Metal Additive Manufacturing (AM) promises an era of highly flexible part production, replete with unprecedented levels of design freedom and inherently short supply chains. But as AM transitions from a technology primarily used for prototyping to a viable manufacturing method, many challenges must first be met before these dreams can become reality. In order for machine users to continue pushing the design envelope, process space must be expanded beyond the limits currently recommended by the machine manufacturers. Furthermore, as usable process space expands and demands for reduced operator burden and mission-critical parts increase, in-situ monitoring of the processes will become a greater necessity. Processing space includes both the parameters (e.g. laser beam power and travel velocity) and the feedstock used to build a part. The correlation between process parameters and process outcomes such as melt pool geometry, melt pool variability, and defects should be understood by machine users to allow for increased design freedom and ensure part quality. In this work, an investigation of the AlSi10Mg alloy in a Laser Powder Bed Fusion (L-PBF) process is used as a case study to address this challenge. Increasing the range (processing space) of available feedstocks beyond those vetted by the machine manufacturers has the potential to reduce costs and reassure industries sensitive to volatile global supply chains. In this work, four non-standard metal powders are successfully used to build parts in an L-PBF process. The build quality is compared to that of a standard powder (supplied by the machine manufacturer), and correlations are found between the mean powder particle diameters and as-built part quality. As user-custom parameters and feedstocks proliferate, an increased degree of process outcome variability can be expected, further increasing the need for non-destructive quality assurance and the implementation of closed-loop control schema. This work presents two Machine Learning-based Computer Vision algorithms capable of autonomously detecting and classifying anomalies during the powder spreading stage of L-PBF processes. While initially developed to serve as the monitoring component in a feedback control system, the final algorithm is also a powerful data analytics tool – enabling the study of build failures and the effects of fusion processing parameters on powder spreading. Importantly, many troubling defects (such as porosity) in AM parts are too small to be detected by monitoring the entire powder bed; for this reason, an autonomous method for detecting changes in melt pool morphology via a high speed camera is presented. Finally, Machine Learning techniques are applied to the in-situ melt pool morphology data to enable the study of melt pool behavior during fusion of non-bulk part geometries.

Laser powder bed fusion (LPBF) remains a predominately open-loop additive manufacturing process with minimal in-situ quality and process control. Some machines feature optical monitoring systems but lack automated analytical capabilities for real-time defect detection. Recent advances in machine learning (ML) and convolutional neural networks (CNN) present compelling solutions to analyze images in real-time and to develop in-situ monitoring. Approximately 30,000 selective laser melting (SLM) build images from 31 previous builds are gathered and labeled as either “okay” or “defect”. Then, 14 open-sourced CNN were trained using transfer learning to classify the SLM build images. These models were evaluated by F1 score and down selected to the top 3 models. The top 3 models were then retrained and evaluated using Dietterich’s 5x2 cross-validation and compared with pairwise student t-tests. The pairwise t-test results show no statistically significant difference in performance between VGG- 19, Xception, and InceptionResNet. All models are strong candidates for future development and refinement. Additional work addresses the entire model development process and establishes a foundation for future work. Collaborations with computer science students has produced an image pre-processing program to enhance as-taken SLM images. Other outcomes include initial work to overlay CAD layer images and preliminary hardware integration plan for the SLM machine. The results from this work have demonstrated the potential of an optical layer-wise image defect detection system when paired with a CNN.

No description available.

No description available.