Research on Convolutional Neural Network Regression forContact Erosion Estimation of AC Contactors
The paper proposes a contact erosion estimation method that models electrical signal data of breaking processes by Convolutional Neural Network Regression(CNNR) for AC contactors.Voltage and current waveform data of breaking operations is acquired during AC-4 tests, and meanwhile the contact mass loss of each phase is measured after every 600 operations.We apply voltage waveforms, current waveforms and voltage-current-combo waveforms as feature arrays, and label them by accumulated contact mass loss values via piecewise linear interpolation from the mass loss measurements.Thus the training and testing datasets for CNN are founded.The breaking waveforms are reformed to two-dimensional feature metrices and input them to a CNN model which consists of three convolution layers, three pooling layers and two full connection layers.For the training process, the model parameters are optimized by the mean square error (MSE) between the ground-truth mass loss labels and the CNN outputs.The test results show that the proposed method achieves a relatively high estimation precision of contact erosion of AC contactors when voltage waveforms are utilized as features.
AC contactors contact mass loss convolutional neural networks regression
Ziran Wu Hechen Cui Guichu Wu Yingmin You
The Key Laboratory of Low-voltage Apparatus Intellectual Technology of ZhejiangWenzhou University No.276, Xueyuan Mid Road, Lucheng District, Wenzhou, Zhejiang Province, 325027, China
国际会议
江苏苏州
英文
45-50
2019-11-04（万方平台首次上网日期，不代表论文的发表时间）