Application In Fault Diagnosis Research Articles

Generalization analysis and improvement of CNN-based nuclear power plant fault diagnosis model under varying power levels

The domain discrepancy caused by power level gap between training set (source domain) and test set (target domain) limits the generalization of data-driven models in practical nuclear power plant fault diagnosis applications. In this study, a highly fine-grained quantitative generalization description of Convolutional Neural Network (CNN) model is conducted with high-dimensional and strong-nonlinear complex nuclear power plant simulation data. Results show that with source domain of single power level, CNN suffers from over-fitting and poor domain discrepancy generalization; with source domain of multiple power levels, the sensitivity of CNN to minor domain discrepancy is greatly reduced and its generalization is significantly boosted. Besides, a novel Artificial Disturbance Method (ADM) based domain discrepancy generalization promotion framework is proposed in this study, which alleviates the over-fitting of CNN by adding disturbed training data to its training set. The feasibility and superiority of the framework is proven when Gaussian noise disturbance or uniformly distributed noise disturbance is adopted to generate disturbed training data. The ADM-based framework reduces the requirement for the number of power levels contained in source domain, this advantage endows it with strong practicability in actual nuclear power plant fault diagnosis tasks where the available source domain data are extremely limited.

Analysis of a class of Lorenz-like stochastic resonance model and its application in induction motor fault diagnosis

Analysis of a class of Lorenz-like stochastic resonance model and its application in induction motor fault diagnosis

Physical model driven fault diagnosis method for shield Machine hydraulic system

The hydraulic system of shield machine is critical component, and any faults in this system can have severe consequences on construction progress and even pose risks to the safety of construction personnel. However, the high manufacturing and construction costs of shield machines, as well as the use of highly reliable hydraulic components, make it challenging to obtain a large number of fault samples that accurately reflect the actual operating conditions of the hydraulic system. This limitation is one of the major obstacles to the practical application of intelligent fault diagnosis in shield machines. Furthermore, due to the large scale and complex structure of the hydraulic system in shield machines, few scholars have studied the bottleneck that restricts the fault diagnosis of the hydraulic system of the shield machine. To address these challenges, this paper proposes physical model-driven approach for fault diagnosis in shield machine hydraulic systems, aiming to overcome the lack of fault feature information during the fault diagnosis process. Firstly, physical models for various hydraulic components of the shield machine hydraulic system are established, and the relationship between the physical models and fault characteristics is investigated to provide a theoretical basis for fault injection. Then, a physical model correction method based on cosine similarity is proposed. Using recorded high-frequency fault data from shield machine hydraulic systems, fault injection is performed to obtain simulated data that effectively reflects the actual fault states and normal states of the hydraulic system. Finally, using the simulated data as training samples and measured data as test samples, various artificial intelligence diagnosis models, such as Support Vector Machines (SVM), Extreme Learning Machines (ELM), and Convolutional Neural Networks (CNN), are employed for diagnostic classification. The classification results demonstrate the feasibility and effectiveness of the physical model-driven approach for fault diagnosis in shield machine hydraulic systems. Furthermore, this approach exhibits strong universality and scalability, providing a solution for fault diagnosis scenarios with limited or even no fault samples.

Sparse random similarity feature decomposition and its application in gear fault diagnosis

As a new signal decomposition method, sparse random mode decomposition (SRMD) does not depend on the signal sampling process and has good anti-mode aliasing performance. However, SRMD needs preset decomposition parameters and cannot perform accurate decomposition for strong noise interference signals. Thus, this paper proposes a sparse random similarity feature decomposition (SRSFD) method to overcome the problems of SRMD. Firstly, SRSFD utilizes the windowed sinusoidal function to randomly generate time–frequency features. Then, by solving sparse optimization problem, the sparse random feature approximation of the input signal can be obtained and its corresponding time–frequency features can be applied to reconstruct the initial signal components. Finally, the initial signal components are regrouped by using periodic similarity to obtain the final components. The analytical results of simulation and actual gear fault signals show that SRSFD can effectively achieve adaptive signal decomposition and has better decomposition performance and noise robustness compared with existing methods.

Metric Learning-Guided Semi-Supervised Path-Interaction Fault Diagnosis Method for Extremely Limited Labeled Samples under Variable Working Conditions.

The lack of labeled data and variable working conditions brings challenges to the application of intelligent fault diagnosis. Given this, extracting labeled information and learning distribution-invariant representation provides a feasible and promising way. Enlightened by metric learning and semi-supervised architecture, a triplet-guided path-interaction ladder network (Tri-CLAN) is proposed based on the aspects of algorithm structure and feature space. An encoder-decoder structure with path interaction is built to utilize the unlabeled data with fewer parameters, and the network structure is simplified by CNN and an element additive combination activation function. Metric learning is introduced to the feature space of the established algorithm structure, which enables the mining of hard samples from extremely limited labeled data and the learning of working condition-independent representations. The generalization and applicability of Tri-CLAN are proved by experiments, and the contribution of the algorithm structure and the metric learning in the feature space are discussed.

A novel momentum prototypical neural network to cross-domain fault diagnosis for rotating machinery subject to cold-start

Cross-domain rotating machinery fault diagnosis has achieved great success recently with the development of deep transfer learning. However, conventional deep transfer learning methods encounter a severe decline in prediction accuracy when fault samples are limited. Moreover, conventional deep transfer learning methods require additional parameter tuning rather than cold-start when applied to the target tasks, hampering their implementation in practical fault diagnosis applications. In this paper, a novel method, named momentum prototypical neural network (MoProNet), is proposed for cross-domain few-shot rotating machinery fault diagnosis. The MoProNet progressively updates the support encoder to address the prototype oscillation problem and enable the model to apply limited source domain samples to predict target domain faults with cold-start. The performance of the proposed MoProNet is tested on a bearing dataset and a hardware-in-the-loop high-speed train simulation platform, respectively, with over forty cross-domain few-shot fault diagnosis tasks. The experimental results demonstrate that the proposed MoProNet achieves satisfactory results and outperforms the other comparable methods in the same cross-domain few-shot scenarios with the simple AlexNet backbone.

A novel feature extraction method based on symbol-scale diversity entropy and its application for fault diagnosis of rotary machines

Multiscale entropy-based methods have made great progress in the field of health condition monitoring and fault diagnosis of machines due to their powerful feature representation capabilities. However, existing multiscale entropy methods suffer from three major obstacles: high fluctuation under large scale-factor, loss of high-frequency information, and poor robustness to noises. Thus, this work proposes a symbol-scale analysis method to deal with the above problems. In one aspect, to capture fault features from the time series over multiple time scales, time-delay process of different intervals is utilized to obtain long-term features and short-term features. In the other aspect, symbol-scale analysis introduces a symbolization procedure and maps time series into a corresponding sequence of symbols to overcome the limitation of weak fault extraction under a low-signal-to-noise ratio environment. Moreover, the symbol-scale entropy approach is developed by integrating with diversity entropy, called symbol-scale diversity entropy. The effectiveness of the proposed strategy is intensively validated using two simulated signals and experimental cases. Results demonstrate its advantages in dynamic change tracking ability and calculation efficiency by comparing it with other state-of-the-art entropy methods. Apart from diversity entropy, the versatility of incorporating the proposed symbol-scale analysis and other entropy methods is also verified using experimental data.

Engine modelling architecture study for hybrid electric vehicle diagnosis application

Model-based control, calibration, and diagnosis are widely used in hybrid electric vehicles (HEVs) studies. For the real-time application, the contradiction lies in the model accuracy and fast response. This problem is particularly evident in engine modeling due to its complexity. In this study, three types of engine model architectures were studied for HEV Hardware In the Loop Simulation (HILS) diagnosis application. It shows that in terms of the real-time factor, the MAP model, obtaining all engine outputs only by looking up maps, has the best real-time performance. Both Mean Value Model (MVM) and Fast Running Model (FRM), with less than 3% steady-state error and preserving the engine's dynamic characteristics, have better real-time performance than the detailed model. What is more, the real-time performance of MVM is about 14% faster than that of FRM. MVM-based and FRM-based vehicle models were verified respectively in the HILS for the fault classification using support vector machine (SVM) and for the engine's misfire fault diagnosis. Since the MVM combines multiple detailed cylinders into a single mean value cylinder through neural network training, the MVM is typically used in diagnosis situations where fast computation speed is especially important and the detailed characterization of engine processes is considered relatively unimportant, such as vehicle-level fault diagnosis applications involving control system design, thermal management system fault, and power system fault, etc. Since the FRM reduces cylinder calculations by using the “cylinder translation method”, which preserves the working process within a single-cylinder cycle, it can also be used for fault diagnosis applications requiring engine performance with a crank-angle resolution, such as cylinder pressure diagnostics, combustion diagnostics, and timing control of valve and fuel injection, etc.

A double-layer fault diagnosis strategy for electric vehicle batteries based on Gaussian mixture model

Battery fault diagnosis is essential to ensure the safe operation of electric vehicles (EVs). In this paper, due to the complexity of EVs’ battery thermal runaway tracing investigation and the limited capacity of on-board computing system, a double-layer fault diagnosis strategy for abnormal cells is proposed. The method bases on probability distribution, which can accurately trace a faulty cell and avoid misinterpreting a normal cell. In this method, unified statistical features are extracted from the big data during vehicle charging, and the corresponding statistical values are analyzed based on Gaussian mixture model and abnormal alarm is made based on the risk accumulation in double-layer diagnostics. The electric vehicles with thermal runaway accident are taken as examples to verify the method, and based on the data of normal-running vehicles, the false alarm tests are carried out. The verification results show that the proposed method can not only successfully identify the outlier cells, but also not generate false alarm, which is conducive to the practical application of fault diagnosis in the on-board battery management system.

A Double Interpolation and Mutation Interval Reconstruction LMD and Its Application in Fault Diagnosis of Reciprocating Compressor

The accuracy and stability of the envelope estimation function are enduring issues throughout the research process of LMD. This paper presents double interpolation and mutation interval reconstruction local mean decomposition (DIMIRLMD) to improve the stability of the demodulation process and the accuracy of PF components. DIMIRLMD first proposes a mutation interval reconstruction envelope algorithm using extreme symmetry points to suppress the demodulation mutation phenomenon, which disturbs the stability of the demodulation process, and then selects the optimal PF component from a double interpolation PF component library based on the index of orthogonality (IO) for a better hierarchical property. DIMIRLMD was employed to analyze the simulation signal and vibration signal of a reciprocating compressor in an oversized bearing clearance state, and the results illustrate its performances are more excellent than those of three other LMD methods. Furthermore, the envelope frequency spectrum obtained from the proposed LMD presents a clear double rotation fault frequency and lower noise disturbance.

Prior knowledge-embedded meta-transfer learning for few-shot fault diagnosis under variable operating conditions

In recent years, intelligent fault diagnosis based on deep learning has achieved vigorous development thanks to its powerful feature representation ability. However, scarcity of high-quality data, especially samples under severe fault states, and variable operating conditions have limited the industrial application of intelligent fault diagnosis. To alleviate this predicament, a novel prior knowledge-embedded meta-transfer learning (PKEMTL) is proposed for few-shot fault diagnosis with limited training data and scarce test data. The method focuses on the problem of few-shot fault diagnosis under variable operating conditions to improve adaptability. Different from traditional models, the PKEMTL employs a metric-based meta-learning framework and embeds prior knowledge to enable cross-task learning under variable operating conditions. Specifically, order tracking is firstly introduced as preliminary prior information for data augmentation, and then the augmented data are divided into a series of meta-tasks. Secondly, the meta-tasks are performed by lightweight multiscale feature encoding to obtain high-level feature representations. Next, the meta-learning module based on diagnostic knowledge embedding guides the model to acquire meta-knowledge of speed generalization by constructing the self-supervised task to embed additional prior knowledge into the meta-training process. The generalization performance of the model is further improved by adaptive information fusion learning as a comprehensive decision-making module. Two case studies under variable operating conditions are implemented to validate the effectiveness and superiority of the proposed few-shot fault diagnosis method.

Intelligent Bearing Fault Diagnosis Based on Feature Fusion of One-Dimensional Dilated CNN and Multi-Domain Signal Processing

Finding relevant features that can represent different types of faults under a noisy environment is the key to practical applications of intelligent fault diagnosis. However, high classification accuracy cannot be achieved with only a few simple empirical features, and advanced feature engineering and modelling necessitate extensive specialised knowledge, resulting in restricted widespread use. This paper has proposed a novel and efficient fusion method, named MD-1d-DCNN, that combines statistical features from multiple domains and adaptive features retrieved using a one-dimensional dilated convolutional neural network. Moreover, signal processing techniques are utilised to uncover statistical features and realise the general fault information. To offset the negative influence of noise in signals and achieve high accuracy of fault diagnosis in noisy settings, 1d-DCNN is adopted to extract more dispersed and intrinsic fault-associated features, while also preventing the model from overfitting. In the end, fault classification based on fusion features is accomplished by the usage of fully connected layers. Two bearing datasets containing varying amounts of noise are used to verify the effectiveness and robustness of the suggested approach. The experimental results demonstrate MD-1d-DCNN’s superior anti-noise capability. When compared to other benchmark models, the proposed method performs better at all noise levels.

An Adaptive Parameterized Domain Mapping Method and Its Application in Wheel-Rail Coupled Fault Diagnosis for Rail Vehicles.

The rapid development of cities in recent years has increased the operational pressure of rail vehicles, and due to the characteristics of rail vehicles, including harsh operating environment, frequent starting and braking, resulting in rails and wheels being prone to rail corrugation, polygons, flat scars and other faults. These faults are coupled in actual operation, leading to the deterioration of the wheel-rail contact relationship and causing harm to driving safety. Hence, the accurate detection of wheel-rail coupled faults will improve the safety of rail vehicles' operation. The dynamic modeling of rail vehicles is carried out to establish the character models of wheel-rail faults including rail corrugation, polygonization and flat scars to explore the coupling relationship and characteristics under variable speed conditions and to obtain the vertical acceleration of the axle box. An APDM time-frequency analysis method is proposed in this paper based on the PDMF adopting Rényi entropy as the evaluation index and employing a WOA to optimize the parameter set. The number of iterations of the WOA adopted in this paper is decreased by 26% and 23%, respectively, compared with PSO and SSA, which means that the WOA performs at faster convergence speed and with a more accurate Rényi entropy value. Additionally, TFR obtained using APDM realizes the localization and extraction of the coupled fault characteristics under rail vehicles' variable speed working conditions with higher energy concentration and stronger noise resistance corresponding to prominent ability of fault diagnosis. Finally, the effectiveness of the proposed method is verified using simulation and experimental results that prove the engineering application value of the proposed method.

A dual-experience pool deep reinforcement learning method and its application in fault diagnosis of rolling bearing with unbalanced data

A dual-experience pool deep reinforcement learning method and its application in fault diagnosis of rolling bearing with unbalanced data

A New Algorithm of Cubic Dynamic Uncertain Causality Graph for Speeding Up Temporal Causality Inference in Fault Diagnosis

Representing and reasoning uncertain causalities have diverse applications in fault diagnosis for industrial systems. Owing to the complicated dynamics and a multitude of uncertain factors in such systems, it is hard to implement efficient diagnostic inference when a process fault occurs. The cubic dynamic uncertain causal graph was proposed for graphically modeling and reasoning about the fault spreading behaviors in the form of causal dependencies across multivariate time series. However, in certain large-scale scenarios with multiconnected and time-varying causalities, the existing inference algorithm is incapable of dealing with the logical reasoning process in an efficient manner. We, therefore, explore the solutions to enhanced computational efficiency. Causality graph decomposition and simplification, and graphical transformation, are proposed to reduce model complexity and form a minimal causality graph. An algorithm, event-oriented early logical absorption, is also presented for logical reasoning. It is mainly intended to minimize the computational costs for compulsory absorption operations in the early stage of reasoning process. The effectiveness of the proposed algorithm is verified on the secondary loop model of nuclear power plant by utilizing the fault data derived from a nuclear power plant simulator. The results show the anticipated capability of efficient fault diagnosis of the proposed algorithm for large-scale dynamic systems.

A local spectrum enhancement‐based method and its application in incipient fault diagnosis of rotating machinery

AbstractIncipient faults of gears and rolling bearings in rotating machineries are very difficult to identify using traditional envelope analysis methods. To address this challenge, this paper proposes an effective local spectrum enhancement‐based diagnostic method that can identify weak fault frequencies in the original complicated raw signals. For this purpose, a traversal frequency band segmentation technique is first proposed for dividing the raw signal into a series of subfrequency bands. Then, the proposed synthetic quantitative index is constructed for selecting the most informative local frequency band (ILFB) containing fault features from the divided subfrequency bands. Furthermore, an improved grasshopper optimization algorithm‐based stochastic resonance (SR) system is developed for enhancing weak fault features contained in the selected most ILFB with less computation cost. Finally, the enhanced weak fault frequencies are extracted from the output of the SR system using a common spectrum analysis. Two experiments on a laboratory planetary gearbox and an open bearing data set are used to verify the effectuality of the proposed method. The diagnostic results demonstrate that the proposed method can identify incipient faults of gears and bearings in an effective and accurate manner. Furthermore, the advantages of the proposed method are highlighted by comparison with other methods.

Research on online monitoring of grid current transformer based on transformers and BiGRU

Transformers play a crucial role in ensuring the safety of power grids. It is of great value to diagnose faults using the large amount of power data generated by the grid. It is possible to detect internal latent faults in transformers in advance of their occurrence if the normal operating condition of the transformer is detected in a timely manner. To perform online fault diagnosis of grid current transformers, we combine the Transformer and BiGRU methods. There is a temporal component to the fault input sample sequences. By using Transformer’s multi-headed attention mechanism to extract deep features from fault input sample sequences, the temporal association between latent variables can be fully exploited. As a result of the extraction of features, BiGRU is used to generate fault category coding as an output. The experimental results indicate that using the proposed algorithm achieves better results than using a single model, which is useful for the study and application of fault diagnosis in power grids for current transformers.

Simulating electrified powertrain noise and vibration measurement signals for machine learning based fault diagnosis applications

Machine learning classification is a common method for vehicle noise and vibration (N&V) fault diagnosis which helps improve vehicle safety, comfort, and reduce maintenance cost. To improve the accuracy of classification, the model requires sufficient training data which is often expensive and time-consuming to acquire. A possible solution to resolve this limitation is to extract representative features from measurement signals and generate realistic simulatied signals based on a smaller set of high-quality N&V measurement signals. A method to simulate N&V measurement signals of rotating machineries such as electrified powertrains is proposed. In its feature extraction process, tonal and broadband components of measurement data are separated. After the separation, time-varying tonal amplitudes, broadband spectra, and various related statistical features are extracted. In the simulation process, tonal and broadband signals were simulated using the extracted features with random variations added to each feature. In the current work N&V measurement signals of rotating machinery are simulated with relatively low speed fluctuation and results are presented.

Data augmentation via variational mode reconstruction and its application in few-shot fault diagnosis of rolling bearings

Currently, the popular fault diagnosis methods based on deep learning encounter a common limitation in that their accuracy heavily relies on an adequate number of training samples. However, collecting fault samples in real-world scenarios is often challenging. To overcome this challenge, this paper develops a novel data augmentation method named variational mode reconstruction (VMR) to generate augmented samples with similar features to the original samples. The first key point is the random weighting of a certain randomly chosen intrinsic mode function (IMF). Another key point is that the mean values and standard deviations of the augmented samples remain consistent with the original samples. Next, the augmented balanced dataset is utilized to train a deep residual shrinkage network (DRSN), which is then employed for the classification of test samples. Finally, the effectiveness and superiority of the developed VMR in few-shot fault diagnosis are verified through a series of experiments.

Normal Sinkhorn Distance: A novel metric for evaluating generated signals and its application in mechanical fault diagnosis

Nowadays, leveraging data augmentation-based methods to address the data-shortage problem in diagnosis field becomes fairly prevailing, while assessing the quality of the generated data receives little attention. Typically, the data quality is evaluated by straightforward employing some existing shallow functions or simple classification models, which have several disadvantages. In this paper, the limitations of existing techniques are comprehensively summarized and illustrated by pathological examples. Accordingly, a novel metric method called Normal Sinkhorn Distance (NSD) is developed to evaluate the quality of vibration data. Firstly, the raw vibration data are directly transformed into time–frequency images with a tensor format and stored to ensure the following computation efficiency. Then an off-the-shelf pretrained Inception model is employed to capture their high-order data structure and amplitude-wise dependence. Finally, the NSD metric is constructed and adopted to measure the similarity between the real data and generated data in a high-dimensional feature space. Extensive results indicate that the proposed NSD provides a more reliable indication for practitioners than the existing evaluation measures on the data quality.