Fatigue performance prediction method based on machine learning

Abstract

The invention discloses a fatigue performance prediction method based on machine learning. The method comprises the steps: firstly collecting material data to form initial sample data, employing a Sobol global sensitivity analysis method to evaluate the contribution degree of each feature variable to an output variable variance, carrying out the sorting of the importance of the feature variables,screening out key feature variables, and forming training sample data; dividing the training sample data into a training set, a verification set and a test set for training the model; performing modeltraining learning based on the training set and verification set data, obtaining corresponding model parameters, outputting a model trained based on existing training set / verification set learning, quantitatively evaluating model prediction performance by adopting a determination coefficient R2, completing establishment of the model, and performing fatigue performance prediction by utilizing thetrained model. According to the method, the nonlinear corresponding relation between the material information and the fatigue performance is established by utilizing the more comprehensive material information, and the method is applied to fatigue performance prediction of a higher-precision and generalized environment.

Description

technical field [0001] The invention belongs to the technical field of fatigue performance evaluation, in particular to a fatigue performance prediction method based on machine learning. Background technique [0002] In industrial fields such as automotive and aerospace, fatigue failure is the main mode of failure of critical structural parts. China's "Mechanical Engineering Handbook" pointed out in the chapter "Structural Fatigue Strength Design" that more than 80% of mechanical components are fatigue damage. Material fatigue performance data is important design information for product development. However, the cost of material durability testing is high and the cycle is long (the test cycle for a single fatigue curve is more than one month, and the test cost is more than 100,000 RMB (excluding material and processing fees )). Since 1870, the connection between the fatigue strength and tensile properties of metal materials has been widely explored. The main purpose is to ...

AI Technical Summary

Problems solved by technology

However, this technology only establishes the corresponding relationship between fatigue strength and tensile properties, and its relationship model adopts the simplest linear fitting. Due to the influence of manufacturing process, macro segregation, composition adjustment and other factors, this method is often unsatisfactory. The prediction results and the generalization ability of the model are also poor

Images