From: Techniques of infrared thermography for condition monitoring of electrical power equipment
Reference | Title of Paper | Results / Analysis |
---|---|---|
Karvelis et al. [17] | An Automated Thermographic Image Segmentation Method for Induction Motor Fault Diagnosis | The image segmentation method used with Naïve Bayes classification yielded 100% accuracy, while C4.5 gave 91.48% |
Eftekhari et al. [11] | A novel indicator of stator winding inter-turn fault in induction motor using infrared thermal imaging | The Mahalanobis distance was used to measure changes in the IR image pixels. Also, forecast hot spots in the stator using Hullindex, Hotarea and Histogrammean values with RMSE error of 0.0097 translating to 99.13% accuracy |
Ali-Younus and Yang [9] | Wavelet Co-efficient of Thermal Image Analysis for Machine Fault Diagnosis | Got the best result at the second level of decomposition |
Fambrini et al. [33] | GPU Cuda JSEG Segmentation Algorithm associated with Deep Learning Classifier for Electrical Network Images Identification | The method resulted to 99.91% detection accuracy in transformers, 86.94% in knife wrenches, 84.88% in splice connectors, and 71.49% in bushings |
Najafi et al. [35] | Fault diagnosis of electrical equipment through thermal imaging and interpretable machine learning applied on a newly-introduced dataset | The technique obtained 93.8% accuracy in 11 classes of equipment condition and 95.6% for 9 classes |
Jadin et al. [13] | Thermal condition monitoring of electrical installations based on infrared image analysis | The normalization cross-correlation method got 95.804% classification accuracy |
Dutta et al. [26] | Condition monitoring of electrical equipment using thermal image processing | The highest Peak Signal-to-Noise Ratio of 63.13Â dB and least MSE of 0.03 was observed when Otsu method was applied |
Huda and Taib [14] | Suitable features selection for monitoring thermal condition of electrical equipment using infrared thermography | The Multi-Layered Perceptron with Scale Conjugate Gradient and Levenberg–Marquardt training got the highest identification training degree of 82.89% and testing rate of 74.25% than other comparable methods |
Huda and Taib [15] | Application of infrared thermography for predictive/preventive maintenance of thermal defect in electrical equipment | The optimum result achieved with the second-fold training dataset were 97.75% accuracy, 95.89% specificity, and 98.88% sensitivity. Whereas on the testing dataset, the performance fell to 80.40% accuracy, 75.29% specificity, and 83.98% sensitivity. And, the Discriminant Analysis classifier yielded the best accuracy of 82.40% |
Zou et al. [19] | Novel intelligent fault diagnosis method for electrical equipment using infrared thermography | Optimization of the classification accuracy up to 97.8495% was achieved with SVM by the application of a coarse-to-fine parametric model |
Liu et al. [28] | Infrared image combined with CNN-based fault diagnosis for rotating machinery | The 6 layered directly trained convolutional neural network achieved the best testing accuracy of 95.8% using 60 × 60 pixel images, without feature extraction |
Janssens et al. [18] | Thermal image-based fault diagnosis for rotating machinery | Two image processing channels were processed by detecting rotor imbalance irrespective of the type of bearing defect and then identifying bearing issues notwithstanding any rotor imbalance. The model was up to 88.25% accurate for 8 machine anomalies |
Singh et al. [22] | Fault diagnosis of induction motor cooling system using infrared thermography | The method obtained a correlation factor of 99.02% between the Hotindex with respect to the percentage of inter-turn short circuits. Hence, the Hotindex is proportional to the stator coil inter-turn faults |
Mahami et al. [37] | Induction motor condition monitoring using infrared thermography imaging and ensemble learning techniques | The proposed ensemble learning technique outperformed the SVM, Decision tree, K-nearest neighbor, Least Square SVM, and Deep Rule-Based methods by returning 100% classification accuracy |
Vidhya et al. [36] | Transformer breather thermal image decomposition for fault diagnosis | Without image decomposition, the MAD of the Symlet wavelet transformations produced 99.68% for normal operation, 99.95% for mild faults, and 98.87% for severe winding faults. And, associated Standard deviations of 106.2, 106.6, and 106.5 of the respective three operational states But, with image decomposition, the symlet wavelet transformation produced significant variation in MAD (8.154, 1.25 and 3.677) as well as the standard deviation (12.66, 2.007 and 6.52) for each respective normal, mild and severe states |
Zou Huang [72] | An Intelligent Fault Diagnosis Method for Electrical Equipment Using Infrared Images | Classification accuracy of 95.6989% was achieved using SVM with bestc2 and bestg2 parameters set to 223.5126 and 0.9639, respectively |
Cui et al. [12] | The Methods in Infrared Thermal Imaging Diagnosis Technology of Power Equipment | The work used Radial Basis Probabilistic Neural Network to identify the level of contamination in string insulators with up to 91.12% accuracy |
Fanchiang et al. [73] | Power Electric Transformer Fault Diagnosis Based on Infrared Thermal Images Using Wasserstein Generative Adversarial Networks and Deep Learning Classifier | The proposed model is a Wasserstein Auto-encoder Reconstruction-based Differential Image Classification (WAR-DIC). It is also a feather-weight network with only 0.223 × 103 parameters, 1.837 MB of weight storage, and 1.781 × 103 floating point calculations. Overheating of conducting wires, inter-turn faults, and overheating in connecting points were investigated for eight fault conditions on a balanced dataset. The classification accuracies that were gotten for four different datasets were 99.95%, 99.89%, 99.71%, and 99.46% |
Fang et al. [74] | Fault diagnosis of electric transformers based on infrared image processing and semi-supervised learning | The application of GAN to synthesize sampled IR images enhanced the classification of equipment defects with an accuracy of 82.2%, recall of 84.7%, and precision of 83.1%. And, recognized overheating with an accuracy of 86.2%, recall of 84.8%, and precision of 83.5% |
Fanchiang, Kuo [75] | Application of thermography and adversarial reconstruction anomaly detection in power cast-resin transformer | Overheating in 1MVA, 24/0.38Â kV cast-resin transformers were investigated with Variational Autoencoder-based-GAN using the difference in the pixel-wise cosine between the real and synthetic images. The results achieved include: F1 score of 94.4%, AP of 94.1%, and AUROC of 94.5% |