Extracting Power Transformer Vibration Features by a Time-Scale-Frequency Analysis Method

The secure and reliable operation of power transformer is of significance to the construction of new power system and the national target of "dual carbon". Acoustic signals are important for the condition monitoring of transformer. In view of the difficulties to obtain abundant data samples of acoustic signals for the recognition of different failures of transformer, this paper proposes an online condition monitoring method for power transformer by the deep Q-network. With the input of the gammatone frequency cepstral coefficients (GFCCs) and the interaction of Q-learning strategy with the environment, the deep Q-network (DQN) model is built to recognize the operation condition of transformer by properly selecting the hyper-parameters of the framework. The acoustic signals of a 500kV power transformer during the running process are measured and analyzed. It is seen that the designed DQN model converges fast with high learning performance and exploring ability and high recognition rate.

This article presents evolving neural nets (ENNs) for protection and condition monitoring of power transformer. Based on the proposed evolutionary algorithm, the ENNs automatically tune the network parameters (connection weights and bias terms) of the neural nets to achieve the best model. The ENNs can identify, classify and detect the fault and issue the trip signal in the case of internal fault only, using the global search capabilities of the evolutionary algorithm and the highly nonlinear mapping nature of the neural nets. The proposed protection scheme has been realized through two different structures using ENNs algorithm. This scheme has been evaluated using simulated data obtained through EMTP/ATP package. The results amply demonstrate the capabilities of fault detector (FD) and condition monitoring (CM) in terms of accuracy and speed with respect to detection of fault, classification and pattern recognition of different event of power transformer.

Vibrations signals of an operated transformer are mainly originated from winding vibration and core vibration. By monitoring the vibration signals in transformer tank via on-line vibration monitoring system, it is possible to capture the incipient mechanical failures at all time while transformer is working. However, a large amount of vibration signals of transformer would produce in the on-line vibration monitoring system, which is a heavy burden for data storage and transmission in the construction of Ubiquitous Electric Internet of Things (UEIOF). To solve this problem, this paper presents a method to compress the vibration signals of transformer according to the compressive sensing theory when considered the spare features of vibration signals. The sparse random matrix, the dictionary and the reconstruction algorithm are built and trained to compress and restore the vibration signals of power transformer. The calculated results of a 35kV oil-immersed transformer in load experiment have shown that the original vibration signals are compressed to 1/8 with the average reconstruction error of 0.06% by the proposed method.

Condition monitoring of power transformers is a significant issue for electrical utilities. Research has demonstrated the efficacy of employing ultra high frequency measurement of partial discharge in the monitoring of transformers. A condition monitoring architecture is described that supports the capture and interpretation of diagnostic data, and provides engineers with meaningful diagnostic advice using intelligent system technologies.

A power transformer outage has a dramatic financial consequence not only for electric power systems utilities but also for interconnected customers. The service reliability of this important asset largely depends upon the condition of the oil-paper insulation. Therefore, by keeping the qualities of oil-paper insulation system in pristine condition, the maintenance planners can reduce the decline rate of internal faults. Accurate diagnostic methods for analyzing the condition of transformers are therefore essential. Currently, there are various electrical and physicochemical diagnostic techniques available for insulation condition monitoring of power transformers. This paper is aimed at the description, analysis and interpretation of modern physicochemical diagnostics techniques for assessing insulation condition in aged transformers. Since fields and laboratory experiences have shown that transformer oil contains about 70% of diagnostic information, the physicochemical analyses of oil samples can therefore be extremely useful in monitoring the condition of power transformers.

This paper provides the information on the diagnostics techniques for condition monitoring of power transformer (PT). This paper deals with the practical importance of the transformer diagnostic in the Electrical Engineering field. A review has been made on the modern development of this practical technology.

ABSTRACTAmong the most expensive assets in power grids, power transformers are essential for the reliability of the power supply chain and the overall stability of the grid. Due to their permanent connection to the network, this equipment is exposed to all kinds of faults and phenomena, including short‐circuit faults and overvoltages caused by lightning and switching. Hence, ongoing monitoring of the transformer's condition is essential to prevent breakdowns and damage to the transformer. Among the different condition monitoring methods, the frequency response analysis (FRA) method is sensitive to the smallest functional changes of the transformer, as it is completely related to the physics and geometry of the transformer. This method stands out as one of the most effective and efficient approaches to transformer monitoring, especially for detecting mechanical faults. However, the FRA method faces an important challenge of interpretation: the correlation between the type of fault that occurred and the way the transformer's function changes is still not well‐known, and studies in this field are ongoing. One of the most widely used methods of interpreting frequency response results is the use of numerical indices, coil modelling, transformer function estimation, and artificial intelligence algorithms. This paper introduces these methods, and their advantages and disadvantages are discussed. Then, the most widely used artificial intelligence algorithms in transformer condition monitoring are presented and compared. Finally, future research directions are anticipated.

Acoustic emission has been applied in identifying evolving failures of electrical, mechanical or thermal nature in power transformers and reactors. Different kinds of failures can be identified by the characteristics of the signals received and also give the location of the source of the signals. Also, operating sequence and contact timing of on-load tap changers can be evaluated with the aid of acoustic emission.The technique has the advantage of being utilized with in-service, energized equipments, avoiding downtime to perform tests. Case studies of acoustic emission tests performed to condition monitoring of several high-voltage power transformers and reactors, rated 230 and 500 kV are presented. Results have demonstrated the effectiveness of the acoustic emission technique as a tool for the condition diagnostic of electric power equipments.KeywordsAcoustic emissionPartial dischargesPower transformersShunt reactors

The power transformer plays a vital role in the power system network. Nowadays, the condition monitoring of oil-filled power transformer is greatest requirement for its reliable operations and quality power supply to the utility. The condition of such transformer can be indeed determined by knowing the ageing status of the dielectric liquid used for insulation and cooling. Though Dissolved Gas Analysis (DGA) has been well-acclaimed in the field of the transformer fault diagnosis, it has certain drawbacks like need of carrier gas, regular calibration. To overcome these drawbacks, in this proposed work, optical methods such as Ultra Violet-Visible (UV-vis) spectroscopy and Nuclear Magnetic Resonance (NMR) have been used. The obtain results of UV-vis spectroscopy demonstrate that all the characteristic peaks are covered and qualitatively analyzed within the wavelength region of 200–400 nm. Further, NMR spectroscopy results precisely identified the molecular groups present in the oil such as the methyl group, methylene group and the aromatic hydrocarbon groups. Moreover, this method is fast and reliable for determining the moisture content in the transformer oil. Overall, the obtained data confirms that UV-visible and NMR spectroscopy methods are highly efficient for analyzing the degraded transformer oils.

The condition monitoring and fault diagnosis of power transformers plays a significant role in the safe, stable and reliable operation of the whole power system. Dissolved gas analysis (DGA) methods are widely used for fault diagnosis, however, their accuracy is limited by the selection of DGA features and the performance of fault diagnosis models, for example, the classical support vector machine (SVM), is easily affected by unbalanced training samples. This paper presents a transformer fault diagnosis model based on chemical reaction optimization and a twin support vector machine. Twin support vector machines (TWSVMs) are used as classifiers for solving problems involving unbalanced and insufficient samples. Restricted Boltzmann machines (RBMs) are used for data preprocessing to ensure the effective identification of feature parameters and improve the efficiency and accuracy of fault diagnosis. The chemical reaction optimization (CRO) algorithm is used to optimize TWSVM parameters to select the optimal training parameters. The cross-validation (CV) method is used to ensure the reliability and generalization ability of the diagnostic model. Finally, the validity of the model is verified using real fault samples and random testing.

The generation, propagation and distribution of power transformer vibration and noise are studied by modeling and simulation. Based on the theoretical analysis of the mechanism of power transformer vibration and noise, the correlation between the characteristics of power transformer vibration noise and the influencing factors is determined. The solid model of single-phase power transformer is established by using multi physical field finite element analysis software. The multi physical field coupling simulation of electromagnetic field, structural force field and pressure acoustic field is carried out. The spatial distribution and frequency characteristics of vibration and noise signal of single-phase power transformer in each solution domain are displayed. It provides theoretical and technical support for smart grid equipment detection and environmental noise comprehensive control.

The condition of locomotive bearings, which are essential components in trains, is crucial to train safety. The morphologic analysis of the locomotive bearing vibration signal, which are always contaminated by certain types of noise, is very important standard for mechanical condition diagnosis of the locomotive bearing and mechanical failure phenomenon. In this paper a novel vibration signal enhancement method based on empirical mode decomposition (EMD) and adaptive filtering is proposed to filter out Gaussian noise contained in raw vibration signal. The reference signal of the adaptive filter is produced by selective reconstruction of the decomposition results of EMD. Real vibration signals from the locomotive bearing are used to validate the performance of the proposed method. Conventional EMD and adaptive EMD are tested to compare the filtering performance. The results of simulation show that the vibration signal can be significantly enhanced by using the proposed method. Besides, the proposed method is used to analyze real acoustic signals of locomotive bearing with inner race and outer race faults, respectively. The results confirm that the periods between the transients, which represent locomotive bearing fault characteristics, can be detected successfully.

In this article, a novel technique that aims at improving the signal recognition effectiveness and real-time performance of distributed optical fiber sensing vibration systems is reported. To this end, a procedure is developed based on the improved Mel-frequency cepstral coefficient (IMFCC), compensation distance estimation technology (CDET), and multiclassification support vector machine (SVM). The IMFCC is first used in the feature extraction stage. The algorithm optimizes the Mel filter according to the energy ratio to extract the characteristics of the optical fiber vibration signal. Then, the proposed method seeks to formulate a feature optimization strategy by calculating the ratio of the feature inter-class distance to the intraclass distance. The CDET removes redundant data in feature extraction and reduces feature dimension. The performance of the proposed method is tested on an experimental example of the optical fiber sensing vibration system. Moreover, the proposed method is compared against a MFCC-based method. The results demonstrate the superiority of the proposed method in long-term condition monitoring. It improves the classification effectiveness of optical fiber vibration signals and improves the real-time performance through the feature dimension reduction method, which can be widely used in optical fiber vibration signal recognition systems.

Understanding the vibrational characteristics of power transformers is significantly important in their design and monitoring. In this contribution, a model with a multi-physics coupling simulation of the electrical circuit, magnetic field, and solid mechanics is developed to investigate the characteristics of the transformer vibration. After describing the model, the harmonic contents of the vibration signals and their variation in the case of mechanical faults are studied. It is shown that under normal operating conditions, the fundamental vibration frequency of 100 Hz has the maximum amplitude, while in the case of mechanical faults, the amplitudes of 200 Hz and 300 Hz harmonics increase dramatically compared to the fundamental harmonic. The influence of vibration sensor position is investigated too, which indicates that the area near the tank bottom is the best position to gather vibration signals. Moreover, the mechanical resonance frequencies of the transformer, along with their mode shapes, are addressed in this paper. Finally, the influence of mechanical changes on the vibration energy distribution in the tank surface is explored. The results of the paper suggest possible diagnosis methods for condition monitoring of transformers, such as using the vibration energy distribution on the tank surface or analyzing the vibration harmonics.

The industry is moving towards maintenance strategies that consider component health, which require extensive collection and analysis of data. Condition monitoring methods that require manual feature extraction and analysis, become infeasible on an industrial scale. Machine learning algorithms can be used to automatically detect and classify faults, however, obtaining sufficient data for training is required for deep learning and other data-driven classification approaches. Data from healthy machine operation is generally available in abundance, while data from representative fault- and operating conditions is limited. This limits both development and deployment of deep learning-based CM systems on an industrial scale. This paper addresses both the challenges of automated analysis and lack of training data. A deep learning classifier architecture utilizing 1-dimensional dilated convolutions is proposed. Dilation of the convolution kernel allows for analysis of raw vibration signals while simultaneously maintaining the receptive field of the classifier enough to capture temporal patterns. The proposed method performs classification in time domain on signal segments of 1 second or shorter. With knowledge of the bearing specification, artificial vibration signals with similar characteristics as an actual bearing fault can be created. In this work, generated fault signals are combined with healthy operational data to obtaintraining data for a deep classifier. Parameters of the vibration model is chosen as distributions rather than fixed values. By using a range parameters in the vibration model, the classifier learns to recognize temporal features from the training data that generalize to unseen data. The effectiveness of the proposed method is demonstrated by training classifiers on generated data and testing on real signals from faulty bearingsat both low and high speed. One dataset containing seeded faults and three run-to-failure tests are used for the demonstration.