Race against the Machine: can deep learning recognize microstructures as well as the trained human eye?

Larmuseau, Michiel; Sluydts, Michael; Theuwissen, Koenraad; Duprez, Lode; Dhaene, Tom; Cottenier, Stefaan

doi:10.1016/j.scriptamat.2020.10.026

Cited by 25 publications

(12 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Indeed, such synergy can be beneficial at three different levels. Firstly, ML has established itself as a robust tool for quantitative phase analysis [ 77 , 78 , 79 ], thus facilitating the collection of the large amounts of data needed for the careful validation of enhanced thermo-metallurgical FE models. Second, Bayesian approaches combined with ML have already been proved powerful for parameters identification, and this could be applied to the set of presently unknown material parameters within the TTB concept [ 79 ].…”

Section: Discussionmentioning

confidence: 99%

A New Concept for Modeling Phase Transformations in Ti6Al4V Alloy Manufactured by Directed Energy Deposition

et al. 2021

View full text Add to dashboard Cite

The microstructure directly influences the subsequent mechanical properties of materials. In the manufactured parts, the elaboration processes set the microstructure features such as phase types or the characteristics of defects and grains. In this light, this article aims to understand the evolution of the microstructure during the directed energy deposition (DED) manufacturing process of Ti6Al4V alloy. It sets out a new concept of time-phase transformation-block (TTB). This innovative segmentation of the temperature history in different blocks allows us to correlate the thermal histories computed by a 3D finite element (FE) thermal model and the final microstructure of a multilayered Ti6Al4V alloy obtained from the DED process. As a first step, a review of the state of the art on mechanisms that trigger solid-phase transformations of Ti6Al4V alloy is carried out. This shows the inadequacy of the current kinetic models to predict microstructure evolution during DED as multiple values are reported for transformation start temperatures. Secondly, a 3D finite element (FE) thermal simulation is developed and its results are validated against a Ti6Al4V part representative of repair technique using a DED process. The building strategy promotes the heat accumulation and the part exhibits heterogeneity of hardness and of the nature and the number of phases. Within the generated thermal field history, three points of interest (POI) representative of different microstructures are selected. An in-depth analysis of the thermal curves enables distinguishing solid-phase transformations according to their diffusive or displacive mechanisms. Coupled with the state of the art, this analysis highlights both the variable character of the critical points of transformations, and the different phase transformation mechanisms activated depending on the temperature value and on the heating or cooling rate. The validation of this approach is achieved by means of a thorough qualitative description of the evolution of the microstructure at each of the POI during DED process. The new TTB concept is thus shown to provide a flowchart basis to predict the final microstructure based on FE temperature fields.

show abstract

Section: Discussionmentioning

confidence: 99%

A New Concept for Modeling Phase Transformations in Ti6Al4V Alloy Manufactured by Directed Energy Deposition

et al. 2021

View full text Add to dashboard Cite

show abstract

“…In recent years, machine-learning methods have been adopted by materials scientists to successfully identify or restore features of interest from microscopic or tomographic images (e.g. Larmuseau et al, 2021;Jiang et al, 2020;DeCost et al, 2017;Dimiduk et al, 2018). Particular interest has also increased within X-ray diffraction imaging.…”

Section: Introductionmentioning

confidence: 99%

Deep learning for improving non-destructive grain mapping in 3D

Fang

Hovad

Zhang

et al. 2021

Int Union Crystallogr J

View full text Add to dashboard Cite

Laboratory X-ray diffraction contrast tomography (LabDCT) is a novel imaging technique for non-destructive 3D characterization of grain structures. An accurate grain reconstruction critically relies on precise segmentation of diffraction spots in the LabDCT images. The conventional method utilizing various filters generally satisfies segmentation of sharp spots in the images, thereby serving as a standard routine, but it also very often leads to over or under segmentation of spots, especially those with low signal-to-noise ratios and/or small sizes. The standard routine also requires a fine tuning of the filtering parameters. To overcome these challenges, a deep learning neural network is presented to efficiently and accurately clean the background noise, thereby easing the spot segmentation. The deep learning network is first trained with input images, synthesized using a forward simulation model for LabDCT in combination with a generic approach to extract features of experimental backgrounds. Then, the network is applied to remove the background noise from experimental images measured under different geometrical conditions for different samples. Comparisons of both processed images and grain reconstructions show that the deep learning method outperforms the standard routine, demonstrating significantly better grain mapping.

show abstract

“…In general, we can specify the following area of application the machine learning in material science application: Classification the element of structures [9,[11][12][13] • Prediction of material properties based on a set of properties -structural and mechanical [14] • Structure-oriented design: design a new material, and predict their properties before conducting the physical test [15,16] • Design new compounds and their structure from an input composition [16] • Invers design -from required properties to the molecular structure [2,17] Due to obtained accurate results of the analysis the large set of data must be prepared as a training data set and it is the major difficulties in widespread application especially in the analysis of the images of material structure. Due to perform the classification of the structure element the training dataset should me between 100 -1000 images of the structure.…”

Section: Training Data Characteristic Feature Extractionmentioning

confidence: 99%

Efficiency and Accuracy of Simulated Microstructure Images Generated by Machine Learning

Gądek-Moszczak¹,

Filipowska²

2021

14th WCCM-ECCOMAS Congress

View full text Add to dashboard Cite

Computer image analysis is a well-known method in material science, mechanical engineering and other branches of science and engineering. Application of the machine learning method in image processing delivers promising results, especially in images with a high level of noise and low contrast. Employing automatic classification, predictive models, simulation models deliver a huge benefit in research. The former model of science-based mainly on experiments slowly passing away. It is still an important part of research but, using advanced computer methods save time and allow significantly reduces the cost of studies. It begins new branch of material science and new research challenges, interdisciplinary filed of materials informatics, incorporating scientists exploring, mathematics, informatics and material science.

show abstract

Race against the Machine: can deep learning recognize microstructures as well as the trained human eye?

Cited by 25 publications

References 14 publications

A New Concept for Modeling Phase Transformations in Ti6Al4V Alloy Manufactured by Directed Energy Deposition

A New Concept for Modeling Phase Transformations in Ti6Al4V Alloy Manufactured by Directed Energy Deposition

Deep learning for improving non-destructive grain mapping in 3D

Efficiency and Accuracy of Simulated Microstructure Images Generated by Machine Learning

Contact Info

Product

Resources

About