Integrating Finite Element Analysis and Machine Learning for Predictive Analysis of the Useful Life of Industrial Cutting Blades: A Case Study

Abstract

This study presents a hybrid methodology that integrates finite element method (FEM) simulations with machine learning to create a highly efficient surrogate model capable of predicting the fatigue life of industrial cutting blades. Initially, an FEM analysis (Ansys v14.5) identifies von Mises stress as the main failure driver. Recognizing the scarcity of real failure data ( n = 30) and the high computational cost of pure simulations, physical insights from FEM are converted into an explicit transfer function to generate a physically informed synthetic dataset ( n = 20,000). The random forest model achieves an R 2 of 0.9840 on real data, with an average error of only 2.2 hours, representing a 94.3% reduction in error compared to pure FEM. This paradigm proves to be a viable solution to the data scarcity paradox in predictive maintenance environments.