•  
  •  
 

Abstract

Using predictions of crack-tip-opening-displacement (CTOD) to measure the extent of fatigue damage has provided the opportunity to prepare an efficient strategy for protecting mechanical structures from damage and developing a structural health monitoring system. The objective is to forecast non-measurable CTOD by using machine learning methods. In this paper, an optical metrology device, which is built by Alicona on a confocal microscope and hereafter referred to as Confocal Microscope, has been used to measure CTOD. However, two factors limit the usage of Alicona Apparatus: (i) the size of optical images, where a CTOD over 400 micrometers cannot be measured; and (ii) the need to protect the device, as a CTOD over 150 micrometers has a significant impact on the safety of Confocal Microscope. Therefore, this paper has utilized Gaussian Processes (GP) and the Support Vector Machine (SVM) to forecast the non-measurable CTOD. Four machine learning metrics, mean average error (MAE), mean square error (MSE), root mean square error (RMSE) and R-squared error have been used in this study to evaluate the performance of the regression models. The results indicate that the GP model provides a better estimate of the current CTOD measurements. However, the SVM model provides a better forecast of future CTOD data based on the behavior of the CTOD rate.

Download

Share

COinS