Planetary gearbox with high speed and high precision is important for sophisticated power equipment, and the effective sensor data are necessary for reliability and stability. However, the traditional models for the fault diagnosis of planetary gearbox are difficult to maintain the high accuracy and effectiveness when the amount of training data are scarce. To solve this problem, inspired by deep learning, a novel intelligent method for planetary gearbox fault diagnosis is proposed by utilizing advantage of gated recurrent neural network (RNN) in the feature extraction. Moreover, the dropout technology is introduced to the proposed method to further reduce the requirement of training data. By dividing the parameters of the classification layer and using a few new fault data to fine-tune the learned network parameters, the proposed method can quickly realize the diagnosis of new type faults while maintaining the original recognition ability. Finally, the test signals of the planetary gearbox are used to verify the proposed method. The results show the proposed method has the advantages of high diagnostic accuracy, fast recognition speed, and well real-time in fault diagnosis. Furthermore, the proposed method can make full use of the less information to diagnose different types of faults under the different working conditions.

The axle box in the bogie system of subway trains is a key component connecting primary damper and the axle. In order to extract deep features and large-scale fault features for rapid diagnosis, a novel fault reconstruction characteristics classification method based on deep residual network with a multi-scale stacked receptive field for rolling bearings of a subway train axle box is proposed. Firstly, multi-layer stacked convolutional kernels and methods to insert them into ultra-deep residual networks are developed. Then, the original vibration signals of four fault characteristics acquired are reconstructed with a Gramian angular summation field and trainable large-scale 2D time-series images are obtained. In the end, the experimental results show that ResNet-152-MSRF has a low complexity of network structure, less trainable parameters than general convolutional neural networks, and no significant increase in network parameters and calculation time after embedding multi-layer stacked convolutional kernels. Moreover, there is a significant improvement in accuracy compared to lower depths, and a slight improvement in accuracy compared to networks than unembedded multi-layer stacked convolutional kernels.

Compressors are important components in various power systems in the field of energy and power. In practical applications, compressors often operate under non-design conditions. Therefore, accurate calculation on performance under various operating conditions is of great significance for the development and application of certain power systems equipped with compressors. To calculate and predict the performance of a compressor under all operating conditions through limited data, the interpolation method was combined with a support vector machine (SVM). Based on the known data points of compressor design conditions, the interpolation method was adopted to obtain training samples of the SVM. In the calculation process, preliminary screening was conducted on the kernel functions of the SVM. Two interpolation methods, including linear interpolation and cubic spline interpolation, were used to obtain sample data. In the subsequent training process of the SVM, the genetic algorithm (GA) was used to optimize its parameters. After training, the available data were compared with the predicted data of the SVM. The results show that the SVM uses the Gaussian kernel function to achieve the highest prediction accuracy. The prediction accuracy of the SVM trained with the data obtained from linear interpolation was higher than that of cubic spline interpolation. Compared with the back propagation neural network optimized by the genetic algorithm (GA-BPNN), the genetic algorithm optimization of extreme learning machine neural network (GA-ELMNN), and the genetic algorithm optimization of generalized regression neural network (GA-GRNN), the support vector machine optimized by the genetic algorithm (GA-SVM) has a better generalization, and GA-SVM is more accurate in predicting boundary data than the GA-BPNN. In addition, reducing the number of original data points still enables the GA-SVM to maintain a high level of predictive accuracy.