researcher Based on surface data, we have developed a machine learning technique to detect internal structures, voids, and cracks in materials.
“Anything that has the potential to produce a smarter-than-human mind, in the form of artificial intelligence, brain-computer interfaces, or neuroscience-based enhancements of human intelligence, will do the most to change the world.” As you are, you beat the competition…even in the same league.”- Eliezer Yudkowski.
Materials engineering problem
Materials engineering problems are difficult to solve because they often require more information. For example, inverse problems with only boundary data, or design problems with simple goals but sufficient search space. Several deep learning (DL) architectures predict missing mechanical information given limited known data in part of the domain, and reconstruct 2D and 3D complex microstructures. Characterize complex geometries using a mechanical field.
2D microstructure
In 2D, we use conditional generative adversarial networks (GANs) to complete partially masked field maps and convolutional models to predict complex geometries with high accuracy and versatility. It does this by making accurate predictions for field data containing mixed stress/strain components, hierarchical structures, different material properties, different types of microstructures, and improperly set inverse problems.
3D microstructure
3D uses a Transformer-based design to predict a full 3D machine field from a snapshot of the input field. This model works well no matter how complex the microstructure is, and can recover the entire bulk region from his single image of the surface region. Furthermore, internal structures can be described using only edge measurements. The whole framework makes it easier to analyze and build things even if you don’t have all the information you need. You can also easily navigate back and forth from the properties to the material structure.
Proposed work
In this job researcher It provides an AI-based framework for predicting global strain and stress fields from partial field data and achieving the inverse transformation from mechanical domains to composite microstructures. Numerical simulations like FEA are forward solvers used to determine structure-property relationships. However, to solve inverse problems such as geometry identification, additional optimizers such as optimization algorithms are often used in conjunction with forward solvers to iteratively search for the best solution.
evaluation
To overcome this limitation, researchers use a combination of deep learning architectures to directly link imperfect strain or stress fields to heterogeneous structures in 2D and 3D environments. In 2D, we use conditional generative adversarial network (GAN) techniques and convolutional models to recover masked regions in the field map and identify composite structures from the recovered mechanical field.
Field completion methods are validated in various scenarios such as:
- When multiple strain/stress components are mixed in the dataset.
- When you need to stretch out-of-distribution microstructures with varying material shapes or grid sizes.
- When the mechanical properties of the constituent material involve plasticity.
- Given a continuous microstructure rather than discrete blocks such as Cahn-Hilliard patterns.
Conclusion
The method proposed in this work can also be used to solve the reverse design problem. The proposed method can handle more complex material responses compared to designing a single feature. These responses can be fed into a deep learning model in the same way as partial field information to obtain the corresponding material structure. For example, by combining our method with an optimization algorithm, we can iteratively search for suitable candidates with boundary responses closest to the goal. Additionally, experimental techniques such as “additive manufacturing” can be used to create real-world examples of designs proposed by computer frameworks.
