Root cause analysis improved with machine learning for failure analysis in power transformers

Velásquez, Ricardo Manuel Arias; Lara, Jennifer Vanessa Mejía

doi:10.1016/j.engfailanal.2020.104684

Cited by 38 publications

(9 citation statements)

References 30 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Yan et al (2019) combine BP neural network with improved Adaboost algorithm, then combined with PNN neural network to form a series of diagnostic models for transformer faults, and finally combined with dissolved gas in oil analysis for transformer fault diagnosis. Velásquez and Lara (2020) propose a new method with the lowest computational cost, using a genetic algorithm to optimize the ANN classifier, which is used to classify faults, replacing the traditional reinforcement learning (RL) action selection process with a genetic algorithmbased optimizer. Wang et al (2016) establish a combination of intelligent methods for transformer fault diagnosis evaluation and neural network case inference based on a knowledge base and an oil chromatography fault diagnosis case base.…”

Section: Neural Networkmentioning

confidence: 99%

“…Intelligent techniques help to resolve the uncertainty of traditional DGA methods due to boundary problems and unresolved codes or multi-fault scenarios (Wani et al, 2021). Researchers have applied many artificial intelligence techniques to DGA fault diagnosis, such as neural networks (Duan and Liu, 2011;Wang et al, 2016;Qi et al, 2019;Yan et al, 2019;Yang et al, 2019Yang et al, , 2020Luo et al, 2020;Velásquez and Lara, 2020;Mi et al, 2021;Taha et al, 2021;Zhou et al, 2021), support vector machine (SVM) (Wang and Zhang, 2017;Fang et al, 2018;Huang et al, 2018;Illias and Liang, 2018;Kari et al, 2018;Kim et al, 2019;Zeng et al, 2019;Zhang et al, 2019;Zhang Y. et al, 2020;Benmahamed et al, 2021), and clustering (Islam et al, 2017;Misbahulmunir et al, 2020). These techniques involve statistical machine learning, deep learning, etc.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Fault diagnosis of transformer using artificial intelligence: A review

et al. 2022

View full text Add to dashboard Cite

Transformer is one of the important components of the power system, capable of transmitting and distributing the electricity generated by renewable energy sources. Dissolved Gas Analysis (DGA) is one of the effective techniques to diagnose early faults in oil-immersed transformers. It correlates the concentration and ratio of dissolved gases with transformer faults. Researchers have proposed many methods for fault diagnosis, such as double ratio method, Rogers method, Duval triangle method, etc., but all of them have some problems. Based on the strong data mining capability and good robustness of AI techniques, many researchers introduced AI techniques to mine the features of DGA data. According to the characteristics and scale of DGA data, researchers select appropriate AI techniques or make appropriate improvements to AI techniques to improve diagnostic performance. This paper presents a systematic review of the literature on the application of artificial intelligence techniques for DGA-based diagnosis and for solving intractable problems in early transformer fault diagnosis, which include neural networks, clustering, support vector machines, etc. In addition to reviewing the applications of these intelligent techniques, the diagnostic thinking proposed in this literature, such as the introduction of temporal parameters for comprehensive analysis of DGA data and the extraction of optimal features for DGA data, is also reviewed. Finally, this paper summarizes and prospects the artificial intelligence techniques applied by researchers in transformer fault diagnosis.

show abstract

Section: Neural Networkmentioning

confidence: 99%

mentioning

confidence: 99%

Fault diagnosis of transformer using artificial intelligence: A review

et al. 2022

View full text Add to dashboard Cite

show abstract

“…This is perhaps why some recent research developments in hybrid ML and statistics have focussed on enabling offline and online failure diagnosis for wear loss in complex components. The application of ML techniques such as Gaussian mixture regression (GMR) support vector machines (SVMs), whereas multiple linear regression (MLR), Gaussian process regression (GPR), and artificial neural networks (ANNs) [31], [34], 46], offer reasonable classification and separation of individual operating conditions during faults diagnosis; however, they have also been criticised for providing limited information about the physics of FMs, which may hinder their ability to achieve comprehensive asset management decisions [58], [61]. Specifically, both GMR and GPR require a large number of datasets and incur high computational costs for accurate and routine failure diagnosis; MLR lacks the consistency to always connect causal inputs to failure outputs [62]; ANN suffers from inconsistent neural network weight allocation values, which in turn produces inconsistent outputs [63]; and SVM uses data subsets (i.e., smaller datasets) that may lead to inaccuracies [64].…”

Section: A Revisiting Hybrid Systems For Failure Managementmentioning

confidence: 99%

Composite Hybrid Framework for Through-Life Multi-Objective Failure Analysis and Optimisation

Appoh

Yunusa‐Kaltungo

2021

IEEE Access

View full text Add to dashboard Cite

Complex engineering systems include several subsystems that interact in a stochastic and multifaceted manner with multiple failure modes (FMs). The dynamic nature of FMs introduces uncertainties that negatively impact the reliability, risk, and maintenance of complex systems. Traditional approaches of adopting standalone techniques for managing FMs independently at various stages of the asset life cycle pose challenges related to utilisation, costs, availability, and in some cases, accidents. Therefore, this paper proposes a composite hybrid framework comprising four independent hybrid models for comprehensive through-life failure management and optimisation. The first hybrid model entails failure mode, effects, and criticality analysis (FMECA) and fault tree analysis (FTA) to identify critical FMs and overall subsystem failure rates. The second hybrid model analyses FMs caused by multiple subsystems using hybrid dynamic Bayesian discretisation. The third hybrid model adopts a hybrid Gaussian process regression machine learning technique to evaluate wear loss. The fourth hybrid model evaluates the overall risk using a Bayesian factorisation and elimination method based on multiple failure causes. Finally, a decision-making step is used to evaluate the results of the previous four steps to decide an appropriate maintenance strategy. The proposed method is verified through a case study of a UK-based train operator's pantograph system. The results show that the maintenance inspection intervals and strategy obtained using the proposed framework strike a good balance between safety and fleet availability.

show abstract

“…By connecting BP-Adaboost in series with PNN, it not only improves the defect of BP-Adaboost algorithm which does not diagnose samples, but also improves the defect of PNN model which has low diagnostic accuracy. (Arias Velásquez and Mejía Lara, 2020) proposed a new method with the lowest computational cost, using genetic algorithm to optimize ANN classifier, which was used to classify faults with genetic algorithm-based optimizer instead of the traditional RL action selection process. However, the DGA method is often limited to fault classification of transformer fault states.…”

Section: Introductionmentioning

confidence: 99%

Research on Variable Weight Synthesizing Model for Transformer Condition Assessment

et al. 2022

View full text Add to dashboard Cite

Transformer is one of the important equipment in the power grid, which helps to integrate renewable energy into the transmission and distribution network efficiently. The safe and stable operation of transformer is of great importance for the reliable transmission of electricity generated from renewable energy and for the reliable use of electricity by the end users. Therefore, it is important to assess the condition to avoid the faults of the transformer. In this paper, a variable weight synthesizing assessment model is presented that combines the G1 method, the entropy weight method, and a variable-weight method proposed in this paper to assess the condition of transformer based on the offset of the transformer equivalent circuit parameters. First, we propose deterioration indexes oriented to the maintenance management needs, which can well reflect the degree of deterioration of each transformer component. Second, the various defects of the transformer are used as the assessment indexes, and the initial weight is given to the assessment indexes according to the damage degree of the defect. The initial weight is calculated comprehensively by the G1 method and the entropy weight method. Then, each index is scored according to the offset of the equivalent circuit parameters, and the weights are adjusted appropriately according to the scores of the indicators using a variable weighting method to emphasize the severity of the defect or the “sub-health” condition of the transformer. Finally, the respective scores and combined weights of the assessment indexes are weighted to obtain a comprehensive score. The simulation shows that the model is more sensitive to abnormal and “subhealth” conditions of the transformer, which verifies the feasibility of the variable weight synthesizing model to assess the condtion of the transformer.

show abstract

Root cause analysis improved with machine learning for failure analysis in power transformers

Cited by 38 publications

References 30 publications

Fault diagnosis of transformer using artificial intelligence: A review

Fault diagnosis of transformer using artificial intelligence: A review

Composite Hybrid Framework for Through-Life Multi-Objective Failure Analysis and Optimisation

Research on Variable Weight Synthesizing Model for Transformer Condition Assessment

Contact Info

Product

Resources

About