Assessment method for wear severity in axial piston pumps based on dynamic identification of leakage coefficient

Abstract

Wear of key friction pairs is a critical failure for swashplate-type axial piston pumps. Real-time wear diagnosis of piston pumps is crucial for their predictive maintenance. While model-data fusion-driven methodology for wear fault diagnosis shows potential, existing methods under this strategy suffer from modeling errors and notable complexity. To address this, an assessment method for wear severity in axial piston pumps based on dynamic identification of the leakage coefficient is proposed. A fluid-thermal coupled model of the piston pump is established, incorporating the leakage coefficient to calculate the leakage flow rate. An identification algorithm is constructed to determine the leakage coefficient in the model using monitored data of both discharge pressure and leakage flow rate of the piston pump. When a piston pump operates under a specific operating condition, the leakage coefficient characterizes its overall wear degree, enabling it to be a quantitative evaluation indicator for the pump’s wear severity. Experiments confirm a positive correlation and sensitivity between the identified leakage coefficient and the pump’s wear severity under a specific operating condition. When operating conditions vary within a certain range, the leakage coefficients of piston pumps with different wear degrees exhibit distinct clustering characteristics, with no overlap between coefficient clusters. In the case study, the distribution ranges of the leakage coefficients for the piston pump under different operating conditions, corresponding to four progressively increasing levels of wear, are as follows: (4.00e⁻¹⁴, 8.37e⁻¹⁴), (2.48e⁻¹³, 6.87e⁻¹³), (1.24e⁻¹², 1.50e⁻¹²), (1.61e⁻¹², 1.96e⁻¹²) m³/(s·Pa). This validates the effectiveness of the proposed method and its applicability across a considerably wide operating condition range. The proposed method exhibits a lower modeling error. Crucially, it requires neither extensive test data nor relies on engineering experience or large training datasets, making it more straightforward and convenient to apply.