A semi-supervised fault diagnosis method for axial piston pump bearings based on DCGAN

Abstract

Recently, deep learning has developed rapidly in the fault diagnosis technology of axial piston pumps. However, when the training data is scarce and the label information is insufficient, many traditional intelligent fault diagnosis models are invalid. To solve these problems, an intelligent fault diagnosis method for axial piston pumps is proposed based on deep convolutional generative adversarial network (DCGAN). Firstly, the continuous wavelet transform (CWT) and DCGAN are designed to enhance the fault features and expand dataset, respectively. Secondly, according to the number of labeled samples, DCGAN and semi-supervised GAN (SGAN) are used to extract the deep features of the image domain. Finally, the clustering algorithm is used to classify the extracted features to realize the fault diagnosis of the axial piston pump bearing. To verify the feasibility of the proposed method, experimental investigation and public dataset are adopted. When the evaluation indicators of the clustering results are close to 1, the proposed method shows the advantages of high diagnostic accuracy, superior generalization ability and excellent anti-noise ability.

FEM simulation-determined band pass filter method with continuously changed bandwidth for fault detection in axial piston pumps

“…Axial piston pumps, which play a crucial role in hydraulic systems, unavoidably fail suddenly owing to the harsh working conditions, causing significant economic losses and even casualties [1][2][3][4][5][6]. Therefore, developing axial piston pump fault diagnosis techniques is necessary to maintain the performance of hydraulic systems [7,8].…”

Maximal overlap discrete wavelet packet transforms and variants of neutrosophic cubic cross-entropy-based identification of rotor defects

“…Step 2. Decomposition of signals through MODWPT Vibration signals are generally decomposed through continuous wavelet transform (CWT) [33][34][35], discrete wavelet transform (DWT) and wavelet packet transform (WPT) for carrying out time-frequency analysis [34]. CWT decays signals smoothly at each scale, and is the most powerful tool to carry out time-frequency analysis but it is computationally expensive.…”

Fault detection of a reciprocating plunger pump with fault-free data based on an unsupervised feature encoder and minimum covariance determinant

“…Based on intra-class and inter-class neighborhood information graphs embedding orthogonal discriminant projection, the global distribution feature information and local geometric structure information of the data were extracted. He et al used continuous wavelet transform to generate time-frequency images and a deep convolution GAN extended dataset [20]. Pan et al [21] proposed a convolutional adversarial AE for mechanical fault recognition with unseen classes via one-class classification.…”

“…Hong and Suh combined UFE and a deep support vector machine network to realize fault diagnosis of rolling bearings [16]. He et al extracted features based on DNN and used a clustering algorithm to classify the extracted features to realize fault diagnosis of axial piston pump bearings [20]. Cao et al [22] trained the AE model for data, and the features extracted by a hidden layer with smaller dimensions were taken as the compressed representation of the data.…”

“…Considering that the phase angle corresponding to the typical fault of a reciprocating plunger pump does not exceed 30 • , the length of the limited signal does not exceed 1/12 of the whole cycle signal. In order to effectively increase the influence of noise and ensure the quality of pseudo-fault samples, the SNR interval of Gaussian white noise is selected as [10,20]. The value range of parameter α, β and γ is set as [0.001, 0.8] and the interval precision is 0.001; the constraint is that the sum of the three weights is equal to 1.…”