Gear pumps are widely used in hydraulic systems, where accurate flow rate prediction is essential for improving system efficiency, reliability, and intelligent monitoring. However, traditional experimental data acquisition are costly and time-consuming, while the coverage of operating conditions is often insufficient, limiting the effectiveness of data-driven prediction models. To address these challenges, this paper proposes a collaborative prediction framework that integrates ANSYS-FLUENT finite element simulation with a gated neural network. In this study, pressure–flow rate time-series data collected at the meshing position of gear pumps are employed. A Generative Adversarial Network (GAN) is utilized to construct an augmented sample set, and a training dataset incorporating physical mechanisms is established. On this basis, a gated LSTM-CNN-CBAM (G-LCC) neural network is designed, which combines multi-scale feature extraction with a dynamic gating mechanism. Experimental results show that under varying inlet velocities and rotational speeds, the proposed G-LCC achieves average R2 values of 0.749 and 0.833, respectively, corresponding to relative improvements of 10.1% (from 0.673 to 0.749) and 16.9% (from 0.692 to 0.833) compared to the traditional LSTM-CNN model. Furthermore, in cross-equipment tests involving three heterogeneous gear pumps, the model achieves an average R2 of 0.692, verifying both the feasibility of substituting simulations for physical experiments and the strong generalization capability of the proposed approach. This study presents an effective integrated framework that combines CFD-based physical modeling with a gated neural network architecture tailored for gear pump flow prediction. By addressing the spatiotemporal coupling of internal flow fields, the G-LCC model demonstrates superior predictive accuracy and provides a robust tool for the intelligent monitoring of hydraulic systems.
Liu et al. (Fri,) studied this question.