PSI - Issue 42

Zafer Yüce et al. / Procedia Structural Integrity 42 (2022) 663–671 Yuce Z., Yayla P., Taskin A / Structural Integrity Procedia 00 (2019) 000 – 000

668

6

3. Results As presented in Section 2, the rainflow histogram and statistical features of load spectra were fed to the machine learning model to predict the CG life of the joint. Regarding error metrics, the root mean square error (RMSE) was employed. Which is;

N 

(

) 2

Predicted Actual −

i

i

(8)

RMSE

=

1

i

=

N

3.1. Random Forest Regression Using the results of 90 load spectra, a random forest regression model was built with 100 estimators. The RMSE for this model was calculated as 9,926 FH. The actual CG life and predicted CG life of this model are presented in Figure 5.

Fig. 5. Comparison of predicted and actual values for random forest regression with original data

To remove outliers, input data was filtered while keeping spectrums with CG life lower than 40,000 FH. RMSE of the filtered model was calculated as 383 FH. The actual CG life and predicted CG life of this model are presented in Figure 6.

Fig. 6. Comparison of predicted and actual values for random forest regression with filtered data

Finally, a grid search algorithm was utilized to filter data to find optimum parameters and reduce the RMSE. According to grid search algorithm results, the number of estimator values should be 292. The RMSE error of the