Stiffness optimization of electric spindle performance based on multi-layer perceptron integrated Bayesian

The performance optimization of electric spindles is critical for enhancing machining accuracy and efficiency. Traditional methods often struggle with high-dimensional parameter spaces and complex nonlinear behavior. This paper proposes a novel hybrid approach integrating a Multi-Layer Perceptron (MLP) with Bayesian Optimization (BO) to address these challenges. The MLP models the intricate relationships between design parameters-including overhang length, support span, and bearing stiffness-and performance metrics such as static stiffness and deformation. BO is then employed to efficiently navigate the design space and identify optimal configurations. Using ANSYS to simulate the electric spindle, various design parameters and spindle stress-deformation data are obtained through simulation analysis. By establishing a response surface, the design parameters sensitive to spindle stress-deformation are identified, including overhang length, support span, radial stiffness of the front bearing, and radial stiffness of the rear bearing. Experimental results demonstrate that the proposed method significantly outperforms conventional techniques: it increased static stiffness by 25.7 N/µm and reduced deformation by 0.2 µm compared to a multi-objective genetic algorithm (MOGA). The study confirms that the MLP-BO framework offers superior optimization efficiency, accuracy, and global search capability, providing valuable insights for the design of high-performance electric spindles and other precision manufacturing systems.