Anomaly detection and classification in a laser powder bed additive manufacturing process using a trained computer vision algorithm

Scime, Luke; Beuth, Jack

doi:10.1016/j.addma.2017.11.009

Cited by 262 publications

(149 citation statements)

References 9 publications

Supporting

Mentioning

148

Contrasting

Unclassified

Order By: Relevance

“…Additionally, an operator will typically select a set of global parameter settings that are used throughout the part, or at most make some layer-to-layer adjustments. [28,29] An Quality control and repeatability of 3D printing must be enhanced to fully unlock its utility beyond prototyping and noncritical applications. Machine learning models are trained on thousands to millions of data points to recognize patterns that are too difficult to identify using deterministic algorithms.…”

mentioning

confidence: 99%

“…[5][6][7] Potential uses for machine learning in 3D printing have also been studied in a limited capacity. [28,29] An Quality control and repeatability of 3D printing must be enhanced to fully unlock its utility beyond prototyping and noncritical applications. [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27] An example of the utility of machine learning in the established quality control method of visual inspection is demonstrated by the use of a neural network to identify flaws in laser powder bed fusion 3D printing.…”

mentioning

confidence: 99%

See 1 more Smart Citation

Machines as Craftsmen: Localized Parameter Setting Optimization for Fused Filament Fabrication 3D Printing

Gardner

Hunt

Ebel

et al. 2019

Adv Materials Technologies

View full text Add to dashboard Cite

FFF is known to have issues with part quality and consistency, which has limited its use to prototyping and noncritical applications where functional reliability does not affect safety. [2] The occurrence of issues such as poor surface finish, layer delamination, and poor dimensional stability depends on a number of parameter settings, including nozzle temperature, print speed, environmental conditions, geometry, and location (Figure 1a,b). [3] Experienced operators must set these parameters according to the material, part geometry, and 3D printer. It can be difficult for even an expert to select optimum parameter settings (Figure 1c,d). [4] An operator can specify settings for over 100 different printing options using a slicing software such as Cura. Attempting all combinations is not practical, so an operator must rely on their knowledge of 3D printing to adjust parameter settings. This leads to operator-dependent part-to-part variations and uncertainty in the performance of the final part. Additionally, an operator will typically select a set of global parameter settings that are used throughout the part, or at most make some layer-to-layer adjustments. A more objective solution for selecting parameter settings is needed to ensure optimal use of the printer/material performance space and consistency in 3D-printed parts, regardless of part geometry, material, system, or operator.Machine learning offers a potential approach to addressing this problem. Machine learning models are trained on thousands to millions of data points to recognize patterns that are too difficult to identify using deterministic algorithms. [5] Machine learning has been successfully applied in applications such as image processing, text classification, and speech recognition. [5][6][7] Potential uses for machine learning in 3D printing have also been studied in a limited capacity. [8] Examples of their use in both monitoring/feedback applications and predictive models include predicting property outcomes based on parameter settings, predicting global parameter settings for specific outcomes, identifying failures during printing, predicting bead geometry, adjusting geometry to prevent failures, and assessing part manufacturability. [9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27] An example of the utility of machine learning in the established quality control method of visual inspection is demonstrated by the use of a neural network to identify flaws in laser powder bed fusion 3D printing. [28,29] An Quality control and repeatability of 3D printing must be enhanced to fully unlock its utility beyond prototyping and noncritical applications. Machine learning is a potential solution to improving 3D printing performance and is explored for areas including flaw identification and property prediction. However, critical problems must be resolved before machine learning can truly enable 3D printing to reach its potential, including the very large data sets required for training and the inherently local nature of 3D print...

show abstract

mentioning

confidence: 99%

mentioning

confidence: 99%

Machines as Craftsmen: Localized Parameter Setting Optimization for Fused Filament Fabrication 3D Printing

Gardner

Hunt

Ebel

et al. 2019

Adv Materials Technologies

View full text Add to dashboard Cite

show abstract

“…

…”

mentioning

confidence: 99%

“…These sen sors provide assessments of spatial and spectral features of the melt pool, [19,20] process plume, [21] degree of spatter, [22][23][24][25] overhang layers, [26] or print bed. High speed image sequences of the melting process, [27] scans of the powder bed, [10,28] beam quality, [29] and/or thermal monitoring [30] are all routinely collected forms of in situ monitoring data. Making use of this data requires methods that can extract rel evant diagnostic information.…”

mentioning

confidence: 99%

Machine‐Learning‐Based Monitoring of Laser Powder Bed Fusion

Yuan

Guss

Wilson

et al. 2018

Adv Materials Technologies

View full text Add to dashboard Cite

variability in the powder properties, [9] bed thickness nonuniformity, [10] and laser parameters and scan paths that result in improper power melting. [11] Thus, even after optimizing LPBF operating para meters and identifying suitable processing windows, [12] rapid build qualification, improved quality, and higher production yields require methods of monitoring the melt pool and/or powder bed in situ, i.e., during a build, that enable realtime process feedback and automated quality detection. [13,14] The majority of LPBF process moni toring approaches rely on noncontact sensing [15] from optical, thermal, [16] and/ or acoustic [17,18] sensors. These sen sors provide assessments of spatial and spectral features of the melt pool, [19,20] process plume, [21] degree of spatter, [22][23][24][25] overhang layers, [26] or print bed. High speed image sequences of the melting process, [27] scans of the powder bed, [10,28] beam quality, [29] and/or thermal monitoring [30] are all routinely collected forms of in situ monitoring data. Making use of this data requires methods that can extract rel evant diagnostic information. For instance, before initiating laser melting, automated computer vision algorithms can characterize metal powder feedstocks, [31] and image analysis of newly spread powder can reveal nonuniformities in the powder bed thickness. [32] Aminzadeh et al. demonstrated layerbylayer detection of fusion defects from images using a Bayesian clas sifier. [33] Realtime events such as material ejecta are detectable by applying manually set thresholds to highspeed nearinfrared images of the melt pool. Also, increasing the µm pixel −1 image resolution relative to the standard deviation of measured track width, σ measured , may result in improved predictions of the final track width, δ predicted , and topography. [34] Reducing laser power proportionally to an integrated signal from a photodiode cali brated against a camera results in smoother overhang struc tures. [35] Using images of the print bed taken after laser melting, a level sets method can detect intentionally created defects, [36] machine vision algorithms can identify pore defects, [37] and multifractal image analysis can characterize layers with balling, cracks and pores, and no defects. [38] Visual imaging equipment is appealing to LPBF monitoring systems because it is relatively inexpensive and provides noncontact sensing. [13] As with most additive manufacturing systems, analysis of LPBF sensor data currently occurs postbuild, rendering A two-step machine learning approach to monitoring laser powder bed fusion (LPBF) additive manufacturing is demonstrated that enables on-the-fly assessments of laser track welds. First, in situ video melt pool data acquired during LPBF is labeled according to the (1) average and (2) standard deviation of individual track width and also (3) whether or not the track is continuous, measured postbuild through an ex situ height map analysis algorithm. This procedure generates three ground truth labeled datasets for supervised...

show abstract

“…Fortunately, the unique manufacturing process of AM affords the possibility of tight integration of in situ measurements and computational modeling. [10][11][12] Significant challenges remain, however, in material/structural qualification and development of process-structureproperties-performance linkages. [13][14][15][16][17] The recent 3rd Sandia Fracture Challenge focused on metal AM is an interesting example of the challenges involved in predicting the performance of AM components, even provided with significant a priori information of the as-built structure.…”

mentioning

confidence: 99%

Integrated Computational and Experimental Methods for Additive Manufacturing

Bishop

Clarke

Wagner

2018

JOM

View full text Add to dashboard Cite

Additive manufacturing (AM) holds the promise of revolutionizing current paradigms in engineering design, manufacturing, and material and structural performance.1-3 However, full realization of the possibilities of metal-based AM will require significant advances in both the process and performance modeling of additively manufactured materials and structures. [4][5][6][7][8] Modeling and simulation of AM processes is particularly challenging, since it involves multiple time and space scales.9 Furthermore, metal-based AM involves multiple physics, and is thus multidisciplinary in nature, requiring interdisciplinary research efforts. Fortunately, the unique manufacturing process of AM affords the possibility of tight integration of in situ measurements and computational modeling.10-12 Significant challenges remain, however, in material/structural qualification and development of process-structureproperties-performance linkages. [13][14][15][16][17] The recent 3rd Sandia Fracture Challenge focused on metal AM is an interesting example of the challenges involved in predicting the performance of AM components, even provided with significant a priori information of the as-built structure. 18With the goal of fostering communication and collaborations between the AM computational and experimental communities, a conference was recently held at the Colorado School of Mines, on September 6-8, 2017. 19 The conference brought together researchers from both the computational mechanics and materials science research communities involved in advancing the state of the art in process and performance modeling of additively manufactured (AM) materials and structures. There were over 31 invited speakers from academia, industry, and national laboratories, organized in a single-track session to foster communication. A central theme of the conference was integration of computational modeling and experiments, leveraging the unique manufacturing aspects of AM, to improve overall modeling predictivity and enable process optimization. This conference was jointly sponsored by the U. While the conference did not have published proceedings, presenters were invited to submit articles for a special topic in JOM. The following three articles have been assembled from this process. The first article by Roehling et al. describes the use of single-track laser-melting experiments to explore process parameters and their effects on microstructure. Experimental results were used to benchmark a phase-field model that simulated the rapid solidification in conditions relevant for AM. The second paper by Lindwall et al. models the creation of bulk metallic glasses using a powder-bed fusion process. The study evaluates computational approaches that can be used to increase the computational speed while maintaining accurate temperatures in critical regions. The accurate prediction of temperature history is particularly important in ensuring that reheated material does not become devitrified. The third paper by Donoso and Peters uses a combined discrete-contin...

show abstract

Anomaly detection and classification in a laser powder bed additive manufacturing process using a trained computer vision algorithm

Cited by 262 publications

References 9 publications

Machines as Craftsmen: Localized Parameter Setting Optimization for Fused Filament Fabrication 3D Printing

Machines as Craftsmen: Localized Parameter Setting Optimization for Fused Filament Fabrication 3D Printing

Machine‐Learning‐Based Monitoring of Laser Powder Bed Fusion

Integrated Computational and Experimental Methods for Additive Manufacturing

Contact Info

Product

Resources

About