Prediction of Fatigue Life of Powder Metallurgy Superalloy Disk via Machine Learning

Author of the article:ZHANG Guodong1 , SU Baolong1 , LIAO Weijie1 , WANG Xiaofeng2 , ZOU Jinwen2 , YUAN Ruihao1 , LI Jinsh

Author's Workplace：1. State Key Laboratory of Solidification Processing, Northwestern Polytechnical University, Xi' an 710072, China; 2.Key Laboratory of Advanced High Temperature Structural Materials, Beijing Institute of Aeronautical Materials of AECC, Beijing 100095, China; 3. Innovation Center NPU Chongqing, Chongqing 401135, China

Key Words：powder metallurgy; machine learning; life prediction; superalloy

Abstract：Due to the pollution of powder making and spalling of container materials, non-metallic inclusions will inevitably be introduced in the preparation of superalloy by powder metallurgy. These inclusions seriously harm the mechanical properties of superalloy turbine disks and may lead to low cycle fatigue failure of turbine disks. At present, the fatigue life model is mainly developed by Manson-Coffin's formula, which does not take into account the change of elastic modulus during fatigue and the influence of alloy defect characteristics, which results in a large error between the predicted results and the actual results. In this paper, five quantitative prediction models between the low cycle fatigue numbers of turbine disk with the distance of inclusions to disk surface and the inclusion sizes were established by using machine learning models. The results showed that compared to support vector machine, random forest, kernel ridge regression and lasso algorithms, the gradient lifting machine algorithm can better predict the fatigue life.