Faulty Samples Research Articles

Fault Diagnosis Method Based on CND-SMOTE and BA-SVM Algorithm

The problem of unbalanced data classification has gotten extensive attention in the past few years. Unbalanced sample data makes the fault diagnosis and classification accuracy rate low, and the capability to classify minority-class fault samples is restricted. To address the problem that the classification algorithm in machine learning has the insufficient capability to identify minority class samples for unbalanced sample data classification problems. Therefore, this paper proposes an improved support vector machine (SVM) classification method based on the synthetic minority over-sampling technique (SMOTE). For the sampler, an improved synthetic minority over-sampling technique based on the characteristics of neighborhood distribution (CND-SMOTE) algorithm is used to equilibrate the minority class samples and the majority class samples. For the classifier, the parameter optimization method of support vector machines based on the bat algorithm (BA-SVM) is used to solve the multi-classification problem of faulty samples. Finally, experimental results prove that the CND-SMOTE+BA-SVM algorithm can synthesize high-quality minority fault samples, increase the classification accuracy rate of fault samples, and decrease the time spent on the classification.

A graph neural network-based bearing fault detection method

Bearings are very important components in mechanical equipment, and detecting bearing failures helps ensure healthy operation of mechanical equipment and can prevent catastrophic accidents. Most of the well-established detection methods do not take into account the correlation between signals and are difficult to accurately identify those fault samples that have a low degree of failure. To address this problem, we propose a graph neural network-based bearing fault detection (GNNBFD) method. The method first constructs a graph using the similarity between samples; secondly the constructed graph is fed into a graph neural network (GNN) for feature mapping, and the samples outputted by the GNN network fuse the feature information of their neighbors, which is beneficial to the downstream detection task; then the samples mapped by the GNN network are fed into base detector for fault detection; finally, the results determined by the integrated base detector algorithm are determined, and the top n samples with the highest outlier scores are the faulty samples. The experimental results with five state-of-the-art algorithms on publicly available datasets show that the GNNBFD algorithm improves the AUC by 6.4% compared to the next best algorithm, proving that the GNNBFD algorithm is effective and feasible.

Aero-engine high speed bearing fault diagnosis for data imbalance: A sample enhanced diagnostic method based on pre-training WGAN-GP

Rolling bearing is the key supporting component of aero-engines, of which fault diagnosis is very important to ensure its reliable operation and continuous airworthiness. However, the data imbalance problem caused by its complex and harsh environment restricts the intelligent diagnosis. This paper proposes a sample enhanced diagnostic method based on pre-training and auxiliary classifier Wasserstein generative adversarial network with gradient penalty (PT-WGAN-GP). Firstly, a pre-training network is proposed and incorporated into the discriminator and classifier of WGAN-GP for feature adaptive and efficient extraction. Meanwhile, a new generator is constructed by introducing a residual network and the instance batch to improve its data-fitting ability. Finally, the data-enhanced model, PT-WGAN-GP, can stably generate high-quality faulty samples, which balances the testing dataset and completes the optimization training of network structure. Two cases under imbalanced data have verified the effectiveness of the proposed method, as well as its superiority over other widely used methods.

Enhancement multi-module network for few-shot leaky cable fixture detection in railway tunnel

The acquisition process of leaky cable fixture in railway tunnel is very complicated, and the number of faulty samples that can be collected is very rare, which seriously restricts the detection accuracy of faulty fixture. Based on the few-shot learning algorithm, we propose the Enhancement Multi-module Network (EMN) to achieve leaky cable fixture detection and further improve the mean accuracy in this paper. Firstly, the multi-scale module and the multi-region module are presented to explore intra-class similarities and inter-class differences through different scale and region information. Then, we propose the fusion module to enhance feature expression, whose role is to distinguish the semantic information between fixture and background, thereby facilitating to identify fixture in highly similar tunnel scenes. Finally, the similarity scores between support and query fixture sample pairs are computed by the relation module, and the mean squared error loss is optimized by adding weight balance coefficient. Experiments are conducted on leaky cable fixture data set, the result demonstrates that our proposed algorithm outperforms excellent accuracy compared to several state-of-the-art methods based on traditional handcrafted feature algorithm, deep learning algorithm, and few-shot learning algorithm.

Deep-learning-based fault detection and recipe optimization for a plastic injection molding process under the class-imbalance problem

AbstractThis paper proposes a supervised learning with a class-balancing loss function (SL-CBL) approach for fault detection and feature-similarity-based recipe optimization (FSRO) for a plastic injection molding process. SL-CBL is a novel method that can accurately classify an input sample as a normal or fault condition, even when the training data are severely class-imbalanced. The proposed class-balancing loss function consists of the weighted focal loss and the loss of the F1 score; together, these are used to correctly classify even a small number of faulty samples. SL-CBL is investigated with four classifiers of different structures; the classifiers consist of several fully connected and batch normalization layers. FSRO is an optimization scheme that finds the optimal recipe whose feature is similar to the features of normal samples. The optimal solution is obtained by minimizing the Euclidean distance to the centroid of the normal features. In this research, the proposed SL-CBL and FSRO methods are validated by applying them to an industrial plastic injection molding dataset. The validation results show that the proposed SL-CBL approach achieves the highest F1 score with the lowest misclassification rate, as compared to the alternative methods. When visualizing the feature space, the optimal recipe found by the FSRO scheme was found to be close to the centroid of the normal features, even if the initial recipe is classified as a fault. Furthermore, each variable of the optimized recipe lies within the confidence interval of 3${\rm{\sigma }}$ for the normal condition. This indicates that the optimal recipe is statistically similar to the normal samples.

Geometrical analysis of consecutive dynamic behaviors in process monitoring

Dynamic information plays an important role in developing monitoring models for industrial processes, and thus accurate evaluation of the dynamic behaviors within multiple consecutive process samples is of great significant. In this study, a geometrical analysis method is introduced to establish monitoring model for reflecting the operation status of the process system. By maximizing the volume of the obtained simplex, the source signals with least joint correlation are extracted from the archived process data. Based on the obtained signals, an offline model is developed with three monitoring statistics. Among them, two statistics are adopted to evaluate the static deviations of current sample, and the other one is used to reveal the dynamic behavior of the process system. After that, an online monitoring policy is designed for identifying the operation status of the industrial process based on the decisions of both static and dynamic statistics. Using the volume of simplex constructed by several consecutive samples, the dynamic behaviors of the process system can be better described and thus the detection sensitivity can be further improved. Besides, the normal shifts of operation conditions can also be distinguished from real process anomalies, so that the false alarms of the established model can be effectively reduced. Process data collected from a simulated process and a real industrial application is adopted to test the performance of the proposed method. Experimental results illustrate that the proposed method has better detection sensitivity for the faulty samples and can accurately identify varying operation condition.

Intelligent Fault Quantitative Identification via the Improved Deep Deterministic Policy Gradient (DDPG) Algorithm Accompanied With Imbalanced Sample

The imbalanced amount of faulty and normal samples seriously affects the performance of intelligent fault diagnosis models. Aiming to solve the above problem, an improved deep deterministic policy gradient algorithm (ResDPG) based on actor-critic architecture is proposed. In ResDPG, a multi-channel time-frequency representation (TFR) is obtained by the synchrosqueezed wavelet transform (SWT) to avoid the non-stationary of the original signal. ResNet is introduced to construct an actor network for extracting representative deep fault features to improve the accuracy of fault diagnosis, while AlexNet is utilized to build a critic network and guide the actor to train in the right direction according to the evaluation mechanism. The model constructs a rational and practical reward function based on the imbalance ratios and uses the minimum distance between the centers of the classes as the feedback of the reward. The optimized state transfer function improves the learning frequency of minority classes. Verified via two datasets of rolling bearing, ResDPG can independently and autonomously achieve accurate fault quantitative identification with high efficiency, stability, and generalization. It also achieves state-of-the-art performance under unbalanced data and variable load.

A Wasserstein generative digital twin model in health monitoring of rotating machines

Artificial intelligence-based rotating machine health monitoring and diagnosis methods often encounter problems, such as a lack of faulty samples. Although the simulation-based digital twin model may potentially alleviate these problems with sufficient prior knowledge and a large amount of time, the more demanding requirements of adaptivity, autonomy, and context-awareness may not be satisfied. This study attempted to address these problems by proposing a novel digital twin model referred to as the Wasserstein generative digital twin model (WGDT). The model employs a Wasserstein generative adversarial network (WGAN) as its core to model virtual samples with high fidelity to the healthy physical samples obtained from different industrial assets, thereby meeting the adaptivity requirement. Further, through a designed consistency test criterion mechanism, samples with high fidelity were generated by checking the similarity of distributions between generated samples and healthy physical samples to ensure that training process in conducted in a timely manner and manual involvement is avoided, thereby catering to the need for autonomy. This mechanism is based on the synchronous evolution of the generator and critic during training. Furthermore, the structure of the critic network can be customized according to the service-end tasks and testing conditions, thereby fulfilling the context awareness requirement. Subsequently, the critic network in the Wasserstein generative adversarial network (WGAN) can be used to perform different service-end tasks. The performance of the digital twin model was evaluated using two experimental cases and the results indicated that the WGDT model can efficiently and stably perform service-end tasks such as health monitoring, early fault detection, and degradation tracking without the requirement of prior knowledge, historical test samples, and faulty samples regarding the asset.

Generative-Adversarial Class-Imbalance Learning for Classifying Cyber-Attacks and Faults - A Cyber-Physical Power System

There has been an increasing interest in the use of data-driven techniques for classifying cyber-attacks and physical faults in cyber-physical systems. In real-world applications, the number of cyber-attack and faulty samples is usually far less than normal samples. This causes the skewed class distribution in data collected from cyber-physical systems. Training an accurate predictive model under skewed class conditions is not an easy task. In this work, we introduce a new generative adversarial framework for learning from skewed class distributions. This novel Adversarial Class-Imbalance Learning (ACIL) scheme has a novel loss function that is used during the adversarial training session. ACIL tries to iteratively adjust weights of an auxiliary multilayer perceptron to learn the minority class (i.e., cyber-attacks and physical faults) distributions along with the majority class (i.e., normal) distribution. Moreover, we devise an inclusive data-driven scheme for classifying cyber-attacks and faults, which includes four experiments of a baseline, nine state-of-the-art class-imbalance learning methods, two different generative-adversarial network-based approaches, and ACIL. These techniques are verified and compared through several experimental cyber-physical power scenarios. The obtained results show the effectiveness of ACIL for classifying samples of cyber-attacks and faults with skewed class distributions.

МЕХАНИЧЕСКИЕ СВОЙСТВА И КАЧЕСТВО ЭЛАСТОМЕРНЫХ ПОКРЫТИЙ ПОСЛЕ ИНФРАКРАСНОЙ ОБРАБОТКИ

One of the main disadvantages of the convective method of drying polymer coatings is their porosity. Porosity during infrared treatment is much less, and the quality of the coatings is significantly higher. However, the analysis of a priori information showed the lack of information about the infrared treatment of polymer coatings during the restoration of the body parts of the equipment. The effect of the infrared treatment regime on the mechanical properties of the F-40S elastomer-based nanocomposite was investigated. The mechanical properties of the elastomeric nanocomposite were assessed by the tensile strength of the films, relative elongation and specific deformation work when breaking the films according to the method of GOST 14236-81, the elastic modulus of the material for tension and compression - according to GOST 9550-81. The optimal mode of infrared processing of the nanocomposite was determined with an active experiment according to the V2 plan. Faulty samples were evaluated according to GOST 9407-84. As a result of regression analysis, a model of the dependence of the specific destruction work of the films of the F-40S elastomer nanocomposite on the infrared treatment mode was obtained. The optimal mode of thermal treatment of the elastomeric nanocomposite is established: 140,00S temperature, time 3.0 hours, at which the material films have the highest specific fracture work of 56.0 MJ/m3. The thermoradiation method, compared to the convective drying method, increases the strength of the nanocomposite films by 1.13 times, from 19.0 to 21.5 MPa. Infrared treatment increases the deformation of samples from 74 to 98%, 1.32 times. The stiffness of the material increased: an increase in the elastic modulus at tension from 158.3 to 161.8 MPa, by 3%, and compression - from 82.7 to 87.2 MPa, by 5.4%. The thermoradiation way, in comparison with a convective way, considerably increases quality of coverings of an elastomeric nanocomposite: the area of the destroyed covering decreases from 20 to 15%, by 1.33 times; concentration of a time in – from 0.67 to 0.54 pieces/cm2, by 1.24 times and the size of a time – from 0.109 to 0.095 mm, for 15%.

An Automated Machine Learning Approach for Real-Time Fault Detection and Diagnosis.

This work presents a novel Automated Machine Learning (AutoML) approach for Real-Time Fault Detection and Diagnosis (RT-FDD). The approach’s particular characteristics are: it uses only data that are commonly available in industrial automation systems; it automates all ML processes without human intervention; a non-ML expert can deploy it; and it considers the behavior of cyclic sequential machines, combining discrete timed events and continuous variables as features. The capacity for fault detection is analyzed in two case studies, using data from a 3D machine simulation system with faulty and non-faulty conditions. The enhancement of the RT-FDD performance when the proposed approach is applied is proved with the Feature Importance, Confusion Matrix, and F1 Score analysis, reaching mean values of 85% and 100% in each case study. Finally, considering that faults are rare events, the sensitivity of the models to the number of faulty samples is analyzed.

Fault Diagnosis of Rotary Machines Using Machine Learning

Classification of faults in rotary machines using machine learning is gaining attention in the field of science and engineering. In rotating machinery, misalignment is a common fault. This type of fault has been extensively studied in the literature using the vibration signals produced by rotary machines. This study proposes an approach based on machine learning techniques to diagnose misalignments in rotary machines under various conditions. A personalized diagnostic fault approach is proposed to detect misalignment faults. The approach includes three steps. First, the data acquisition model is developed to obtain signals (fault samples). The rotor vibration signals in stationary rotation conditions were obtained by two inductive proximity sensors with analog output, and the data were collected by a data acquisition device. Then, to generate the faulty training samples, each data acquisition signal is transformed to the frequency domain using Fast Fourier Transform (FFT). Finally, using the samples obtained through the feature selection techniques, machine learning algorithms Random Forest, Naïve Bayes and SVM were evaluated, resulting in classifications with different efficiencies. The results show that the SVM algorithm outperforms the Naïve Bayes and Random Forest algorithms when the same number of features is used. The proposed personalized diagnostic approach was applied to detect faults of rotating electrical machines misalignment with success.

Multiscale One-Class Classification Network for Machine Health Monitoring

Currently, most machine health monitoring methods require labeled abnormal or faulty samples to build a supervised classification model. However, it is difficult and expensive to collect enough faulty samples for accurate and reliable fault detection and diagnosis. To address this issue, we formulate machine health monitoring as a one-class classification problem and propose a new unsupervised health monitoring model named multiscale one-class classification network (MOCN) only relying on normal data without any label information. It consists of two sequential learning modules: a multiscale feature representation learning module to extract effective features from raw one-dimensional vibration signals and a one-classifier training module to build a hypersphere model to contain all normal samples with a minimum radius. Unlike traditional two-step one-class classification methods, both modules in our proposed MOCN are jointly optimized, and therefore it can simultaneously learn compact representations and train a deep one-class classifier. Our proposed MOCN method is evaluated through experiments on three datasets, and experimental results have proved the superiority of our proposed MOCN model compared with several existing methods.

An incipient fault diagnosis methodology using local Mahalanobis distance: Fault isolation and fault severity estimation

Recently, a detection methodology based on anchor estimation and the Mahalanobis distance, named the local Mahalanobis distance (LMD) technique, has been proposed for incipient fault detection. As an extension, this paper proposes the LMD based faulty variable isolation and fault severity estimation methods to offer an integrated incipient fault diagnosis solution. The LMD technique has been proved to be highly sensitive to incipient faults, robust to outliers, and free of distribution assumptions. Within the LMD framework, the faulty variable is recognized by analyzing the relative position of faulty samples and their corresponding anchors. Then an analytical expression of fault severity derived from the LMD index is established for the fault severity estimation task. Performance evaluations of the proposed methods based on a benchmark case of the Continuous-flow Stirred Tank Reactor (CSTR) process are provided in this work. The result shows that even for tiny deviation, such as 5dB fault to noise ratio, the total isolation performance still achieves 100% accuracy, and the relative error of two tricky faults in the case of severity estimation is less than 10%. The comparison study of different methods indicates that our solution outperforms divergence based techniques and reconstruction based contribution (RBC) methods for the two tasks.

A Modified Generative Adversarial Network for Fault Diagnosis in High-Speed Train Components with Imbalanced and Heterogeneous Monitoring Data

Data-driven methods are widely considered for fault diagnosis in complex systems. However, in practice the between-class imbalance due to limited faulty samples may deteriorate their classification performance. To address this issue, synthetic minority methods for enhancing data have been proved to be effective in many applications. Generative Adversarial Networks (GANs), capable of automatic features extraction, can also be adopted for augmenting the faulty samples. However, the monitoring data of a complex system may include not only continuous signals, but also discrete/categorical signals. Since the current GAN methods still have some challenges in handling such heterogeneous monitoring data, a Mixed Dual Discriminator GAN (noted as M-D2GAN) is proposed in this work. In order to render the expanded fault samples more aligned with the real situation, and improve the accuracy and robustness of the fault diagnosis model, different types of variables are generated in different ways, including: floating-point, integer, categorical, and hierarchical. For effectively considering the class imbalance problem, proper modifications are made to the GAN model, where a normal class discriminator is added. A practical case study concerning the braking system of a high-speed train is carried out to verify the effectiveness of the proposed framework. Compared to the classic GAN, the proposed framework achieves better results with respect to F-measure and G-mean metrics.

Imbalanced Sample Selection With Deep Reinforcement Learning for Fault Diagnosis

An imbalanced number of faulty and normal samples causes serious damage to the performance of the conventional diagnosis methods. To settle the data-imbalance fault diagnosis problem, this article presents a novel general imbalanced sample selection strategy (DiagSelect) based on deep reinforcement learning. In DiagSelect, the problem of imbalanced sample selection from the training set is formulated as a multiarmed bandit problem of deep reinforcement learning. The nondifferentiable optimization problem of imbalanced sample selection can be solved by the Markov decision process. The parameters of DiagSelect can be optimized by REINFORCE with the feedback of the validation set. DiagSelect performs intelligent imbalanced sample selection to obtain better diagnosis performance autonomously. As a data-level technique, DiagSelect can be easily used in conjunction with the conventional diagnosis models. DiagSelect is validated in a synthetic dataset and an industrial process dataset. The results have shown the effectiveness, stability, and transferability of DiagSelect.

Data-Driven Fault Diagnosis Techniques: Non-Linear Directional Residual vs. Machine-Learning-Based Methods.

Linear dependence of variables is a commonly used assumption in most diagnostic systems for which many robust methodologies have been developed over the years. In case the system nonlinearities are relevant, fault diagnosis methods, relying on the assumption of linearity, might potentially provide unsatisfactory results in terms of false alarms and missed detections. In recent years, many authors have proposed machine learning (ML) techniques to improve fault diagnosis performance to mitigate this problem. Although very powerful, these techniques require faulty data samples that are representative of any fault scenario. Additionally, ML techniques suffer from issues related to overfitting and unpredictable performance in regions which are not fully explored in the training phase. This paper proposes a non-linear additive model to characterize the non-linear redundancy relationships among the system signals. Using the multivariate adaptive regression splines (MARS) algorithm, these relationships are identified directly from the data. Next, the non-linear redundancy relationships are linearized to derive a local time-dependent fault signature matrix. The faulty sensor can then be isolated by measuring the angular distance between the column vectors of the fault signature matrix and the primary residual vector. A quantitative analysis of fault isolation and fault estimation performance is performed by exploiting real data from multiple flights of a semi-autonomous aircraft, thus allowing a detailed quantitative comparison with state-of-the-art machine-learning-based fault diagnosis algorithms.

Fault clustering by small‐entropy nonnegative matrix factorizations

Fault clustering attempts to partition a set of faulty samples into several clusters, allowing the exploration of the underlying pattern of faults. Nonnegative matrix factorizations (NMFs) are good candidates for fault clustering since they are inherently capable of data clustering and variants of the k-means algorithm. However, NMFs always show poor performance in real-world clustering applications for their naive data clustering mechanism. To improve the clustering performance of the existing NMFs and solve the fault clustering problem, this paper proposes a new type of NMFs, called small-entropy nonnegative matrix factorizations (SENMFs). SENMFs impose a small amount of entropy on the cluster probability distribution of each sample to avoid ambiguous clustering results. Moreover, the algorithm for SENMFs is convergent in theory. We selected three types of faulty samples of the penicillin fermentation process for fault clustering. The case study results showed that SENMFs exceed the state-of-the-art NMFs and k-means in terms of fault clustering performance.

Online reconstruction and diagnosibility analysis of multiplicative fault models for process-related faults

Multiplicative faults generally refer to the change of process parameters or structures which are well-suited to represent the process-related anomalies. Unlike sensor faults and external disturbances that are added into process observations and independent with process states, process-related faults directly influence process states such that it is more challenge to reconstruct and diagnose. To address the process-related fault diagnosis, an online fault reconstruction method based on the multiplicative fault model is proposed with the commonly used multivariate statistical process monitoring framework. The fault reconstruction strategy based on the multiplicative fault representation is given by minimizing reconstruction errors. The diagnosability of the proposed reconstruction method is guaranteed for the change of a single parameter, also known as a unidimensional fault. Moreover, the reconstruction-based contribution is derived for providing heuristic references when diagnosing multidimensional faults. Experiments on a numerical example and a simulated continuous stirred tank heater process benchmark are carried out to investigate the effectiveness of the proposed method. The results show that this method can accurately diagnose the faulty variable or loop and further reconstruct the faulty samples into normal ones.

CNN-DST: Ensemble deep learning based on Dempster–Shafer theory for vibration-based fault recognition

Nowadays, using vibration data in conjunction with pattern recognition methods is one of the most common fault detection strategies for structures. However, their performances depend on the features extracted from vibration data, the features selected to train the classifier, and the classifier used for pattern recognition. Deep learning facilitates the fault detection procedure by automating the feature extraction and selection, and classification procedure. Though, deep learning approaches have challenges in designing its structure and tuning its hyperparameters, which may result in a low generalization capability. Therefore, this study proposes an ensemble deep learning framework based on a convolutional neural network (CNN) and Dempster–Shafer theory (DST), called CNN-DST. In this framework, several CNNs with the proposed structure are first trained, and then, the outputs of the CNNs selected by the proposed technique are combined by using an improved DST-based method. To validate the proposed CNN-DST framework, it is applied to an experimental dataset created by the broadband vibrational responses of polycrystalline nickel alloy first-stage turbine blades with different types and severities of damage. Through statistical analysis, it is shown that the proposed CNN-DST framework classifies the turbine blades with an average prediction accuracy of 97.19%. The proposed CNN-DST framework is benchmarked with other state-of-the-art classification methods, demonstrating its high performance. The robustness of the proposed CNN-DST framework with respect to measurement noise is investigated, showing its high noise-resistance. Further, bandwidth analysis reveals that most of the required information for detecting faulty samples is available in a small frequency range.