Eventos Anais de eventos
COBEM 2021
26th International Congress of Mechanical Engineering
CREEP LIFETIME PREDICTION OF LOW-CARBON STEELS WITH ARTIFICIAL NEURAL NETWORKS
Submission Author:
Giovanni Corsetti Silva , PR
Co-Authors:
Giovanni Corsetti Silva, Diogo B. Pitz
Presenter: Giovanni Corsetti Silva
doi://10.26678/ABCM.COBEM2021.COB2021-0141
Abstract
Current energy systems frequently operate under extreme conditions at high pressure and high temperature to increase the overall performance. Materials are often subjected to creep for the aforementioned conditions, which is the tendency of a material to fail due to a constant stress load at an elevated temperature, usually higher than one-third of its melting temperature. Effective creep investigation remains an arduous task, since reproducing high stresses and high temperatures for an extended period of time is not only hard to perform, but also very expensive. In recent years, Artificial Neural Networks have demonstrated the ability to solve complicated problems when a reasonable amount of data is available, predicting unseen situations after being trained with past situations. In the era of big data, where tons of observations are available, a reasonable approach to tackle the creep prediction problem is through a data-driven approach, which consists of gathering a large number of observations from past experiments and developing a unified model for creep prediction. The present work aims to implement an Artificial Neural Network for predicting the creep lifetime of low-carbon steels for high-temperature applications from the chemical composition, thermal treatment performed on the steel and working conditions (stress and temperature). The dataset utilized was created by combining an open- source dataset from the University of Cambridge containing 2,066 observations plus additional 157 observations collected from different publications in the scientific literature. The Neural Network was trained with 85% of the data, and its efficiency was checked on the remaining 15% for an unbiased estimation. The Artificial Neural Network used in this study successfully predicts creep lifetime from chemical composition, thermal treatment and working conditions with an R² of 88.6%. For few cases, the error is large, which indicates that more observations should be collected for enhancing the algorithm performance. The authors provide the code utilized in the present work for general use.
Keywords
Creep Lifetime, Low-carbon Steels, Artificial neural networks
DOWNLOAD PDF VIEW PRESENTATION
