With the development of machine learning in recent years, the application of machine learning to machine fault diagnosis has become increasingly popular. Applying traditional feature extraction methods for complex systems will weaken the characterization capacity of features, which are not conducive to subsequent classification work. A reciprocating compressor is a complex system. In order to improve the fault diagnosis accuracy of complex systems, this paper does not use traditional fault diagnosis methods and applies deep convolutional neural networks (CNNs) to process this nonlinear and non-stationary fault signal. The valve fault data is obtained from the reciprocating compressor test bench of the Daqing Natural Gas Company. Firstly, the single-channel vibration signal is collected on the reciprocating compressor and the one-dimensional CNN (1-D CNN) is used for fault diagnosis and compared with the traditional model to verify the effectiveness of the 1-D CNN. Next, the collected eight channels signals (three channels of vibration signals, four channels of pressure signals, one channel key phase signal) are applied by 1-D CNN and 2-D CNN for fault diagnosis to verify the CNN that it is still suitable for multi-channel signal processing. Finally, further study on the influence of the input of different channel signal combinations on the model diagnosis accuracy is carried out. Experiments show that the seven-channel signal (three-channel vibration signal, four-channel pressure signal) with the key phase signal removed has the highest diagnostic accuracy in the 2-D CNN. Therefore, proper deletion of useless channels can not only speed up network operations but also improve diagnosis accuracy.

Reciprocating compressors are widely used in petroleum industry. Due to containing complex nonlinear signal, it is difficult to extract the fault features from its vibration signals. This paper proposes a new method named Convolutional Neural Network based on Multisource Raw vibration signals (MSRCNN). The proposed method uses multisource raw vibration signals collected by several sensors as input and uses the designed CNN to operate both the feature extraction and classification. The gas valve signals of reciprocating compressor in different states are used as the experimental data. In order to test the effectiveness of the proposed method, it is compared with the traditional BP (Back‐Propagation) neural network fault diagnosis method based on power spectrum energy and wavelet packet energy. In order to further test the antinoise performance of the proposed method, some noisy signals with different signal‐to‐noise ratios were constructed by adding white noise into sampled signals for testing. The results show that the MSRCNN model has higher fault recognition rate than the traditional methods. This indicates that the MSRCNN method not only has good fault recognition effect, but also has certain antinoise performance.

Domain shift is the main reason that the recognition accuracy of traditional intelligent fault diagnosis methods greatly decreases. However, previous methods mainly focus on global domain matching, without considering the finer internal structure and making full use of source domain data, resulting in unsatisfactory adaptive effects. In this paper, we present an unsupervised joint subdomain adaptation network (JSAN), which reduces the discrepancy between the two domains by joint local maximum mean discrepancy (JLMMD). JLMMD divides the features of the specific activation layer into different subdomains according to the labels and matches the subdomains which have the same labels. At the same time, the joint distribution of the feature space and the prediction space of the target domain is matched utilizing the joint distribution of the source domain feature and the label. The experimental results show that the proposed method has achieved remarkable results in the domain adaptation task.