Fault-free Data Research Articles

An enhanced digital twin-driven fault detection and isolation method based on sensor series imaging mechanism for gas turbine engine

Condition based maintenance plays a crucial role in ensuring the safety and reliability of gas turbine engine. Specifically, fault detection and isolation (FDI) can improve efficiency and reduce costs during maintenance. The traditional FDI methods, which developed only using physical engine data, lack a comprehensive consideration of high accuracy and strong applicability across different engine individuals. In this paper, an enhanced digital twin-driven FDI method is proposed, which combines a digital engine, sensor series imaging mechanism and a convolutional neural network. The digital engine is constructed to mirror the operational status of the physical engine. The output residuals between the physical and digital engines are transformed into multi-channel images to train a convolutional neural network for FDI of the physical engine. The network is then fine-tuned using fault-free data from the target engine to further strengthen diagnostic performance across different engine individuals. The effectiveness of the digital engine is verified using actual ground test data, and the performance of the proposed method is demonstrated through comparative simulations. The network, trained on operational data from Engine A and fine-tuned with fault-free data from other engines, exhibits the highest diagnostic accuracy and fault isolation rate. Compared with the physical engine data-based method, the accuracy is increased from 68.25% to 97.95%, the fault isolation rate is improved from 85.91% to 99.26%, and the false alarm rate is decreased from 4.41% to 0.34%. The results show that the proposed method has better FDI capabilities across different engine individuals.

Voltage fault diagnosis and prognostic of lithium-ion batteries in electric scooters based on hybrid neural network and multiple thresholds

Voltage fault diagnosis and prognostics of lithium-ion batteries (LIBs) are critical in ensuring their safe and reliable operation, and timely voltage prediction can help detect the oncoming fault of LIBs effectively. To this end, this study develops an efficient voltage fault diagnosis and prognostic method for LIBs based on voltage prediction and multiple thresholds. Firstly, a hybrid neural network integrating convolutional neural network and gated recurrent unit is established to extract both temporal and spatial features from the current operation data and predict voltage. Then, a residual calculator is devised to generate the voltage residuals, and the maximum voltage residuals of the fault-free data during charge and discharge operations are leveraged to constitute the multiple thresholds of fault diagnosis and prognostics strategy. The robustness, reliability and feasibility of voltage prediction are validated under the conditions of different driving scenarios, different dynamic temperature and real-world operation data. Moreover, the validation results on the fault electric scooter reveals that the proposed method can reliably diagnose the battery over-voltage and under-voltage faults and their alarm levels. Additionally, the warning moment by this method is at least 630s earlier than the alarm time in the battery management system, highlighting its efficient prognostic functionality.

Towards domain shifts: Stream fine-tuning via feed-forward fault data generation for on-board aero-engine gas-path diagnosis

On-board gas-path diagnosis is integral to condition-based maintenance, with the false alarm rate as a primary metric. Engine variations from manufacturing and performance deterioration lead to a domain shift, increasing the false alarm rate. Consequently, stream fine-tuning is imperative to align diagnostic methods with the shifting domain. However, using only fault-free data for fine-tuning leads to class imbalance, reducing fault detection capability. This paper introduces a method to mitigate class imbalance in fine-tuning and handle ongoing domain shifts. It involves inserting static fault features into new data, generating artificial fault data. This artificial fault data, along with actual fault-free data, are used for stream fine-tuning. An ablation study involving a 20-hour simulation and tests with actual engine data confirms the method’s effectiveness in maintaining low false alarm rates and minimizing diagnostic omissions under continuous domain shifts.

Automatic fault detection in grid-connected photovoltaic systems via variational autoencoder-based monitoring

Anomaly detection is indispensable for ensuring the reliable operation of grid-connected photovoltaic (PV) systems. This study introduces a semi-supervised deep learning approach for fault detection in such systems. The method leverages a variational autoencoder (VAE) to extract features and identify anomalies. By training the VAE on normal operation data, a compact latent space representation is created. Abnormal observations, indicating faults, exhibit distinct feature vectors in this latent space. Multiple anomaly detection algorithms, including Isolation Forest, Epileptic Envelope, Local Outlier Factor, and One-Class SVM, are employed to discern normal and abnormal observations. This semi-supervised approach only requires fault-free data for training, without labeled faults, making it attractive in practice. A publicly available dataset, the Grid-connected PV System Faults (GPVS-Faults) dataset, which includes data from a PV plant operating in both maximum power point tracking (MPPT) and intermediate power point tracking (IPPT) switching modes, is used for evaluation. The proposed approach is assessed across various fault scenarios, such as partial shading, inverter faults, and MPPT/IPPT controller faults in boost converters. The outcomes underscore the effectiveness of VAE-based techniques in accurately identifying these faults, with accuracy rates reaching up to 92.90% for MPPT mode and 92.99% for IPPT mode, thus contributing to the robustness of fault detection in grid-connected PV systems.

A fast data-driven fault detection and location method for unknown distributed thermal processes

The distributed thermal processes are widely present in various industrial operations, but the presence of time delay effect and unknown model information make achieving fast and timely fault detection challenging. Furthermore, the involvement of spatiotemporal coupled data in measurement processes further exacerbates the situation. To solve these problems, this paper proposes a data-driven fault detection and location method for thermal processes described by distributed parameter systems (DPS). Firstly, a Time/Space separation is employed to decoupling the spatio-temporal data into spatial basis functions and temporal coefficients. Subsequently, dynamic partial least squares (DPLS) is employed to obtain the temporal information of the model with fault-free data. Monitoring statistics in both the temporal and spatial domains are then constructed, and their thresholds are estimated by kernel density function. Finally, an online strategy for fault detection and location is presented. The method was validated on the catalytic rod, and an experimental snap curing oven.

Addressing class-imbalanced learning in real-time aero-engine gas-path fault diagnosis via feature filtering and mapping

Condition-based maintenance of aero-engines requires real-time gas-path fault diagnosis, which is crucial for reducing costs and enhancing aircraft attendance. It is imperative to ensure a low false alarm rate in fault diagnosis algorithms. Inherent variations between engines and age-related performance deterioration can lead to an increase in false alarms, necessitating online algorithm tuning. However, insufficient fault data in online tuning may lead to class imbalance and algorithmic forgetfulness of fault features, compromising effectiveness. To tackle this issue, this paper introduces a novel hybrid diagnosis algorithm that utilizes feature filtering and mapping. This method entails an engine model, a feature extractor, a feature filter, a fault classifier, and several feature mappers. The feature mappers are designed to generate fault features based on the incoming fault free data for rebalancing the tuning dataset. And the feature filter is introduced to prevent weight collapse during the online tuning. Validation through a 400-hour life cycle simulation test and a real data test indicate that the method outperforms the benchmark methods. It overcomes the class-imbalanced learning problem, showcasing its potential for aero-engine gas-path fault diagnosis.

ANFIS and Takagi–Sugeno interval observers for fault diagnosis in bioprocess system

This paper develops a data-driven approach for incipient fault diagnosis based on ANFIS and Takagi–Sugeno (TS) interval observers. First, the nonlinear bioreactor system is identified using an adaptive neuro-fuzzy inference system (ANFIS), which results in a set of polytopic TS models derived from measurement data. Second, a bank of TS interval observers is deployed to detect sensor and process faults using adaptive thresholds. Unlike other works that require training fault data, only fault-free data is considered for ANFIS learning in this work. Fault insolation is based on residual generation and evaluated on a fault signal matrix (FSM). Parametric uncertainty and measurement noise are considered to guarantee the method’s robustness. The effectiveness of the proposed method is tested on a well-known bioreactor Continuous stirred tank reactor system (CSTR) reference simulator.

Transformer encoder based self-supervised learning for HVAC fault detection with unlabeled data

Data driven methods are the most studied fault detection and diagnostics (FDD) type in buildings HVAC systems. However, most studies rely on labeled data for specific faults which are hard to find and collect for real systems. While the fault-free data is easier to collect, it is still time consuming to label for large systems operation. Moreover, most of the studies rely on the usage of supervised learning algorithms which do not generalize well beyond the training data making unseen faults hard to detect. In this paper, we define a methodology to use a self-supervised learning method for HVAC systems' FDD using a Transformer encoder, moreover, we tested it on a real case study. By strategically masking portions of the multivariate time-series data using Markov chain approach with two states. The model is trained by predicting these concealed segments. This approach, independent of labeled data, offers a scalable solution for practical HVAC applications. Anomalies are labeled using the Peak Over Threshold (POT) method, which dynamically determines thresholds by fitting reconstruction errors to a generalized Pareto distribution. Subsequent fault diagnostics emphasize features with pronounced reconstruction errors, pinpointing potential HVAC malfunctions. This methodology reduces dependence on labeled datasets and augments the model's generalization, facilitating detection of unobserved faults. This approach was applied to data from a real building. As a results multiple faults were detected mainly due to the malfunctioning of the monitoring system. The model demonstrates the ability to detect both sequential and individual faults. The period from October 19th to December 23rd was detected as a fault period due to the change in the trend of the data because of the monitoring system.

Fault detection and diagnosis of electric bus air conditioning systems incorporating domain knowledge and probabilistic artificial intelligence

The air conditioning systems in electric city buses usually operate in rapidly changing ambient conditions and are more likely to suffer from mechanical faults. Although many fault detection and diagnosis (FDD) methods have been developed for building air conditioning systems, they are difficult to be applied to bus air conditioners since its operation is highly dynamic and fault-free data are usually unavailable. Therefore, this paper proposes an FDD method for electric bus air conditioners to tackle the above issues. First, the method identifies faults in an unsupervised manner by comparing selected features among a group of peer systems. Then, considering the features are influenced by the operating conditions, Gaussian process regression (GPR) models are established to find the relationships between each feature and its influential parameters. The probabilistic nature of the GPR is used to differentiate predictions with large uncertainty, which are then excluded from FDD. In this way, robustness of the method is evidently improved. Finally, fault indexes are defined to detect and diagnose mechanical faults. We applied the method to a group of air conditioners in a city bus fleet. Results showed that it can effectively identify refrigerant undercharge and indoor and outdoor fan problems with low false positive/genitive rates. Also, the method is highly robust and not sensitive to the faulty systems in the bus fleet.

A Risk Evaluation Framework in System Control Subject to Sensor Degradation and Failure.

Sensor degradation and failure often undermine users' confidence in adopting a new data-driven decision-making model, especially in risk-sensitive scenarios. A risk assessment framework tailored to classification algorithms is introduced to evaluate the decision-making risks arising from sensor degradation and failures in such scenarios. The framework encompasses various steps, including on-site fault-free data collection, sensor failure data collection, fault data generation, simulated data-driven decision-making, risk identification, quantitative risk assessment, and risk prediction. Leveraging this risk assessment framework, users can evaluate the potential risks of decision errors under the current data collection status. Before model adoption, ranking risk sensitivity to sensor data provides a basis for optimizing data collection. During the use of decision algorithms, considering the expected lifespan of sensors enables the prediction of potential risks the system might face, offering comprehensive information for sensor maintenance. This method has been validated through a case study involving an access control.

Voltage measurement-based recursive adaptive method for internal short circuit fault diagnosis in lithium-ion battery packs

Internal short circuit (ISC) has been identified as a major cause of thermal runaway in lithium-ion (Li-ion) battery systems, making the investigation of ISC fault diagnosis a focal research topic in electric vehicles and battery energy storage systems. Recently, several studies have applied multivariate analysis techniques to analyze abnormal behavior in cell voltage measurements and achieved an accurate diagnosis of ISC faults. However, model parameter variations under different operating conditions and state of charge (SOC), which results in unacceptable false alarm rates, is yet to be fully considered. To effectively address this problem, this study introduces a recursive approach on the basis of principal component analysis (PCA) to update the monitoring model and adapt to parameter changes. Moreover, an adaptive threshold with fusion factors is devised to reduce fault false alarm rates. In particular, this study employs sliding window techniques to amplify weak fault features to tackle the challenge of early-stage detection difficulty for ISC faults and provides an optimal window width determination algorithm based on fault-free data. Experimental results on a real Li-ion battery pack test platform under various operating conditions and SOC levels demonstrate that the proposed method maintains extremely high fault detection rates for ISC faults of different intensities. Furthermore, it is established that the adaptive updating of the model in the absence of faults greatly reduces false alarm rates.

Graph attention network with Granger causality map for fault detection and root cause diagnosis

Unsupervised data-driven methods are widely used for fault detection and diagnosis in modern industrial processes. However, accurately distinguishing faults from normal feedback control system adjustments and promptly identifying their root causes are among unresolved challenges. To address these issues, we propose a neural network model consisting of one-dimensional convolutional neural networks and a graph attention network (CNN-GAT) that uses a causal map derived from fault-free data using conditional Granger causality analysis. The CNN-GAT model produces a monitoring index that accurately reflects the operating conditions of the process and distinguishes faults from normal control adjustments. Using the causal map and prediction results from the CNN-GAT model, the root cause diagnosis can be performed promptly after faults are detected, providing operators with more time to address the fault. We demonstrate the performance of the proposed framework using the benchmark Tennessee Eastman process case studies and through comparison with other fault detection methods.

Fault detection of a reciprocating plunger pump with fault-free data based on an unsupervised feature encoder and minimum covariance determinant

It is well known that complex mechanical equipment has many differnt the failure modes, and monitoring data for fault conditions are scarce. Therefore, research on fault detection for reciprocating plunger pumps with fault-free data is significant for theory and application. Due to the lack of prior knowledge of faults, it is always a great challenge for researchers to extract fault features from signals. In this paper, an innovative fault detection method for a reciprocating plunger pump is proposed based on an unsupervised feature encoder (UFE) and minimum covariance determination (MCD). Firstly, a criterion based on mutual information maximization for local and global features is proposed for unsupervised feature extraction. In addition, an unsupervised training strategy based on negative sampling is proposed to train the encoder, so that the model can converge quickly and improve the stability of model training. A fault detection algorithm for a reciprocating plunger pump is proposed based on a UFE and MCD. Finally, the effectiveness and superiority of the proposed method are verified with the measured data for a reciprocating plunger pump. The results show that the proposed method can accurately detect the faults in the reciprocating plunger pump with a detection accuracy of more than 98%. Compared with other methods, the proposed fault detection algorithm has better applicability and accuracy for fault detection with fault-free data.

Deep learning GAN-based data generation and fault diagnosis in the data center HVAC system

The automated fault diagnosis and smart management of heating, ventilation and air conditioning (HVAC) system is essential to the reliability of data centers. The machine learning is an efficient way to develop the smart management tools for HVAC system using historical data. In real data centers, however, the quantity of fault-free data is much more than that of fault data. As a result, the imbalanced training data limits the diagnosis capacity of machine learning models. The deep learning-based generative adversarial network is proposed to integrate with an incremental learning SVM model to diagnose the commonly occurred faults of data center air conditioning system. The adversarial learning between generator and discriminator generates the data of minority class for training purpose in HVAC system. The 8 sensitive features are selected as the inputs through three-step optimal selection strategy: manual screening, correlation feature selection and redundancy feature removing. The incremental learning strategy is proposed to update the FDD model regularly. The refrigerant leakage faults with intensities of 10%, 20%, 30% and 40% are tested and validated under various operational conditions. The experimental results show that the incremental learning SVM integrated with deep learning GAN reaches the acceptable diagnosis accuracies.

Fault detection and isolation in wind turbines based on neuro-fuzzy qLPV zonotopic observers

This article develops a hybrid approach to fault detection and isolation (FDI) based on a machine learning technique and quasi-Linear Parameter Varying (qLPV) zonotopic observers. First, the dynamical model of a wind turbine is identified using an adaptive network-based fuzzy inference system (ANFIS), which results in a set of qLPV polytopic models whose form is derived using structural analysis. Second, a bank of qLPV zonotopic observers is implemented to detect sensor and actuator faults. Unlike other works that consider different fault scenarios to train a neuronal network, in this work, only fault-free data is considered for the ANFIS. The FDI is based on the residual generation obtained by a bank of qLPV zonotopic observers of the identified models. Disturbances related to aerodynamic loads and measurement noise are considered to guarantee the robustness of the proposed method. The effectiveness of the proposed method is tested in a 5MW WT well-known benchmark simulator based on fatigue, aerodynamics, structures, and turbulence under different fault scenarios.

Neuro-fuzzy Takagi Sugeno observer for fault diagnosis in wind turbines

This work proposes a method for fault diagnosis based on Takagi Sugeno (TS) observers and convex models identified with a multioutput adaptive neuro-fuzzy inference system (MANFIS) derived from structural analysis. A bank of zonotopic TS observers is implemented to detect sensors and actuators faults. Unlike other works that require data from fault scenarios to train the MANFIS neural network, only fault-free data are considered. In addition, uncertainty related to aerodynamic loads and measurement noise is considered for testing the proposed method's robustness. The method performance is evaluated using measurements from a 5 MW wind turbine benchmark.

An Approach to Oscillatory Failure Case Detection in the Aerospace Industrial Benchmark

This paper proposes an approach to detect the oscillatory failure case (OFC) and reconfigure the flight control law after the OFC detection in the Aerospace Industrial Benchmark. To cope with system nonlinearity, a square root unscented Kalman filter is utilized to generate the residual signal. An estimation of the parameter uncertainty based on the fault-free data is incorporated to reduce the influence of model uncertainty on the residual signal significantly. To evaluate the residual signal, a modified integral absolute error based index is introduced along side a check of the oscillation regularity. Our approach is capable of detecting OFCs with very low amplitudes within at most 3 oscillation periods without producing any false alarm, regardless of load factor command, turbulence level or OFC location, type and frequency. The minimum detectable OFC amplitude is 0.6 mm and 0.9 mA for liquid OFCs located at sensor signal and current input signal, respectively, and 0 mm and 0.9 mA for solid OFCs.

Fault diagnosis in wind turbines based on ANFIS and Takagi–Sugeno interval observers

Wind turbine power generation is becoming one of the most critical renewable energy sources. As wind power grows, there is a need for better monitoring and diagnostic strategies to maximize energy production and increase its security. In this paper, a fault diagnosis approach based on a data-driven technique, which represents the system behavior employing a Takagi–Sugeno (TS) model, is developed. An adaptive neuro-fuzzy inference system (ANFIS) method is used to obtain a set of polytopic-based linear representations and a set of membership functions to interpolate the linear models of the convex TS model. Then, considering the TS model, a fault diagnosis strategy based on convex state observers generate residuals to detect and isolate sensor faults. Unlike other methods, this proposal only needs to be trained with fault-free data. The proposed methodology is tested under different fault scenarios on a well-known wind turbine benchmark built upon fatigue, aerodynamics, structures, and turbulence (FAST). The results demonstrate the method’s effectiveness in detecting and isolating different sensor faults.

Kantorovich Distance Based Fault Detection Scheme for Non-Linear Processes

Fault detection is necessary for safe operation in modern process plants. The kernel principal component analysis (KPCA) technique has been widely utilized for monitoring non-linear processes because it enhances dimension reduction and fault detection in non-linear space. In this paper, an improved non-linear fault detection strategy based on Kantorovich distance (KD) and kernel principal component analysis is proposed. The KD statistic is based on the optimal mass transport theory where the distance between two distributions is computed with respect to a cost function. The addressed fault detection problem models the data using the KPCA framework and utilizes the ability of the KD statistical indicator to detect faults. The detection stage involves comparing the residuals of training fault-free data and testing faulty data using the KD statistic. Additionally, the reference threshold for the KD statistic is computed using the kernel density estimation (KDE) approach as compared to the previously utilized three-sigma rule approach. The detection performance is illustrated with the help of three benchmark case studies: a continuous stirred tank reactor (CSTR) process, Tennessee Eastman (TE) process and an experimental distillation column process. The performance analysis suggests the superiority of the KPCA-KD fault detection scheme in monitoring various sensor faults. Moreover, comparison with traditional statistical indicators of PCA and KPCA schemes shows that the proposed scheme enhances fault detection and achieves an improved detection rate in monitoring different categories of faults.

Fault Detection With LSTM-Based Variational Autoencoder for Maritime Components

Maintenance routines on ships today follow either a reactive maintenance (RM) or preventive maintenance (PvM) approach. RM can be regarded as post-failure repair, which might create large costs. PvM uses predetermined maintenance intervals, which often involves unnecessary maintenance. Recently, prognostics and health management (PHM) has emerged as a potential way to develop an ideal maintenance policy. PHM aims to provide optimal maintenance schedule through the use of sensor measurement for fault detection and fault prognostics, among which fault detection is the first and fundamental action. In this paper, a long-short term memory based variational autoencoder (LSTM-VAE) is proposed for fault detection of maritime components onboard. It is a semi-supervised approach that requires only fault-free data for training. Therefore, it is widely applicable in the maritime industry since operational data in normal conditions already exists. Real-world operation data collected from a diesel engine on the research vessel (RV) Gunnerus is used to validate the method. Results show that the LSTM-VAE can detect the fault accurately.