Condition monitoring of axial piston pumps based on machine learning-driven real-time CFD simulation
Wentao Wang, Qun Chao, Junjie Shi, Chengliang Liu
Abstract
Condition monitoring of axial piston pumps is essential for the stable and safe operation of the entire hydraulic system in various electro-hydraulic equipment. Model-based condition monitoring methods are widely used due to their interpretability. However, they require significant time for prediction, which does not meet the requirements of real-time condition monitoring. This paper proposes a real-time simulation method for the condition monitoring of axial piston pumps. First, a high-fidelity 3D transient computational fluid dynamics (CFD) model of a healthy pump is established to generate simulated discharge pressure signals under different operating conditions. Second, a POD-Kriging surrogate model is constructed and trained by the CFD simulation dataset to predict healthy reference pressure signals in real time under new operating conditions. Finally, the monitoring discharge pressure signals and the real-time healthy reference ones are compared for the anomaly detection of the pump. The experimental results show that the surrogate model enables a CFD simulation in 1.2 s with a relative error below 1%. The proposed method provides a new paradigm for the real-time anomaly detection of axial piston pumps.