Material Fracture Life Prediction Under High Creep Conditions Using Decision Trees and Rule-based Techniques

Several elements in power plants suffer from high creep conditions in its normal service life. This fact makes that high efficiency materials are used to manufacture critical components of the system. Among them, high chrome content alloy steels are widely applied in these cases to optimize the final mechanical properties of critical elements. Although knowing the mechanical properties at creep conditions is not an easy task, since creep tests are a high time consuming process. Due to these problems, the use of regression models to get a better understanding and a prediction of mechanical properties of the material is a really helpful technique. In this work, several regression techniques, based on decision trees and decision rules, are applied to predict the time when the material fracture happens. In order to build these models, a representative dataset of the problem was studied to get a better knowledge of the problem using several techniques of multivariate analysis. Then, a validation methodology based on cross validation training and simple validation testing was applied to verify the generalization of the models. The algorithms applied in this methodology show how decision trees and decision rules techniques can achieve accurate results in their prediction, obtaining low RMSE close to a 7%. And finally, among the studied algorithms, the one based on rule-based cubist technique performed the most accurate results.