Defect Characterization and Classification for Metal Additive Manufacturing Process Optimization Using Machine Learning

Public Deposited

Downloadable Content

open in viewer

Metal 3D Printing, or Additive Manufacturing (AM), is a layer-by-layer manufacturing process that involves adding the material to create a structure instead of removing it. AM is a complex process governed by many parameters, which are often interdependent and impacted by the manufacturing environment. Manual quality inspection of these parts is not only a time-consuming process but also depends on the skill of the quality control inspector. Due to the novelty of this area of manufacturing, the process-structure-property relationship is yet to be fully explored. We investigated more closely the relationship between porosity and process parameters. In particular, we developed a semi-automated method for the extraction of defect characteristics from image data. We also examined the relationship between these properties and the type of defect to discover the best geometric indicators of pore type. Using these characteristics, we propose an alternative vision-based defect classification model. We conducted a comparative study of classification methods on 1794 different pores, which indicated Random Forest as the best approach to the problem with a mean cross-validation accuracy of 93.81% and 98.10% Average Precision. Additionally, we found that providing additional information regarding the processing parameters to the model can provide a statistically significant improvement to the classifier’s performance, increasing mean cross-validation accuracy to 96.15% and Average Precision to 98.76%. Furthermore, we compared the performance of six different regression algorithms for porosity prediction in parts built with additive manufacturing. We showed that Gradient Boosting Regressor is the best model with a Mean Absolute Error (MAE) on porosity precision of 0.3753 and Root Mean Squared Error (RMSE) of 0.4558, and Mean Average Percentage Error (MAPE) of 0.4514 on solid ratio prediction. Finally, we built separate regression models for predicting the amount of different types of porosity, keyhole porosity prediction model obtaining MAE, RSME, and MAPE of 0.0029, 0.0034, and 0.0029, respectively.

  • etd-71811
Defense date
  • 2022
Date created
  • 2022-08-11
Resource type
  • etd-71811
Rights statement


In Collection:



Permanent link to this page: