Prediction Model of Aluminized Layer Thickness Based on X-Ray Fluorescence and Extreme Gradient Boosting

Abstract

As an important means of high-temperature protection for aero-engine turbine blades, the quality of aluminized coatings is closely related to flight safety. The thickness of the aluminized layer is an essential factor in evaluating its performance. However, it is not easy to measure it accurately by current nondestructive testing methods. For this problem, the X-ray fluorescence technology is combined with the extreme gradient boosting (XGBoost) algorithm, and the feature element extraction by Pearson correlation coefficient screening (PCCS) is used to build a prediction model for the thickness of the aluminized layer. The average relative error of the prediction results is compared with K nearest neighbor regression, linear regression, support vector machine, and random forest models. The results show that the PCCSXGBoost model had the smallest average error of 1. 60% in predicting thickness compared with other models. The study provides a new prediction method for nondestructive testing of the thickness of the aluminized layer.