5001.004.001.001 Frameworks for Utilizing Process Monitoring in Conjunction with Ex-situ Inspection for Qualification

AI software predicts presence of pores in each build layer with 100% accuracy

Artificial intelligence (AI) will evaluate datasets to determine an optimal in-situ monitoring framework balancing framework cost and complexity with part quality predictive performance resulting in a prioritized in-situ monitoring specification.


Laser powder bed fusion (LPBF) and directed energy deposition (DED) additive manufacturing (AM) processes have the potential to reduce Department of Defense (DoD) and industrial sustainment costs and delays via on-demand part production and repair. LPBF and DED part quality is dependent on in-process variables requiring significant post-build inspections adding cost and schedule to builds offsetting benefits of these AM processes. An in-situ process monitoring framework with the ability to predict post-build outcomes with high statistical accuracy is needed.


The goal of the project was to foster conditional inspection practices by evaluating different in-situ monitoring framework qualities and features for their ability to predict post-build non-destructive examination (NDE) outcomes with high statistical confidence and use the results to develop a framework specification for in-situ monitoring.

Technical Approach

The project evaluated two different artificial intelligence (AI) correlation applications: Artisan-AI©, and Pyramid Ensemble Convolution Neural Network (PECNN) on their ability to predict post-build part inspection outcomes on two existing data sets. The project team developed requirements and metrics for evaluating the AI applications and extracted metrics from NDE results and sensor data from the two existing datasets. With metrics established, both datasets were evaluated by the two AI correlation applications to determine a predicted NDE result. Statistically significant results were reviewed to determine which sensor data and AI correlation application produced the highest confidence NDE prediction outcomes. From this analysis, a draft in-situ monitoring specification was delivered.


AI algorithms were able to predict pore defect volume in a validation build with high confidence using in-situ process monitoring sensor data in Inconel 718 test coupons. Further, the project team developed a limited case in-situ monitoring framework for Inconel 718 LPBF aerospace applications through guidance provided by American Society for Testing and Materials (ASTM) and America Makes working group members. The AI correlation applications processed the training files created from existing datasets to correlate sensor metrics with NDE outcomes and the statistical confidence was evaluated. The results were sorted by statistical significance (R2) to determine the combination of in-situ sensors that best predicted NDE outcomes. The project team was able to conclude a model featuring both off-axis and coaxial sensors, intensity-based and emissivity independent true temperature sensors, and machine parameters (e.g., build plate lead screw rotation angle) are important in predicting NDE outcomes relative to the project datasets. An additional Inconel 718 dataset was compiled and configured into a training file. This dataset was processed by the AI algorithms to develop an in-situ sensor model to predict NDE outcomes to validate the previous results. The validation dataset in-situ sensor model was able to predict the presence of a pore (greater than 0.0 mm3) with 100% accuracy and pore volume over prescribed threshold value with 99% accuracy.

Project Participants

Project Principal

Other Project Participants

  • America Makes
  • Air Force Research Laboratory
  • Macy Consulting, LLC
  • The Ohio State University
  • University of Louisville

Public Participants

  • U.S. Department of Defense

Success Story

Back to Projects