The laser powder bed fusion (LPBF) industry, also known as 3D metal printing or additive manufacturing, uses a highly complex combination of physics, chemistry, materials science, and process engineering to create parts for a growing number of industries. But the parts building process has many adjustable inputs. These directly affect the ability to create a part and its final quality: equipment type and condition, materials quality and type, and engineering knowledge.

Several kinds of anomalies caused by the nature of the printing process can also affect the final quality of 3D printed parts. Additionally, there is often little experience ensuring consistent, repeatable quality in the manufacturing process. Unfortunately, those anomalies and how they impact in-process measurement data can be difficult to interpret. Certifying that a part meets quality assurance standards by destructive testing and CT scanning is expensive and time-consuming. And these methods do not always yield accurate results.

The In-Process Quality Imperative

As manufacturers increase their production of 3D printed parts, so does the importance of using in-process measurements and non-destructive testing. What if manufacturers could monitor and modify part and process quality in real time? They’d be able to make quick adjustments of materials and machine settings, or even redesign the parts. That would save time, money and materials. They could also ensure that each part was the same quality as the one before and the one after it.

One effective method of data analysis uses advanced computer algorithms (i.e., machine learning). Machine learning is based on the idea that systems can learn from data, identify patterns and make decisions with minimal human intervention. As Techopedia.com defines machine learning:

Machine learning is an artificial intelligence (AI) discipline geared toward the technological development of human knowledge. Machine learning allows computers to handle new situations via analysis, self-training, observation and experience. Machine learning facilitates the continuous advancement of computing through exposure to new scenarios, testing and adaptation, while employing pattern and trend detection for improved decisions in subsequent (though not identical) situations.

Sigma Labs has proven that machine learning can accurately predict where anomalies in 3D printed parts are likely to occur. Sigma found, though a series of experiments, that machine learning did predict anomalies in 3D additively printed parts better than any other single metric.

Sigma Labs developed a machine learning process for:

• Semi-automatic labelling of anomalies in parts that have already been tested using non-destructive CT scanning.
• Training machine learning models using in-process data and these anomaly labels.
• Applying the models to accurately predict where new production parts will exhibit similar patterns.
• Measuring the machine learning models’ accuracy at predicting anomalies.

PrintRite3D^® Machine Learning – A QA Case Study

Sigma Labs applied machine learning models using PrintRite3D^® IPQM^® metrics to recognize four common anomaly types: lack of fusion, gas porosity, keyhole, and tungsten inclusions. After running the part builds, we examined the PrintRite3D^® in-process metrics to choose one part for each anomaly type. We sent the physical specimens and metric data to the Jesse Garant Metrology Center for objective third-party analysis. The Center provides industrial CT scanning and x-ray services. The Center’s results showed the location and kind of detected anomalies in the same part coordinates as Sigma’s. Thus, Sigma’s machine learning models accurately predicted each specific anomaly’s presence.

Our team at Sigma Labs used the in-process metrics from the parts builds to train the machine learning models to diagnose how well the model predictions compared to each of the training labels (lack of fusion, gas porosity, keyhole, and tungsten inclusions). We found that the models could be well trained to recognize their own anomaly types and avoid flagging other anomaly types.

In addition, we also evaluated the machine learning models by visualizing the predictions in the PrintRite3D^® user interface. Again, we found that the models could be well trained to recognize their own anomaly types and avoid flagging other types.

Conclusion

Sigma Labs demonstrated that machine learning models can be trained to recognize four different types of anomalies that occur in the 3D printing process: lack of fusion, gas porosity, keyhole, and tungsten inclusions. The predictions show a strong, accurate correlation with CT labels, as verified by third-party testing.

Prediction metrics like these can make it possible in the future to close the loop on controlling the build process. When production managers can predict anomalies, they can decide whether to interrupt or adjust the build process or even redesign the part. This will ultimately save time, money and materials, enhancing the manufacturer’s bottom line.

In any experiment the ultimate objective is to provide a rational basis for action. A problem or question exists, and something is to be done or answered about it. This experiment was performed as an enumerative study, where problems are generated and subsequent detection and modeling is attempted, regardless of why those problems are large or small. However, the AM industry quest is to solve the analytic problem, where something must be done to regulate and predict the results of the causal system that made these anomalies in the past and will continue to produce them in the future.

About Sigma Labs

Founded in 2010, Sigma is a software company that specializes in the development and commercialization of real-time computer aided inspection (CAI) solutions known as PrintRite3D® for 3D advanced manufacturing technologies.

Our people, processes and technologies are recognized leaders in disruptive technologies. Our products and services are engineered, manufactured, and qualified for use in highly demanding production environments for the aerospace & defense industries. Sigma’s innovative approach to quality control/assurance is a proactive, comprehensive and process-focused methodology that allows prediction with adequate confidence of product conformance to defined acceptance requirements.

Since inception in September 2010, Sigma has been able to establish credibility and acceptance within the aerospace and defense community and has become the go-to AM experts for real-time computer aided inspection (CAI) solutions. Sigma is ITAR Certified.

Read the detailed report.