Intelligent Prediction of Control Force of Active Hydrodynamic Pressure Dry Gas Seal under Random Conditions and Comparative Study of Its Algorithm Performance

This study delved into optimizing the performance and longevity of dry gas seal, a crucial component frequently impacted by operational disturbances, such as inconsistencies in rotational speed and alterations in sealed medium pressure. The research innovatively employed active hydrodynamic pressure and a nuanced controllable closing force, aspiring to bolster the adaptability of the dry gas seal amidst these challenges. Utilizing foundational principles from gas lubrication theory, a strategic equation had been developed, instrumental in proficiently navigating the film pressure within the spiral groove. This initiative facilitated a more profound and precise understanding of the ramifications of various disturbance conditions on sealing performance parameters. Strategic data collection methods, specifically Latin hypercube sampling were harnessed to glean valuable insights into the requisite control force essential for sustaining a stable film thickness in the face of prevailing disturbances. A meticulous comparative assessment was conducted, encompassing four pivotal intelligent prediction algorithms: BP neural network, RBF neural network, multiple linear regression, and locally weighted linear regression. This comparative approach aimed to discern the predictive capabilities and effectiveness of these algorithms in navigating the complexities of seal performance under variable conditions. Consequential findings from the study unveiled that the end face film thickness was intrinsically susceptible to significant variations induced by operational disturbances. By optimizing the adjustment of the closing force, a practical methodology had been illuminated, contributing significantly to the stabilization of the film thickness amidst the spectrum of encountered disturbances. The study also revealed the pivotal influence of augmenting the volume of training samples, demonstrating a marked improvement in the refinement, accuracy, and overall predictive aptitude of the analyzed models. Comparing the prediction results of the four models, the prediction ability and stability of BP neural network were superior, and the theoretical film thickness control results achieved by applying the dry gas seal meet the accuracy requirements.