Automatic detection of fibers orientation on composite laminates using convolutional neural networks

Abstract

The mechanical behaviour and failure of the composite materials are highly influenced by the fibre’s orientation. It is important to decide which type of layers and orientations to use for the layup sequence such that the composite laminate is as light and/or cheap as possible while being capable to carry the load for which it is designed. During the production process it is critical to cut and lay the composite woven according to the optimal layup sequence resulting from the optimization process. Therefore, it is important to develop an automatic system which is capable to accurately detect the fibres orientation from a material picture. In our previous work we proposed such a system based on image processing and computational geometry. The main disadvantage of this system is that it has a lot of parameters which need to be manually tuned for each material. In this paper we propose a simple system which doesn’t need any parameter tuning. It is based on convolutional neural networks and we demonstrate with a lot of examples that it is very accurate, stable and insensitive to image variations. The method was tested on diverse composite laminates and woven with different chromatic and morphological properties.

Automatic Detection of the Orientation of Strain Gauges Bonded on Composite Materials with Polymer Matrix, in Order to Reduce the Measurement Errors

“…As it can be seen from the description of the SG filament orientation determination procedure, this is a deterministic one, not using neural models like those applied in the automatic fiber orientation detection method [ 30 ]. There are two reasons for choosing this deterministic approach: The chromatic characteristics of the region representing the SG are favorable in the sense that they allow for the markers present in this region to be easily detected.…”

“…The detection of the θ fibres orientation of the composite material’s fibers upon which the SG is mounted can be done using the method described in [ 29 ] (method 1) or the neural model described in [ 30 ] (method 2). The only precaution to be taken is to select one or more rectangular regions from the acquired image that contain only the composite material, without portions of the SG.…”

“…[ 25 , 26 , 27 , 28 ]. An automatic method for the detection of fiber orientation on composite laminates using convolutional neural networks is presented in papers [ 29 , 30 ]. This method does not require sophisticated equipment, is cheap, quite fast and has high accuracy.…”

“…In this article, references to the determination of fiber orientation will also be made. For this purpose, the method described in [ 29 , 30 ] will be used because both methods, the one used to determine the orientation of fibers and the one used to determine the orientation of strain gauges, respectively, use the same equipment and different, but compatible, software.…”

“…In the paper [ 29 ], it is shown that the fiber orientation of composites can be automatically detected with good accuracy based on classical image processing and computational geometry. Another method can do the same using convolutional neural networks [ 30 ]. It turns out that these methods give good results, even for cases where the orientation of the composite’s fibers can be observed with difficulty with the naked eye.…”