Flaws in Powder Bed Fusion Additive Manufacturing: Formation Mechanisms, Detection Methods, and Effect on Fatigue
Open Access
- Author:
- Snow, Zackary K
- Graduate Program:
- Engineering Science and Mechanics
- Degree:
- Doctor of Philosophy
- Document Type:
- Dissertation
- Date of Defense:
- June 18, 2021
- Committee Members:
- Edward Reutzel, Chair & Dissertation Advisor
Jayme Keist, Outside Field Member
Timothy Simpson, Outside Unit Member
Abdalla Nassar, Major Field Member
Albert Segall, Program Head/Chair - Keywords:
- additive manufacturing
qualification
process monitoring
fatigue
titanium
machine learning - Abstract:
- Laser-based powder bed fusion (L-PBF) additive manufacturing (AM) offers several advantages over traditional manufacturing techniques, but is associated with variability in mechanical properties that has prevented widespread use in fatigue-critical applications. Post-process treatments, such as hot isostatic pressing (HIP), reduce variability, but low lifetime outliers persist. This variability has been attributed to internal flaws, but the size, morphology, and locations of flaws which most impact fatigue properties is unclear. Flaws are often detected post-build through X-ray computed tomography (XCT), but XCT can be expensive and limited for large components made of high atomic number components. Flaw detection via real time analysis of in situ monitoring data may allow flaws to be detected as they form, possibly removing the need for post-build XCT and enabling automated corrective action. To address these concerns, the following research objectives were established: (1) improve the understanding of flaw formation mechanisms in L-PBF, (2) evaluate the efficacy of current post-build flaw detection methods, (3) determine the effects of flaws on the fatigue properties of AM material, and (4) explore novel flaw detection techniques. First, a thorough review of relevant literature was conducted to understand the current perspective on formation mechanisms for L-PBF flaws. To evaluate the efficacy of XCT, flaw sizes and locations were extracted from XCT scans of a L-PBF test specimen, the results of which were compared to a ground truth flaw population generated using automated serial sectioning. To assess the impact of flaws on fatigue properties, Ti-6Al-4V fatigue coupons were built and tested to failure, and the resulting fatigue lives and locations of the failure origins were correlated to both pre- and post-HIP flaws detected via XCT. Flaw detection from process monitoring data was also assessed and demonstrated through the application of machine learning techniques. While still an effective quality inspection tool, flaw detection using XCT is inherently probabilistic. The canonical three voxel requirement for detecting flaws of a given size was found to be just a minimum — less than 4% of flaws approximately three voxels in dimeter were detected via XCT. Through a combination of fatigue testing, fractography, and analysis of in situ monitoring data, a connection between process ejecta and stochastic, fatigue-critical lack-of-fusion was established for the first time. Stochastic lack-of-fusion generated by the spatter particles were identified in layerwise images and were found to detrimentally affect both pre- and post-HIP fatigue properties. Some pre-HIP flaws were confirmed to not truly heal during HIP but instead reduced in size below post-HIP XCT detection limits, complicating NDI requirements. Finally, ML-based flaw detection of process monitoring data was found to be accurate enough to not incur a significant debit on processing time. However, strategies for loading and analyzing sensor data in real-time must also not seriously affect total build time and will require additional research. Data fusion of disparate process monitoring modalities was found to greatly improve classifier performance and generalizability. Future work should investigate the XCT acquisition parameters that most affect flaw detection. More research is also required to quantify the relationship between flaw size, morphology, and location in real AM components with complex geometries and surface topologies. The relative importance of oxygen embrittlement and microstructural texturing on the fatigue life of L-PBF Ti-6Al-4V remains unclear. The source of contaminants, which were highly detrimental to fatigue life, should also be investigated. Strategies for real-time implementation of flaw detection and interlayer repair must be developed to accelerate qualification of AM components. Such a quality control system could greatly reduce the cost and time required for post-build inspection, while interlayer flaw repair could improve the repeatability and reliability of AM processes to accelerate industry adoption of the technology in critical applications.