“…Assuming ξ i (i = 1, 2, · · · , c) are the new chosen vibration features of DC bias obtained by linear combination of the original features x i (i = 1, 2, · · · , p), the relation between original and new features can be expressed as [29,30]:…”
DC bias is a great threat to the safe operation of power transformers. This paper deals with a new vibration-based technique to diagnose DC bias in power transformers. With this technique, the DC bias status of power transformers can be automatically recognized. The vibration variation process of a 500 kV autotransformer is tested under the influence of DC bias in the monopole trail operation stage of a ±800 kV HVDC transmission system. Comparison of transformer vibration under normal and DC-biased conditions is conducted. Three features are proposed and are validated by sensitivity analysis. The principal component analysis method is employed for feature de-correlation and dimensionality reduction. The least square support vector machine algorithm is used and verified successful in DC bias recognition. A remote on-line monitoring device based on the proposed algorithm is designed and applied in field DC bias diagnosis of power transformers. The suggested diagnostic algorithm and monitoring device could be useful in targeted DC bias control and improving the safe operation level of power transformers.
“…Assuming ξ i (i = 1, 2, · · · , c) are the new chosen vibration features of DC bias obtained by linear combination of the original features x i (i = 1, 2, · · · , p), the relation between original and new features can be expressed as [29,30]:…”
DC bias is a great threat to the safe operation of power transformers. This paper deals with a new vibration-based technique to diagnose DC bias in power transformers. With this technique, the DC bias status of power transformers can be automatically recognized. The vibration variation process of a 500 kV autotransformer is tested under the influence of DC bias in the monopole trail operation stage of a ±800 kV HVDC transmission system. Comparison of transformer vibration under normal and DC-biased conditions is conducted. Three features are proposed and are validated by sensitivity analysis. The principal component analysis method is employed for feature de-correlation and dimensionality reduction. The least square support vector machine algorithm is used and verified successful in DC bias recognition. A remote on-line monitoring device based on the proposed algorithm is designed and applied in field DC bias diagnosis of power transformers. The suggested diagnostic algorithm and monitoring device could be useful in targeted DC bias control and improving the safe operation level of power transformers.
“…Mahadevan et al [7] employed one-class SVM to detect odd behaviors during semiconductor etching process. You et al [8] applied principal component analysis (PCA) to extract features from acquired signals before using them as inputs for feedforward neural network. This model succeeded in detecting defects.…”
Predicting the status of flight vehicle in advance can have huge advantages in maintenance and early warning areas. Accurate forecast helps reduce maintenance costs and improve safety during the aircraft's life cycle. Combining the ability of convolutional neural network to extract features of different levels and its computational efficiency, a novel convolutional neural network --fault prognosis convolutional neural network(FP-CNN) is proposed in this paper, the purpose of which is to predict the Remaining Useful Life (RUL) by learning sequential information and extracting sensor features from noisy datasets under different operating modes. An experiment on CMPASS data is conducted to prove the efficiency and accuracy of this framework.
“…In addition to the papers having been reviewed in the previous part, this second part of the Special Section presents eight papers in total. According to their contents, the selected papers have been categorized into three types, i.e., four papers with the topic "data-driven controller tuning/design" (see [2]- [5]), three papers with the topic "data-driven process monitoring" (see [6]- [8]), and one state-of-the-art (SoA) paper providing an overview of industrial data-based techniques (see [9]). To make the published contributions explicit for readers' further reference, brief summaries of the included papers are given as follows.…”
mentioning
confidence: 99%
“…In [6], databased methods have been considered for laser welding process monitoring and welded defect diagnosis. A promising opinion goes that the laboratory-scale sensor can be replaced with the industrial-scale sensor after the proper establishment of data-driven models through multivariate statistics and machine learning methods.…”
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.