Fault detection of axial piston pumps is of great significance to improve the reliability and life of fluid power systems. However, it is difficult to detect multiple faults on key lubricating interfaces due to the liquid-solid coupling. This paper proposes a fault detection strategy of the three key lubricating interfaces based on the one against all (OAA) and spare support vector machine (SSVM). The parameter sparsity is imposed to deal with the performance degradation of OAA-SVM model as a result of the imbalanced dataset. Experimental investigations on the benchmark dataset and axial piston pumps are carried out. Results show that the OAA-SSVM model accuracies of the benchmark dataset and axial piston pump are 96.67% and 95.83%, respectively, which are better than the OAA-SVM model. The recall rates of the bearing fault 3 and pump fault 2 can decrease by 13.33% and 10.00%, respectively. And the false discovery rates of the normal bearing and normal pump can be reduced by up to 7.58% and 6.24%, respectively. Besides, the OAA-SSVM model can improve the feature sparsity. Results show that the proposed method is effective in detecting multiple faults of axial piston pumps.

The output flow ripple of the axial piston pump is one of the excitation sources for the hydraulic system vibration. The amplitudes of its specific harmonics must be reduced to avoid the resonance with the hydraulic pipeline. In this paper, a method on the nonuniform distribution of the pistons is put forward to adjust the flow ripple. The deflection angles of the pistons are used to describe the distribution rule. The distribution rule is imported to the Fourier expansion of the flow rate of each single-piston chamber, and then every single flow rate is superposed to obtain the Fourier coefficient of total flow rate that becomes the function of deflection angles. After this, objective optimization design is carried out to reduce the amplitudes of specific harmonics. Finally, the dynamic simulation model of the nonuniformly distributed axial piston pump is established to verify the effects of objective optimization. The results show that the amplitude of the ninth harmonic of the flow ripple can be reduced by about 40%, and the reductions are about 99% for the 18th and 27th harmonic.

Online wear state recognition of key friction pairs in the axial piston pump is of great significance for stable operation and predictive maintenance of the whole hydraulic system. Edge computing (EC) meets the real-time and low-cost requirements of online wear state recognition whereas two challenges limit its application. One is that current fault diagnosis methods only focus on local fault information, causing inaccuracy and poor generalization ability in different working conditions. The other is that the computing power and storage of EC devices are limited. Therefore, a lightweight knowledge-distilled vision transformer (ViT) is proposed for online wear state recognition. A novel time-frequency domain stacked and channel-weighted pooling structure is proposed to directly process raw time series. To realize high accuracy and high generalization ability, a ViT-based teacher model is pretrained to learn local and global information. To narrow model capacity gap and adapt to the limited resource of the edge node, a novel student model with a simplified self-attention mechanism is proposed to mimic the structure of the ViT and learn from the pretrained teacher model through knowledge distillation. An edge node with functions of signal acquisition, data preprocessing, and wear state recognition is designed and the distilled student model is deployed into it. Comparison with other state-of-the-art methods, ablation experiment, and online verification experiment demonstrate that the proposed method trades off wear state recognition performance and hardware limitations.

Multisource fault patterns usually occur simultaneously in the axial piston pump. The coupled fault problem makes different fault patterns affect each other on the failure mechanism, leading to the aggravation of each fault pattern. Meanwhile, the features of different fault patterns will be aliased on the monitored signals, making some relatively weak fault patterns hard to diagnose. In this article, a novel framework named temporal–spatial attention network (TSAN) is proposed to solve this problem. The key of the proposed method is to effectively extract fault-sensitive features from multisource sensors on both temporal and spatial scales. First, to extract periodic fault-sensitive features of each sensor on the temporal scale, the multihead attention-based temporal feature extraction model is constructed. Then, to fuse the extracted temporal features of each sensor and extract fault-sensitive features between different sensors on a spatial scale, the multihead attention-based spatial feature extraction model is constructed. The proposed learnable temporal feature token and spatial feature tokens effectively transmit the temporal features and spatial features and improve the efficiency of the fault-sensitive temporal–spatial feature extraction. Compared with other state-of-the-art methods, comparison experiments demonstrate that the proposed framework improves the fault diagnosis accuracy under coupled fault problems by at least 22.06%.