Machine learning-based fatigue life evaluation of the pump spindle assembly with parametrized geometry

Artificial intelligence-based fatigue analysis methods have gained considerable attention in recent years, but little research has been conducted on the fatigue behaviors of assembled structures, which involve numerous factors that impact structural fatigue life and complicated assembling associations in parameter selection. This paper presents a hybrid machine learning (ML) model-based evaluation platform for data-driven fatigue life analyses of a parameterized industrial pump spindle that considers the impacts of variable loads and characteristic design parameters. Firstly, the numerical fatigue evaluation is validated by comparing experimental and simulation results of spindle samples with a specific dimension. Secondly, datasets for training and testing are constructed to assess high cycle and low cycle fatigue lives of parameterized pump spindle assemblies under two real working conditions of cyclic start-stop and continuous running. The proposed hybrid ML method exhibits excellent overall performance for both running conditions, validating its accuracy and reliability. Additionally, the superiority of the hybrid method for fatigue evaluation of the assembly is confirmed through performance comparisons between single forward models. Finally, sensitivity-based parametric analysis is performed to interpret the model, revealing that pre-Tightening load and diameter of the shaft seal have the most significant impacts on the fatigue properties under both cyclic working conditions. This work provides an efficient tool for quantifying the fatigue lives of parametrized spindles and contributes to the safety assessment in structural design and optimization.