between the members (nodes) of such a network. As another approach, monitoring the number of communications between the nodes is also an important way to identify network anomalies. Previous works have focused on conventional statistical control charts, but there are no machine learning approaches, although the performance of machine learning based control charts has been shown to be excellent in several other applications. This paper aims to develop machine learning based control charts for network surveillance problems in the case of monitoring the number of communications between the nodes. The results of extensive Monte Carlo simulations reveal the superiority of the proposed method over conventional competitors. [ABSTRACT FROM AUTHOR]

Classification tasks are of great importance in machine learning. However, class imbalance is a universal problem that needs to be solved in classification and can greatly affect the performance of machine learning classifiers. Developing from the basic belief rule base (BRB) system, the hierarchical belief rule-based system can integrate expert knowledge and has the potential to alleviate the negative effect of class imbalance. To utilize BRB to solve the imbalanced multi-classification task and avoid the combinational explosion problem, a novel hierarchical BRB structure based on the extreme gradient boosting (XGBoost) feature selection method, abbreviated as HFS-BRB is proposed in this paper in order to deal with any number of classes. In the hierarchical BRB structure, there is one main-BRB in the first level and several sub-BRBs in the second level. The XGBoost technique is used for feature selection in the modelling process of each abovementioned BRB model. The output of the main BRB represents the approximated classification between confusable classes. Then, these samples were transmitted to a certain sub-BRB for binary classification to make a precise prediction. Thus, a multi-classification problem can be transformed into several binary classification problems. The class imbalance is alleviated. To test the effectiveness of the proposed method, seven classical benchmark problems for imbalanced classification and a real asteroid orbit classification were performed. [ABSTRACT FROM AUTHOR]

• A multi-objective online multi-label feature selection method is proposed. • A novel metric based on mutual information was proposed to evaluate features. • A novel algorithm was designed to update a fixed-sized feature pool. • Features are added to the pool or removed from the pool based on their importance. • Our method assigns a high rank to this highly relevant and non-redundant feature. Multi-label classification methods aim at assigning more than one label to each instance. In many real-world classification problems such as image multi-label classification tasks such as cancer detection, and text classification, we faced with thousands of thousand features. The performance of machine learning methods will be reduced while faced with high dimensional problems. To tackle this issue, feature selection methods are introduced to choose a small set of prominent features which best describe the data. Traditional multi-label feature selection methods are required to access to whole feature space, while in online platforms such as Facebook and Twitter, we faced with streams of data added by the users of these platforms over the time. Traditional multi-label feature selection methods are failed while applied on data streams. To solve this issue, online methods are introduced to deal with data streams. Existing streaming multi-label feature selection methods consider the task as a single optimization process while there are several contradictory objectives that need to be optimize simultaneously. To solve this issue, this paper uses a multi-objective search strategy to choose streaming features by using the mutual information and Pareto optimal set theories. There are several objectives such as minimizing the redundancy of features, and maximizing the relevancy of features to a set of labels that are need to be optimized during the feature selection process. Here, we used the Pareto set theory to identify a set of no-dominant solutions which best describe the problem. The proposed method has compared with a set of state-of-the-art online feature selection methods and the obtained results demonstrate the effectiveness of the proposed strategy. [ABSTRACT FROM AUTHOR]

This paper proposes searching for network architectures which achieve similar performance while promoting diversity, in order to facilitate ensembling. We explain prediction performance and diversity of various network sizes and activation functions applied to semantic segmentation of the CityScapes dataset. We show that both performance and diversity can be predicted from neural network architecture using explainable boosting machines. A grid search of 144 models is performed, and many of the models exhibit no significant difference in mean performance within a 95% confidence interval. Notably, we find the best performing models have varied network architecture parameters. The explanations for performance largely agree with the accepted wisdom of the machine learning community, which shows that the method is extracting information of value. We find that diversity between models can be achieved by varying network size. Moreover, homogeneous network sizes generally show positive correlation in predictions, and larger models tend to converge to similar solutions. These explanations provide a better understanding of the effects of network parameters to deep learning practitioners; they could also be used in place of naïve search methods or a model pool to inform growing an ensemble. • Confirm explainable boosting machines predict performance of network architectures. • Show pairwise diversity of neural networks is predictable similar to performance. • Explain network architecture parameter effect on diversity. • Show that diversity can be promoted through architectural differences. • Show that diversity through architecture can maintain underlying model performance. [ABSTRACT FROM AUTHOR]

• WAFs diagnosis model of mine ventilation system based on SVM is constructed. • The influence law of sample attributes on WAFs diagnosis model is obtained. • Provide technical support for realizing intelligent mine ventilation. Rapid diagnosis and mitigation of windage alteration faults (WAFs) in the mine ventilation system is of great significance to create a good working environment and ensure safe production. The complexity of the ventilation system state evolution in time and space, and the high dimensionality of the feature set characterizing the system state leads to an exceptionally complex fault diagnosis. Therefore, this paper proposes to build a WAFs diagnosis model based on support vector machine (SVM). Meanwhile, the generalization performance of the model is significantly affected by the sample attributes. Therefore, the generalization errors of the model under the combination of four factors, i.e., sample dispersion, sample numbers, input features, and feature numbers, are further analyzed. The results show that the prediction effect of the model leads to a law that with the increase of the degree of sample dispersion, the number of samples, and the number of input features, it improves at the beginning and becomes stable in the end. Furthermore, the best feature for fault diagnosis is found as the combination of wind volume and wind pressure features. Finally, the correctness of the conclusion is verified by comparative experiments. This research provides useful guidance and valuable reference for how to build machine learning (ML) models for mine ventilation system fault diagnosis. [ABSTRACT FROM AUTHOR]

Feature selection is an important data pre-processing procedure in the classification task, whose main purpose is to remove negligible/redundant features to reduce the computational cost, while improving the performance of the subsequent machine learning method. However, most feature selection methods can only work well with the complete data, but show poorly in the presence of the missing information, especially with a high missing rate, due probably to the severe irrelevance and redundancy of features. To address this issue, we proposed a novel feature selection method, namely non-negative latent factor incorporated duplicate maximal information coefficient (NLF-DMIC), which improves the effectiveness of feature selection for the classification of incomplete data. The NLF-DMIC method is fulfilled by the following three steps: (1) select category-friendly features by MIC based on "partial sample strategy" roughly; (2) use the NLF model to make an imputation for the missing data in terms of the filtered features; and (3) select features again by an improved maximal information coefficient (i.e., low-redundancy MIC, LMIC) method on the complete dataset. Finally, experiments on an artificially synthetic dataset and 8 real datasets show that the proposed NLF-DMIC method outperforms some state-of-the-art feature selection methods. • A three-step method is proposed for dimensionality reduction of incomplete data. • The initial feature selection improves the effectiveness of imputation. • The improved maximal information coefficient effectively reduces redundancy. • The combination of two filters achieves less computation and higher accuracy. [ABSTRACT FROM AUTHOR]

This study relies on using a Moth–Flame Optimization (MFO) method as a search algorithm and a Decision Tree (DT) as an evaluation algorithm to generate an efficient feature subset for intrusion detection systems (IDS). The target is to find a feature subset using the minimum number of traffic network features that later obtains the maximum performance by the machine algorithms used in the classification task. This depends on enhancing the MFO by adopting new operators besides the embedded spiral operator to balance the exploration and exploitation alleviating the local minima problem. The main contribution of this work is the adoption of the cosine similarity measure to binarize the continuous MFO into a binary problem. Cosine similarity overcomes the limitations of the commonly used sigmoid function that depends on using a threshold value for conversion. However, cosine similarity computes the similarity ratio between the current solution and the optimal solution. The augmented MFO wrapper framework is applied as an IDS to detect anomalous traffic in the network. The proposed method is compared against several well-known state-of-the-art algorithms on three network datasets (KDDCUPP9, NSL-KDD, and UNSW-NB15), using IDSACC, IDSTPR, IDSFPR, IDSF-score, and convergence evaluation measures to assess the performance of the proposed method. The experimental results show the superiority of the proposed cosine similarity method compared to other algorithms with an accuracy of 97.8%, F-score of 99%, TPR of 99.6%, and FPR of 8.1% using only five selected features from the KDDCUPP99 dataset. It achieved the accuracy of 89.7%, TPR of 89.1%, FPR of 2.9%, when four selected features from the NSL-KDD dataset are used. And finally, it achieved an accuracy of 92.4%, TPR of 92.3%, FPR of 3%, and F-score 94.2% when the UNSW-NB15 dataset is used. • A revision of feature selection methods for intrusion detection systems (IDS). • A wrapper approach for IDSs based on moth–flame optimizer (MFO). • Adoption of the cosine similarity measure to binarize the continuous MFO. • IDS framework assessment against six feature selection methods. [ABSTRACT FROM AUTHOR]

The language barrier is one of the practical challenges human being face during communication. To overcome this, researchers are focusing on using machines to translate a source language to a target language using the textual representations of the languages. Thus, machine translation (MT) could achieve a near human-level performance in terms of translation quality for several resource-rich languages. However, machine translation performance is still far from a production-level quality for the low resource languages. This work reports a semi-supervised neural machine translation system to boost the translation quality for an extremely resource constraint language pair, i.e. English–Manipuri. Our proposed approach exploits self-training and back-translation in a combined technique. The quantitative evaluation shows that the system performance improves by +0.9 BLEU score after introducing external noise to the input data. Additionally, a multi-reference test dataset developed in-house is used to evaluate the linguistic diversity of the highly agglutinative and morphologically rich Manipuri language. Experimental result attests that the proposed semi-supervised system outperforms the supervised, the pretrained mBART and existing semi-supervised baselines in terms of automatic score and subjective evaluation parameters by a significant margin up to +4.5 and +1.2 BLEU improvements against the supervised and mBART baselines respectively. • Backtranslation and forward-translation improve the low resource machine translation. • External perturbations to the noisy synthetic data help in converging the model. • Linguistic variations are tackled via the inclusion of multiple test references. • The proposed method is competitive with pre-trained models. [ABSTRACT FROM AUTHOR]

• New Membership Functions for FSVM. • New Method of Estimating Bias Value. • Effect of the Relationship between New Membership Function and Slack Variable. The effect of an outlier in classification is a problem in the Fuzzy Support Vector Machine (FSVM). In this paper, two new methodologies have been implemented to compute fuzzy membership values for the fuzzy data to address this issue. The membership values that arrived through these methods have been inputted in a fuzzy support vector machine. A method has been devised to select the best bias value to bring out the classifier using the bias values obtained through the support vectors. The Geometric Mean (Gm) and consistency in the newly proposed method for selecting bias values are used as the criteria to choose the best bias value. The methods such as average bias value, individual bias value, weighted average bias value, and median bias value are considered for this purpose. The obtained classifier improves the classification accuracy by reducing the fuzziness in data. The relation between the new membership function and the slack variable ξ i has been exposed in terms of how new membership function reduces the chances of ξ i become a basic variable. An experiment has been conducted with seventeen datasets, including fourteen real-world datasets. A comparison of the performance of the proposed Membership function with other membership functions is made. The robustness of the proposed membership functions for outliers is verified. A statistical test has been performed to ensure the significance of the proposed model. [ABSTRACT FROM AUTHOR]

A Machine Learning (ML) system learns from a set of samples called the training set. In some cases, the training set may have learning conflicts that affect the performance of the machine learning system. A learning conflict is produced when two or more samples in a dataset have similar input values but different target values. We propose a method to remove learning conflicts from a dataset in this work. Our method is based on a genetic algorithm tries to keep those samples that free of conflicts and intents to remove those samples with conflicts. Each individual in the genetic algorithm represents a possible dataset. We introduce the concept of retention error in the fitness function, which describes how many samples are kept while removing learning conflicts. Additionally, the fitness function comprises the Mean-Squared Error (MSE) that validates the machine learning performance. The algorithm is designed to keep as many samples as possible while the machine learning system exhibits the highest possible performance. Therefore, the proposal consists in cleaning first the dataset that compares and highlights the individual with the best performance in the Genetic Algorithm (GA), recommending which samples must be included for training and testing. Three different datasets with learning conflicts are used to test the proposed methodology. Besides, one artificial neural network is trained using the datasets with learning conflicts for each dataset. After removing the conflicts, a second artificial neural network is trained using the cleaned datasets. A noticeable reduction in the mean-square error is observed when the neural network is trained using the cleaned dataset. • Learning conflicts that affect machine learning system performance. • Samples in a dataset with similar values but different target. • Retention error kept meaningful samples while removing learning conflicts. • Artificial neural networks trained with learning conflicts datasets. • Generic algorithm improves the machine learning system performance. [ABSTRACT FROM AUTHOR]

Class imbalance is a major issue that degrades the performance of machine learning classifiers in real-world problems. Oversampling methods have been widely used to overcome this issue by generating synthetic data from minority classes. However, conventional oversampling methods often focus only on the minority class and ignore relationships between the minority and majority classes. In this study, we propose an oversampling method called minority oversampling near the borderline with a generative adversarial network (OBGAN). To consider the minority and majority classes, OBGAN employs one independent discriminator for each class. Each discriminator competitively affects the generator to be trained to capture each region of the minority and majority classes. However, the sensitivity of the generator to the discriminator of the minority class is greater than that of the majority class. Hence, the generator learns the minority class with a focus near the borderline. In addition, the architecture and loss function of OBGAN are designed to avoid the mode collapse problem, which commonly occurs in GANs trained on relatively small datasets. Experimental results, involving 21 datasets and 6 benchmark methods, reveal that OBGAN exhibits excellent performance and stability. • OBGAN: A novel minority oversampling method with GAN for class imbalance problems. • OBGAN generates data from minority region, but near the borderline. • A novel GAN architecture with two discriminators and a loss function is proposed. • Experimental results showed the proposed OBGAN outperformed benchmark methods. • Hyperparameters were analyzed to be not too sensitive to achieve good performances. [ABSTRACT FROM AUTHOR]

Satellite images have become more easily accessible due to government programmes and the availability of commercial earth observation satellites. The increased volume of data and the recent advancement in deep learning-based techniques have made possible the development of fast and accurate methods for satellite image segmentation, which aims at automatically giving each pixel in the satellite image a label of the underlying object/ground type. However, despite the abundant supply of raw satellite images, the labelling of these images is a manual task and is therefore time and labour intensive. When a large labelled dataset is absent, resorting to a smaller data source significantly hampers the performance of the machine learning model. Therefore, an approach which mitigates the effects of using small labelled datasets should be investigated. This paper provides a solution to the problem of insufficient data samples for satellite image segmentation. To this end, two methods are developed and compared with the benchmark U-Net model: the use of transfer learning on an Xception architecture, and the use of steerable filters. The first model transfers knowledge from a pre-trained Xception model and predicts with a multi-resolution feature fusion module, specifically designed to recover fine details. The latter further improves the capability of our neural network model in handling data with high rotational variation. Both models are experimentally shown to significantly enhance the segmentation performance despite a small-sized training dataset. • Lower the impact of small labelled datasets on deep-learning-based methods. • Proposed the Xception-based model using transfer learning. • Proposed the novel steerable-filter-based model for handling rotation. • Verified to achieve superior performance on small labelled dataset. [ABSTRACT FROM AUTHOR]

• A novel SC-GAN data augmentation scheme for stable raw vibration signal generation. • An interpretation method to open the black box of the proposed neural network model. • The model can learn frequency domain patterns from time domain vibration signals. • Data augmentation improves the performance of the machine fault diagnosis model. Vibration signal-based methods have been widely utilized in machine fault diagnosis. Usually, a lack of sufficient training data can prevent these methods from achieving satisfactory performance. The generative adversarial network (GAN) is a feasible solution to this problem. However, existing GAN-based methods struggle to stably generate raw vibration signals. To achieve vibration signal generation, a novel sparsity-constrained GAN (SC-GAN) method containing a two-stage training process is developed, which can perform data augmentation for machine fault diagnosis with a simple structure. Autoencoder (AE)-based pretraining and sparsity regularization constraints are implemented in the proposed method. Furthermore, to understand the internal mechanisms of vibration signal generation, we propose a method for analyzing the network's weight matrix to interpret the generation mechanism. In a case study on rolling element bearings, the SC-GAN is verified to be able to generate raw vibration signals under 10 different health conditions with a more stable training process than other models. In a fault diagnosis task, the data augmentation by SC-GAN significantly improves the diagnostic accuracy by 7.44%. An analysis of the well-trained SC-GAN shows that the model captures key frequency components, which provides a credible interpretation for the generation mechanism. Another case study on the gearbox illustrates the good generalization ability of SC-GAN to other machines and more complicated signals. [ABSTRACT FROM AUTHOR]