AI Approach To Identify Binder And Fibers
The FiberFind module provides the possibility of using Artificial Intelligence (AI) approaches for the separation of binder and fibers as well as the identification of individual fibers in a fibrous structure.
In the image taken from a material, fibers and binder often have the same gray values, and therefore cannot be separated based on this value. The same holds true for the separation of different fibers. However, it is possible to separate binder from fibers and fibers from each other based on the shape in the image.
The fundamental idea of AI is that a neural network is trained to perform a special task, here the identification of separate fibers or the differentiation between fibers and binder. To train the neural network, enough examples need to be available to teach it. Clearly, for 3D-scans it is a very hard and time-consuming task to manually label materials or objects for the number of examples required, which are typically hundreds of millions. This is where GeoDict’s unique structure-generation capabilities come in.
As long as the models are close enough to the real 3D-scans, GeoDict is capable of providing the needed ground truth data for the neural networks that embody the AI capability in FiberFind. Since GeoDict 2021, the capability to create and parameterize these neural networks is available in the module GeoDict-AI.
Based on the trained neural network, FiberFind decides for each voxel of the structure whether it is binder or fiber and sets the material ID accordingly.
In the same way, a trained neural network for FiberFind-Identify Fibers (AI) decides for each voxel to which fiber it belongs.
The current version of FiberFind-Identify Fibers (AI) and FiberFind-Identify Binder (AI) come with a single neural network each that can be applied to segment data sets in the FiberFind module. Therefore, currently some constraints on the input data exist which may make it necessary to preprocess the input data before applying the AI approach.
