134 B. Jones et al. was able to classify the amount of damage that was present, with a root squared mean error (RSME) of less than 0.5g which means that our model had a high predictive accuracy of how much mass is removed from the MEL. Destructive Cases The same methodology and experimental setup was applied for the non-destructive case. The only difference is that the MEL would be damaged by removing mass around the lug. To execute this the MEL was cut by a milling machine, carefully and not excessively damaging the structure. To measure the amount of mass removed from the structure we weighed the MEL before and after it was cut. The first cut removed 1.6 grams and increased to 2.3 grams and ended with 4.1 grams. The cuts that were made on the structure are pictured in Figure 4. Using the same SVR model to see if the machine learning algorithm could predict how much mass was removed from the MEL. In Figure 5, the SVR predictions show a strong correlation in our model identifying how much damage is on the MEL. At this point the model was proven successful and yield promising results. Fig. 4 Damage Test Cases. Fig. 5 SVR Results Destructive Case.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==