A remaining creep life prediction modeling method based on physical information machine learning

CN121279073BActive Publication Date: 2026-05-29EAST CHINA UNIV OF SCI & TECH
View PDF 2 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
EAST CHINA UNIV OF SCI & TECH
Filing Date
2025-09-08
Publication Date
2026-05-29

AI Technical Summary

Technical Problem

Existing creep life prediction methods rely on idealized assumptions, making them difficult to adapt to complex and ever-changing engineering practices. Furthermore, data-driven methods have poor generalization ability when high-quality data is insufficient, and cannot accurately predict the remaining life of materials.

Method used

A physical information-based machine learning approach is adopted, combining Robinson's linear damage accumulation rule and the time-temperature parameter method to construct a residual creep life prediction model that integrates a physical information layer and a data-driven layer. A recurrent neural network is used for dynamic prediction, with the physical information layer predicting the baseline value and the data-driven layer compensating for the deviation.

Benefits of technology

It improves the accuracy and robustness of remaining creep life prediction, enabling reliable life prediction with a small amount of data and adapting to complex service environments.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121279073B_ABST
    Figure CN121279073B_ABST
Patent Text Reader

Abstract

The application provides a remaining creep life prediction modeling method based on physical information machine learning, comprising: obtaining characteristic parameters and time series of remaining creep life of different test pieces to form a creep data set; establishing a physical information machine learning model, a remaining creep life processing unit being a component of each cycle unit of a recurrent neural network; the processing unit establishes a physical information layer based on a robinson linear damage accumulation rule and a time-temperature parameter method to predict a remaining creep life benchmark value; a data driven layer is established based on a machine learning algorithm to compensate for prediction deviation caused by complex creep behavior under different working conditions; the outputs of the two layers are added as a final output; finally, the model is trained by using the creep data set, and the established remaining creep life prediction model can predict the remaining creep life according to the historical characteristic parameters of the test piece. The method of the application can realize accurate prediction of the remaining creep life.
Need to check novelty before this filing date? Find Prior Art