Prediction model for oil seal performance parameters based on PSO-MLP-KAN

Oil seals are critical sealing components in mechanical systems, and their sealing performance directly impacts the operational reliability of the entire system. To accurately, efficiently, and stably predict the sealing performance of oil seals, this study proposes a prediction method based on a combination of the Kolmogorov–Arnold network (KAN) and multi-layer perceptron (MLP) optimized by particle swarm optimization (PSO) algorithm. First, an oil seal performance prediction model is developed, taking into account macroscopic contact forces and contact curve parameters. To address the challenges posed by incomplete experimental data and limited data quantity, a hybrid approach integrating simulation data and experimental data is employed, complemented by the use of the Latin hypercube sampling method to construct an oil seal performance dataset. To further enhance model prediction accuracy, a PSO-MPL-KAN multi-output prediction model is established. Comparative analysis with existing prediction models demonstrates that the proposed PSO-MPL-KAN model achieves significantly higher accuracy in predicting oil seal sealing performance.