A modified divide-and-conquer based machine learning method for predicting creep life of superalloys

TL;DR

This paper enhances a divide-and-conquer machine learning approach for predicting superalloy creep life by integrating dimensionality reduction and automated hyperparameter tuning, significantly boosting accuracy and intelligence.

Contribution

The authors introduce a modified method that improves clustering and prediction accuracy, and automates hyperparameter selection, advancing the original approach.

Findings

Prediction errors are substantially reduced.

Clustering results are more desirable.

Hyperparameter automation improves method intelligence.

Abstract

Recently Liu et al. (Acta Mater., 2020) proposed a new divide-and-conquer based machine learning method for predicting creep life of superalloys. The idea is enlightening though, the prediction accuracy and intelligence remain to be improved. In the present work, we modify the method by adding a dimensionality reduction algorithm before the clustering step and introducing a grid search algorithm to the regression model selection step. As a consequence, the clustering result becomes much more desirable and the accuracy of predicted creep life is dramatically improved. The root-mean-square error, mean-absolute-percentage error and relevant coefficient of the original method are 0.2341, 0.0595 and 0.9121, while those of the modified method are 0.0285, 0.0196, and 0.9806, respectively. Moreover, the ad-hoc determination of hyperparameters in the original method is replaced by automated determination of hyperparameters in the modified method, which considerably improves the intelligence of the method.