Machine Learning-based Fault Diagnosis Research Articles

Interpretable intelligent fault diagnosis strategy for fixed-wing UAV elevator fault diagnosis based on improved cross entropy loss

The current popular machine learning-based fault diagnosis methods make it difficult to explain the diagnostic results, leading to low user trust in such diagnostic techniques. In this regard, this paper explores the study of the interpretability of intelligent fault diagnosis algorithms using the elevator of a fixed-wing unmanned aerial vehicle (UAV) as a diagnostic object. The Transformer model combines excellent modeling capability and efficient sequence data processing, is chosen to mine fault signal features to guarantee accurate diagnosis. Among the proposed interpretable fault diagnosis models, the local interpretable model-agnostic explanations (LIME) model is used to provide explicit interpretability for the decision-making process of the diagnosis model. In addition, a loss function called RDCE (reinforced diagnostic cross-entropy) Loss is designed to minimize the negative impact of different sample sizes for different fault types on the diagnostic performance. This loss function is designed to weigh the various types of faults to speed up the convergence of the model and improve the diagnostic accuracy. By comparing the proposed diagnostic strategy with other commonly used machine learning models, including long short term memory and recurrent neural network (RNN), the average diagnostic accuracy of the proposed diagnostic strategy is 99.97%, significantly better than that of the comparison algorithms. At the same time, this paper provides an in-depth interpretable analysis of the diagnostic process of the Transformer. The diagnostic process of the Transformer model gives the reasons for the diagnostic results from the point of view of the kind of features processed by the model. Based on this, the diagnostic model is simplified. After streamlining the number of features from 40 to 24 according to their importance, the diagnostic accuracy of the model is improved by 0.26%, and the diagnostic efficiency is improved. In addition, the proposed diagnostic strategy also shows significant advantages in terms of noise robustness.

Chiller fault diagnosis based on improved variational autoencoder and co-training framework: A case study of insufficient samples

Implementing efficient automatic fault diagnosis is critical for saving energy and minimizing financial losses in the heating ventilation air-conditioning (HVAC) systems of commercial buildings. However, the limited quantity and weak features of fault samples acquired during HVAC operations hinder the effectiveness of conventional machine learning-based fault diagnosis methodologies. This paper proposes a method based on an improved conditional variational autoencoder (MCVAE) and co-training (CT) ideology to address the issue of insufficient training samples. Initially, we employ MCVAE to synthesize an extensive dataset of chiller fault samples from the original training dataset. Subsequently, the beneficial samples for training our fault diagnostic classifier, namely high-quality samples, are selected from the generated dataset using the CT-based framework. Finally, the selected high-quality samples are merged into the original training dataset to train the ultimate fault classifiers. Experimental results demonstrate that our proposed method outperforms in effectiveness and efficiency compared to recently published methods. For instance, in the case of fault level 1 compared to the suboptimal model, our approach exhibits improvements of 2.41% when each type has 5 fault samples.

Small Sample Fiber Full State Diagnosis Based on Fuzzy Clustering and Improved ResNet Network

The optical time domain reflectometer (OTDR) curve features of communication fibers exhibit subtle differences among their normal, subhealthy, and faulty operating states, making it challenging for existing machine learning-based fault diagnosis algorithms to extract these minute features. In addition, the OTDR curve field fault data are scarce, and data-driven deep neural network that needs a lot of data training cannot meet the requirements. In response to this issue, this paper proposes a communication fiber state diagnosis model based on fuzzy clustering and an improved ResNet. First, the pretrained residual network (ResNet) is modified by removing the classification layer and retaining the feature extraction layers. A global average pooling (GAP) layer is designed as a replacement for the fully connected layer. Second, fuzzy clustering, instead of the softmax classification layer, is employed in ResNet for its characteristic of requiring no subsequent data training. The improved model requires only a small amount of sample training to optimize the parameters of the GAP layer, thereby accommodating state diagnosis in scenarios with limited data availability. During the diagnosis process, the OTDR curves are input into the network, resulting in 512 features outputted in the GAP layer. These features are used to construct a feature vector matrix, and a dynamic clustering graph is formed using fuzzy clustering to realize the fiber state diagnosis. Through on-site data detection and validation, it has been demonstrated that the improved ResNet can effectively identify the full cycle of fiber states.

A Novel Machine Learning Based Fault Diagnosis Method for All Gas-Path Components of Heavy Duty Gas Turbines With the Aid of Thermodynamic Model

A Novel Machine Learning Based Fault Diagnosis Method for All Gas-Path Components of Heavy Duty Gas Turbines With the Aid of Thermodynamic Model

Machine learning-based fault diagnosis for three-phase induction motors in ventilation systems

ABSTRACTInduction motors are crucial in industrial drive systems and nuclear and radiological facilities due to their durability and ease of maintenance. Even though the systems are reliable, they have several faults. One of these faults is the instability of the facility’s electrical network. This may lead to catastrophic consequences such as instability in HVAC and ventilation systems, which affects operational safety. Air instability in the HVAC system is a critical concern in the radio pharmaceutical industry. Controlled environments are vital for safety and regulatory compliance. Maintaining air quality and stability is crucial to protect radioactive materials and ensure personnel safety. A healthy and faulty motor condition dataset proposed and used in the study was generated based on a simulated actual supply of electric data. A fault diagnosis is necessary to increase reliability and minimise risks. This paper proposes a simple, reliable, and economical machine-learning-based fault classifier for induction motors. The method analyzes stator currents, motor speed, torque, and three-phase voltage supply to classify some faults resulting from high amplitude and zeros of voltage. A case study of an air handling unit with a 2 hp, 4-pole, 50 Hz three-phase induction motor was modelled and tested with the proposed dataset using various classifiers, including decision trees, logistic regression, discriminant analysis, Naive Bayes, Ensemble, and K-Nearest Neighbor (KNN).

Machine learning-based fault diagnosis for high-voltage transmission lines

In order to comply with the speed and diversification process of human economic and technological development, the power system gradually develops in the direction of intelligence and complexity, and it is urgent to combine transmission line fault diagnosis methods with machine learning pattern recognition problems in order to reflect the representative characteristics of transmission lines under different operating conditions and different operation modes. In this paper, a convolutional neural network (CNN) model is established for transmission line fault diagnosis, and a series of parameters such as weights, bias, learning rate, batch size and loss function of the proposed CNN model are debugged through experimental comparison, and the best CNN model structure is determined through experimental comparison and debugging, and the recognition accuracy is higher than other traditional machine learning algorithms, which can be as high as 98.38%, with validity, practicality and advancedness.

A novel semi-supervised fault diagnosis method for chillers based on neighbor-optimized graph convolutional network

The fault diagnosis of chillers is of great significance in reducing building energy consumption and extending the operational lifespan of refrigeration equipment. Several popular machine learning-based fault diagnosis methods rely heavily on many labeled samples. However, such samples are difficult to obtain in practice due to sparse fault data and high labeling costs. This limits the application of ML-based fault diagnosis methods based on supervised learning. To reduce the dependence on labeled samples, this paper proposes a novel chiller fault diagnosis method based on neighbor-optimized graph convolutional network. The method improves the utilization efficiency of unlabeled samples by mining the spatio-temporal relationship between a large number of unlabeled samples and a limited number of labeled samples. And it dynamically adjusts the graph's structure by optimizing the number of adjacent samples in the correlation graph to obtain better diagnostic results. Its effectiveness is validated on the authoritative dataset ASHRAE RP-1043 and a more challenging dataset of real-world chillers in a building. Experimental results show that the proposed method can achieve better diagnostic performance than the state-of-the-art methods.

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.

Fault-Prototypical Adapted Network for Cross-Domain Industrial Intelligent Diagnosis

Despite rapid advances in machine learning based fault diagnosis, their identical distribution assumption of the training (source domain) and testing data (target domain) is generally challenged in industrial applications due to the variation of working conditions. In this article, a fault-prototypical adapted network (FPAN) is proposed, which enables cross-domain industrial intelligent fault diagnosis aided by deep transfer learning. First, a similarity learning-based discrimination module is designed to learn fault prototypes (FPs) that are representative for each fault and discriminative across different faults. Then, a fault prototypical-adaptation module is developed, which adapts the multiple FPs to the target dataset and enables more precise category-wise domain invariance. The two modules are trained simultaneously to extract transferrable and discriminative FPs, by which the cross-domain intelligent diagnosis can be readily achieved. Experimental results on two industrial cases illustrate that the proposed approach learns transferable feature representations that better reduce domain discrepancy, and provides improved diagnosis performance on target data. Note to Practitioners—As the distribution of source and target domain data may differ due to the varying working conditions, cross-domain intelligent fault diagnosis plays an increasingly important role in practical industrial cases. This work presents the FPAN model, in which the fault discrimination and the category-wise adaptation are naturally connected and unified by exploring representative virtual FP for each category, irrespective of which domain that the category comes from. Connected by the adaptive FPs, the similarity learning-based discrimination and the fault-prototypical adaptation modules can benefit from each other to boost the target learning. Finally, the cross-domain fault diagnosis can be performed based on the similarities with different FP representations. The proposed method is readily applicable to various intelligent fault diagnosis problems, including the mechanical cases and the industrial processes.

Machine learning-based fault diagnosis for research nuclear reactor medium voltage power cables in fraction Fourier domain

Fault diagnosis of Medium Voltage power Cables (MVCs) research nuclear reactor, incredibly inaccessible/remote ones, has to be carefully identified, located, and fixed within a short time. Therefore, this paper proposes a perfect simultaneous fault diagnosis scheme based on Multiclass Support-Vector Machine (MCSVM) in the fractional Fourier domain. First, the three-phase sending currents are simulated under different conditions then their features are extracted using Discrete Fractional Fourier Transform (DFRFT). Afterward, the features reduction process occurs via the Singular Value Decomposition (SVD) approach. MCSVM scheme is used to diagnose faults (i.e., discover, categorize, and trace) using reduced features outcome from DFRFT and SVD stages. Alternating Transient Program/Electromagnetic Transient Program (ATP/EMTP) simulations have been carried out for 22 kV unreachable MVC. Different kernels of SVM, i.e., linear, quadratic, or polynomial, and diverse factors of DFRFT, i.e., α, are investigated in simulations to obtain the optimum performance (i.e., best α and kernel pair). Hence, performance analysis of the proposed diagnosis method under different conditions (i.e., various fault resistances, locations, and inception angles) concluded two highest accuracy and lowest time settings, which were found at α = 0.5 (for both) quadratic kernel, and linear kernel, respectively. Moreover, the linear kernel achieves 99.8% accuracy rate, the lowest execution time (10 ms), and fault tracing error rate of 0.525789%, which is proper for real-time applications. Besides, our proposed method is more reliable and accurate against variable operating conditions (fault resistances, distances, and inception angles), leading to more reliable power production systems.

Development of a plug-and-play anti-noise module for fault diagnosis of rotating machines in nuclear power plants

The health of rotating machines is crucial to the stable operation and safety of nuclear power plants. However, research on machine learning-based fault diagnosis of rotating machines in the nuclear industry is still in its infancy. The signal noise generated in the plant may negatively affect the effectiveness of the analysis. In this paper, a plug-and-play anti-noise machine learning module is proposed to fill the knowledge and capability gap. The modules are loaded into a convolutional neural network called deep residual network (ResNet) to obtain a new model with noise reduction capability. The basic idea is that the module is able to identify noise features and include them in the subsequent analysis, effectively filtering noise at the feature level of the network. Nine variants of the new model are compared with the original ResNet as well as four classical machine learning models to test the effectiveness of the module and to examine the impact of the module's loading modes on the performance of the new model. This research helps facilitate the application of machine learning in the plant noise environment.

Review of Machine Learning Based Fault Detection for Centrifugal Pump Induction Motors

Centrifugal pumps are an integral part of many industrial processes and are used extensively in water supply, sewage, heating and cooling systems. While there are several review papers on machine learning-based fault diagnosis on induction motors, its application to centrifugal pumps has received relatively little attention. This work attempts to summarize and review recent research and development in machine learning-based pump condition monitoring and fault diagnosis. The paper starts with a brief explanation of pump operation including common pump faults and the main principles of the motor current signature analysis (MCSA) method. This is followed by a detailed explanation of various machine learning-based methods including the types of detected faults, experimental details and reported accuracies. The performances of different approaches are then presented systematically in a unified table. Finally, the authors discuss practical aspects and challenges related to data collection, storage and real-world implementation.

Condition Monitoring and Feature Extraction of Stator Current Phasors for Enhanced Fault Diagnosis in AC Drive

AC drives are employed mainly in process plants for various applications. In most industrial applications, Induction motor drives are preferred as they are robust, reliable, and efficient. Process industries have seen a paradigm shift from manual control to automatic control. Advancements in power electronics technology have led to smooth control of the induction motor using variable frequency drives over an entire speed range. Variable Frequency Drives (VFD) comprises of Voltage source inverter and a three phase squirrel cage induction motor. Various electric faults that are incipient in the VFD cause an abrupt change in circuit parameters resulting in insulation damage, reduced efficiency, and leading to catastrophic failure of the entire system. Hence, continuous monitoring of the system parameters such as stator current, speed, and the vibration of the machine is essential to diagnose incipient faults in the system. AI techniques have been effectively used in the fault diagnosis of electrical systems. In the proposed work, simulation results of machine learning-based fault diagnosis techniques are presented. Real-time IoT-based condition monitoring of the Variable Frequency Drive is also implemented for enhanced fault diagnosis of various incipient electrical faults in AC drives. The experimental results obtained are validated with the simulation data.

Satellite based fault diagnosis of photovoltaic systems using recurrent neural networks

Due to manufacturing defects and wear, faults in photovoltaic (PV) systems are often unavoidable. The effects range from energy losses to risk of fire and electrical shock, making early fault detection and identification crucial. Literature focuses on PV fault diagnosis using dedicated on-site sensors or high-frequency current and voltage measurements. Although these existing techniques are accurate, they are not economical for widespread adoption, leaving many PV systems unmonitored. In contrast, we introduce a machine learning based technique that relies on satellite weather data and low-frequency inverter measurements for accurate fault diagnosis of PV systems. This allows one to adopt machine learning based fault diagnosis even for PV systems without on-site sensors. The proposed approach relies on a recurrent neural network to identify six relevant types of faults, based on the past 24h of measurements, as opposed to only taking into account the most recent measurement. Additionally, whereas state-of-the-art methods are limited to identifying the fault type, our model also estimates the output power reduction stemming from the fault, i.e., the fault severity. Comprehensive experiments on a simulated PV system demonstrate that this approach is sensitive to faults with a severity as small as 5%, reaching an accuracy of 96.9%±1.3% using exact weather data and 86.4%±2.1% using satellite weather data. Finally, we show that the model generalizes well to climates other than the climate of its training data and that the model is also able to detect unknown faults, i.e., faults that were not represented in the training data.

Low-Pass Filtering Empirical Wavelet Transform Machine Learning Based Fault Diagnosis for Combined Fault of Wind Turbines.

Fault diagnosis of wind turbines is of great importance to reduce operating and maintenance costs of wind farms. At present, most wind turbine fault diagnosis methods are focused on single faults, and the methods for combined faults usually depend on inefficient manual analysis. Filling the gap, this paper proposes a low-pass filtering empirical wavelet transform (LPFEWT) machine learning based fault diagnosis method for combined fault of wind turbines, which can identify the fault type of wind turbines simply and efficiently without human experience and with low computation costs. In this method, low-pass filtering empirical wavelet transform is proposed to extract fault features from vibration signals, LPFEWT energies are selected to be the inputs of the fault diagnosis model, a grey wolf optimizer hyperparameter tuned support vector machine (SVM) is employed for fault diagnosis. The method is verified on a wind turbine test rig that can simulate shaft misalignment and broken gear tooth faulty conditions. Compared with other models, the proposed model has superiority for this classification problem.

Wasserstein distance-based asymmetric adversarial domain adaptation in intelligent bearing fault diagnosis

Addressing the phenomenon of data sparsity in hostile working conditions, which leads to performance degradation in traditional machine learning-based fault diagnosis methods, a novel Wasserstein distance-based asymmetric adversarial domain adaptation is proposed for unsupervised domain adaptation in bearing fault diagnosis. A generative adversarial network-based loss and asymmetric mapping are integrated to alleviate the difficulty of the training process in adversarial transfer learning, especially when the domain shift is serious. Moreover, a simplified lightweight architecture is introduced to enhance the generalization and representation capability and reduce the computational cost. Experimental results show that our method not only achieves outstanding performance with sufficient data, but also outperforms these prominent adversarial methods with limited data (both source and target domain), which provides a promising approach to real industrial bearing fault diagnosis.

Machine Learning-Based Fault Diagnosis of Self-Aligning Bearings for Rotating Machinery Using Infrared Thermography

Bearings are considered as indispensable and critical components of mechanical equipment, which support the basic forces and dynamic loads. Across different condition monitoring (CM) techniques, infrared thermography (IRT) has gained the limelight due to its noncontact nature, high accuracy, and reliability. This article presents the use of IRT for the bearing fault diagnosis. A two-dimensional discrete wavelet transform (2D-DWT) has been applied for the decomposition of the thermal image. Principal component analysis (PCA) has been used for the reduction of dimensionality of extracted features, and thereafter the most relevant features are accomplished. Furthermore, support vector machine (SVM), linear discriminant analysis (LDA), and k-nearest neighbor (KNN) as the classifiers were considered for classification of faults and performance assessment. The results reveal that the SVM outperformed LDA as well as KNN. Noncontact condition monitoring shows a great potential to be implemented in determining the health of machine. The utilization of noncontact thermal imaging-based instruments has enormous potential in anticipating the maintenance and increased machine availability.

Hybrid multi-mode machine learning-based fault diagnosis strategies with application to aircraft gas turbine engines

In this work, a novel data-driven fault diagnostic framework is developed by using hybrid multi-mode machine learning strategies to monitor system health status. The coexistence of multi-mode and concurrent faults and their adverse coupling effects pose serious limitations for developing reliable diagnostic methodologies. A novel framework is proposed by exploiting inherent embedded health information contained in the I/O sensor data. The proposed hybrid strategies consist of optimal integration of recurrent neural network-based feature generation and self-organizing map diagnostic modules. To construct reliable fault diagnostic modules, a systematic clustering and modeling methodology is developed that has two primary advantages: (i) it does not require any a priori knowledge of data set characteristics or system mathematical model, and (ii) it does address and resolve the key limitations and challenges in conventional self-organizing map approaches. The effectiveness of our proposed framework is validated by utilizing sensor data including healthy and various degradation modes in application to compressor and turbine of an aircraft gas turbine engine. Comparisons with other machine learning-based methods in the literature are provided to demonstrate the performance and superiority of our proposed framework in fault diagnostic accuracy, false alarm rates, and in dealing with multi-mode and concurrent fault scenarios.

Assessment of characteristics of acoustic emission parameters for valve damage detection under varying compressor speeds

Acoustic emission technique is often employed to detect valve abnormalities. With the development of technology, machine learning-based fault diagnosis methods are prevalent in the nondestructive testing industry as they can automatically detect valve problems without any human intervention. Nevertheless, feeding in all possible input parameters into the learning algorithm without any prior assessment may result in high computational cost and time, while adding to the risk of having false alarms. This study intended to obtain characteristics of acoustic emission signal for various valve conditions and compressor speeds by examining the four most commonly used parameters, namely the acoustic emission root mean square, acoustic emission crest factor, acoustic emission variance, and acoustic emission kurtosis. The study begins with time–frequency analysis of one revolution acoustic emission signal acquired from a faulty suction valve through discrete wavelet transform to obtain the signal characteristics of valve events. To associate signals with valve movements, the reconstructed discrete wavelet transform signals are further segregated into six time segments, and the four acoustic emission parameters are computed from each of the time segments. These parameters are analyzed through statistical analysis namely the two-way analysis of variance, followed by the Tukey test to obtain the best parameter which can differentiate each valve condition clearly at all speeds. The results revealed that acoustic emission root mean square is the best parameter especially in identification of heavy grease valve condition during suction valve opening event while acoustic emission crest factor is capable to detect leaky valve during the suction valve closing event at all speeds. It is believed that effective valve diagnosis strategy can be delivered by referring to the features of parameters and the characteristic valve event timing corresponding to each valve condition and speed.

Single- and Multi-Fault Diagnosis Using Machine Learning for Variable Frequency Drive-Fed Induction Motors

In this article, an effective machine learning-based fault diagnosis method is developed for induction motors driven by variable frequency drives (VFDs). Two identical 0.25 HP induction motors under healthy, single-, and multifault conditions were tested in the lab with different VFD output frequencies and motor loadings. The stator current and the vibration signals of the motors were recorded simultaneously under steady-state for each test, and both signals are evaluated for their suitability for fault diagnosis. The signal processing technique, discrete wavelet transform, is chosen in this article to extract features for machine learning. Four families of machine learning algorithms in the MATLAB Classification Learner Toolbox, decision trees, support vector machines, k -nearest neighbors, and ensemble, with 20 classifiers are evaluated for their classification accuracy when used for fault diagnosis of induction motors fed by VFDs. To allow fault diagnosis for untested motor operating conditions, the feature calculation formulas are developed through surface fitting using experimental data of a range of tested frequencies and loadings of the motor to provide training data for untested conditions.